PUBH5010

一句话先救命

• 这门课的 final 不是背公式的考试，而是：
  • 先认出研究设计
  • 再选对测量指标
  • 再用一句外行也听得懂的话解释数字
  • 最后做一套批判性评价（critical appraisal）
• 因为这是 open book，而且课程提供公式表，所以真正拉开分数的不是“会不会背公式”，而是“你能不能迅速做对判断”。[1]Source: asksia-bible-pubh5010-bilingual.pdfstandardisation。这些都是不受版权保护的范本,平实陈述。每个范例都使用 AskSia 自拟的数字,每个考试风格的情 景都经过改写和重新编号 -- 我们绝不复制任何往届试卷的具体题干或数据。书面状态(开卷、提供公式表、允许计算 器、禁止AI、考试门槛)均已引述并遵守。请在你的 Canvas 站点上核实。 AskSia Library · PUBH5010 · 双语 Bilingual THE BLUEPRINT - THE EXAM BLUEPRINT 60-85% FINAL . OPEN BOOK . EXAM HURDLE Where every mark lives 每一分藏在哪里 An open-book final - formula sheet supplied, so judgement and speed win it 一场 open-book 期末 -- 已提供公式表,所以靠判断与速度取胜 TL;DR. The final is worth 60% or 85% (whichever scheme is greater), it is open book, and the course supplies a "Helpful Formulae" sheet - so the formulas are not the bottleneck. Its make-or-break skills are (1) which measure / decision applies, (2) interpreting a number in plain words (never "two times higher" for RR = 2), and (3) a critical-appraisal pipeline. And you must clear the exam hurdle: ≥50% on the exam itself, or you fail the unit. TL;DR. 期末占 60% 或 85%(两套方案取分高者),它是 open book,而且课程提供一张“Helpful Formulae”表 -- 所以 公式不是瓶颈。它的成败技能是(1)该用哪个指标/该做哪个决策、(2)用大白话解读一个数字(对 RR=2 永远不要说“two times higher"),以及(3) critical-appraisal 流程。而且你必须过 exam hurdle:考试本身≥50%,否则整门课挂科。 60-85% FINAL EXAM WEIGHT 期末考权重 2h+10 WRITING + READING MIN 书写 + 阅读 分钟 OPEN BOOK . FORMULAE SUPPLIED 开卷 · 提供公式表 ≥50% EXAM HURDLE TO PASS 通过所需的考试 HURDLE The four assessment pieces 四个考核组成部分 Component Weight When Format Final exam 60- Exam 85% period In-person, open book; 2 h + 10 min reading; 3 sections A/B/C, separate booklets; non-prog calculator OK; no AI; HURDLE ≥50% Mid-sem assignment 25% / - Wk 6 0%[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

1）你最先要知道的：Final 到底怎么考

考试结构

• Final 占比很高：
  • 60% 或 85%（取更高方案）[1]Source: asksia-bible-pubh5010-bilingual.pdfstandardisation。这些都是不受版权保护的范本,平实陈述。每个范例都使用 AskSia 自拟的数字,每个考试风格的情 景都经过改写和重新编号 -- 我们绝不复制任何往届试卷的具体题干或数据。书面状态(开卷、提供公式表、允许计算 器、禁止AI、考试门槛)均已引述并遵守。请在你的 Canvas 站点上核实。 AskSia Library · PUBH5010 · 双语 Bilingual THE BLUEPRINT - THE EXAM BLUEPRINT 60-85% FINAL . OPEN BOOK . EXAM HURDLE Where every mark lives 每一分藏在哪里 An open-book final - formula sheet supplied, so judgement and speed win it 一场 open-book 期末 -- 已提供公式表,所以靠判断与速度取胜 TL;DR. The final is worth 60% or 85% (whichever scheme is greater), it is open book, and the course supplies a "Helpful Formulae" sheet - so the formulas are not the bottleneck. Its make-or-break skills are (1) which measure / decision applies, (2) interpreting a number in plain words (never "two times higher" for RR = 2), and (3) a critical-appraisal pipeline. And you must clear the exam hurdle: ≥50% on the exam itself, or you fail the unit. TL;DR. 期末占 60% 或 85%(两套方案取分高者),它是 open book,而且课程提供一张“Helpful Formulae”表 -- 所以 公式不是瓶颈。它的成败技能是(1)该用哪个指标/该做哪个决策、(2)用大白话解读一个数字(对 RR=2 永远不要说“two times higher"),以及(3) critical-appraisal 流程。而且你必须过 exam hurdle:考试本身≥50%,否则整门课挂科。 60-85% FINAL EXAM WEIGHT 期末考权重 2h+10 WRITING + READING MIN 书写 + 阅读 分钟 OPEN BOOK . FORMULAE SUPPLIED 开卷 · 提供公式表 ≥50% EXAM HURDLE TO PASS 通过所需的考试 HURDLE The four assessment pieces 四个考核组成部分 Component Weight When Format Final exam 60- Exam 85% period In-person, open book; 2 h + 10 min reading; 3 sections A/B/C, separate booklets; non-prog calculator OK; no AI; HURDLE ≥50% Mid-sem assignment 25% / - Wk 6 0%[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS
• 考试形式：
  • 线下
  • open book
  • 2小时作答 + 10分钟阅读
  • 允许不可编程计算器
  • 禁止 AI / 电子辅助
  • 考试本身必须 ≥50% 才能过课（exam hurdle）[1]Source: asksia-bible-pubh5010-bilingual.pdfstandardisation。这些都是不受版权保护的范本,平实陈述。每个范例都使用 AskSia 自拟的数字,每个考试风格的情 景都经过改写和重新编号 -- 我们绝不复制任何往届试卷的具体题干或数据。书面状态(开卷、提供公式表、允许计算 器、禁止AI、考试门槛)均已引述并遵守。请在你的 Canvas 站点上核实。 AskSia Library · PUBH5010 · 双语 Bilingual THE BLUEPRINT - THE EXAM BLUEPRINT 60-85% FINAL . OPEN BOOK . EXAM HURDLE Where every mark lives 每一分藏在哪里 An open-book final - formula sheet supplied, so judgement and speed win it 一场 open-book 期末 -- 已提供公式表,所以靠判断与速度取胜 TL;DR. The final is worth 60% or 85% (whichever scheme is greater), it is open book, and the course supplies a "Helpful Formulae" sheet - so the formulas are not the bottleneck. Its make-or-break skills are (1) which measure / decision applies, (2) interpreting a number in plain words (never "two times higher" for RR = 2), and (3) a critical-appraisal pipeline. And you must clear the exam hurdle: ≥50% on the exam itself, or you fail the unit. TL;DR. 期末占 60% 或 85%(两套方案取分高者),它是 open book,而且课程提供一张“Helpful Formulae”表 -- 所以 公式不是瓶颈。它的成败技能是(1)该用哪个指标/该做哪个决策、(2)用大白话解读一个数字(对 RR=2 永远不要说“two times higher"),以及(3) critical-appraisal 流程。而且你必须过 exam hurdle:考试本身≥50%,否则整门课挂科。 60-85% FINAL EXAM WEIGHT 期末考权重 2h+10 WRITING + READING MIN 书写 + 阅读 分钟 OPEN BOOK . FORMULAE SUPPLIED 开卷 · 提供公式表 ≥50% EXAM HURDLE TO PASS 通过所需的考试 HURDLE The four assessment pieces 四个考核组成部分 Component Weight When Format Final exam 60- Exam 85% period In-person, open book; 2 h + 10 min reading; 3 sections A/B/C, separate booklets; non-prog calculator OK; no AI; HURDLE ≥50% Mid-sem assignment 25% / - Wk 6 0%[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

试卷固定结构：A / B / C 三段

• Section A
  • 认研究类型
  • 做少量短计算
  • 可能涉及 RCT 基本概念（如 ITT、blinding）[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.
• Section B
  • 一个完整情境题
  • 从头做到尾：
    • 研究类型 + justify
    • 画 $2\times 2$
    • 选对测量
    • 计算
    • 通俗解释
    • NNT
    • 若干 1 分 mini-calcs[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.
• Section C
  • 给你一篇较长研究
  • 做 critical appraisal
  • 核心四块：
    • selection
    • exposure measurement
    • outcome measurement
    • confounding
  • 还可能涉及 effect modification、chance、causation 等[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

时间分配

• 文件明确建议按分值分配时间：
  • 大约 1 分钟/1 分
  • 要特意给 Section C 长题干留阅读时间[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").

2）这门课真正的“主线”是什么

这门课整个逻辑可以压成一条流水线：

• 先建 $2\times 2$
• 认研究设计
• 选合法测量
• 解释数字
• 识别偏倚
• 判断偏倚方向
• 完成批判性评价
• 再延伸到 screening / SR / MA / causation[12]Source: asksia-bible-pubh5010-bilingual.pdf- CONTENTS The whole course, in one ordered book 整门课,浓缩成一本有序的书 Thirteen teaching weeks - one decision-and-appraisal toolkit 十三个教学周 →一套决策与 appraisal 工具箱 TL;DR. The book follows the exam's own logic - build the 2x2 (W1-3: name the design, count frequency, pick the measure, then say what it means), name the bias (W4-6: selection, confounding & standardisation, measurement error and its direction), judge the study (W7-8: the critical-appraisal pipeline and RCTs), then synthesise & screen (W9-13: screening & diagnostic tests, systematic reviews and meta-analysis, Bradford Hill causation) - before turning it all into marks with the glossary, practice bank and exam decoder. TL;DR. 本书遵循考试自身的逻辑 -- 建 2×2 (W1-3:说出设计、计算频率、选指标,再说出它意味着什么)、说出偏倚 (W4-6: selection、confounding & standardisation、测量误差及其方向)、评判研究(W7-8: critical-appraisal 流程 与 RCTs),然后综合 & 筛查(W9-13: screening & 诊断检验、systematic reviews 与 meta-analysis、Bradford Hill causation) -- 最后借助术语表、习题库与考试解码器把这一切转化成分数。 Ch Topic Core ideas Build the 2x2 . Weeks 1-3 Study types the design tree · ecological / cross-sectional / case-control / cohort / RCT · well- vs → poorly-defined study base 2 Measures of frequency ratio / proportion / rate · point & period prevalence · cumulative incidence & incidence rate · person-time → 3 Measures of association RR, RD, OR · which measure for which design · attributable & population-attributable fractions → 4 Interpreting in plain words the lay-sentence stems . the 'two times higher' trap . RRR . NNT framed for a layperson → Name the bias . Weeks 4-6 5 Selection bias selection of cases / controls . healthy-worker . loss to follow-up . how it distorts the measure 6 Confounding, EM & standardisation the 3 criteria . direction of bias . effect modification vs confounding . direct & indirect standardisation → 7 Measurement error differential vs non-differential . bias toward the null · exposure vs outcome misclassification · sens/spec of an instrument → Judge the study . Weeks 7-8 8 Critical appraisal the Section-C pipeline . selection - measurement - confounding - chance . what flaw matters most 9 Randomised controlled trials randomisation & blinding · intention-to-treat vs per-protocol . efficacy vs effectiveness[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

文件对整门课的总结非常清楚：

• Weeks 1–3：study types、frequency、association、plain-language interpretation
• Weeks 4–6：selection bias、confounding、effect modification、standardisation、measurement error
• Weeks 7–8：critical appraisal、RCTs
• Weeks 9–13：screening、diagnostic tests、systematic reviews、meta-analysis、Bradford Hill causation[12]Source: asksia-bible-pubh5010-bilingual.pdf- CONTENTS The whole course, in one ordered book 整门课,浓缩成一本有序的书 Thirteen teaching weeks - one decision-and-appraisal toolkit 十三个教学周 →一套决策与 appraisal 工具箱 TL;DR. The book follows the exam's own logic - build the 2x2 (W1-3: name the design, count frequency, pick the measure, then say what it means), name the bias (W4-6: selection, confounding & standardisation, measurement error and its direction), judge the study (W7-8: the critical-appraisal pipeline and RCTs), then synthesise & screen (W9-13: screening & diagnostic tests, systematic reviews and meta-analysis, Bradford Hill causation) - before turning it all into marks with the glossary, practice bank and exam decoder. TL;DR. 本书遵循考试自身的逻辑 -- 建 2×2 (W1-3:说出设计、计算频率、选指标,再说出它意味着什么)、说出偏倚 (W4-6: selection、confounding & standardisation、测量误差及其方向)、评判研究(W7-8: critical-appraisal 流程 与 RCTs),然后综合 & 筛查(W9-13: screening & 诊断检验、systematic reviews 与 meta-analysis、Bradford Hill causation) -- 最后借助术语表、习题库与考试解码器把这一切转化成分数。 Ch Topic Core ideas Build the 2x2 . Weeks 1-3 Study types the design tree · ecological / cross-sectional / case-control / cohort / RCT · well- vs → poorly-defined study base 2 Measures of frequency ratio / proportion / rate · point & period prevalence · cumulative incidence & incidence rate · person-time → 3 Measures of association RR, RD, OR · which measure for which design · attributable & population-attributable fractions → 4 Interpreting in plain words the lay-sentence stems . the 'two times higher' trap . RRR . NNT framed for a layperson → Name the bias . Weeks 4-6 5 Selection bias selection of cases / controls . healthy-worker . loss to follow-up . how it distorts the measure 6 Confounding, EM & standardisation the 3 criteria . direction of bias . effect modification vs confounding . direct & indirect standardisation → 7 Measurement error differential vs non-differential . bias toward the null · exposure vs outcome misclassification · sens/spec of an instrument → Judge the study . Weeks 7-8 8 Critical appraisal the Section-C pipeline . selection - measurement - confounding - chance . what flaw matters most 9 Randomised controlled trials randomisation & blinding · intention-to-treat vs per-protocol . efficacy vs effectiveness[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

3）Final 最值钱的核心能力

这门课最值钱的不是多会算，而是这 4 个动作：

• 动作 1：先说出研究设计
• 动作 2：从设计推出合法测量
• 动作 3：把数字翻译成“普通人听得懂的话”
• 动作 4：遇到 bias 时一定说方向，不只报名字[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS

文件原话反复强调：

• “绝不只是叫个名，要 define → apply → because”
• “绝不只说某个 bias，要说 likely? differential or not? direction?”[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

4）最核心知识块 1：研究设计 Study Types

这是全卷起点，几乎每个题都从这里开始。[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

你必须会分的 6 类研究

• Ecological
  • 单位是群体，不是个体[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.
• Cross-sectional
  • 单次观察，没有随访[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• Case-control
  • 按结局选人，再回头问暴露[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• Cohort
  • 按暴露分组，向前随访，看发病[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• RCT
  • 本质像 cohort，但关键区别是随机分配暴露/干预[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• Systematic review / meta-analysis
  • 合并多个研究证据[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection

研究设计决策树：考试里怎么快速认

• 先问：
  • 是在描述，还是在比较暴露和结局？
• 再问：
  • 单位是群体还是个体？
  • 有随访吗？
  • 是按暴露分组还是按结局分组？
  • 有没有随机化？[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

高频混淆点

• Cohort vs RCT
  • 唯一关键区别：是否随机化[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.
• Cross-sectional vs Cohort
  • 有没有随访[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.
• Ecological vs Cross-sectional
  • 一个是群体单位，一个是个体单位[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.
• Case-control
  • 关键特征是按结局选人[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

5）最核心知识块 2：从设计选测量

这是全课最容易“白丢分”的地方之一。文件直接说：

• 给 case-control 算 RR
• 或给 cohort 算 OR
• 是最常见错误之一。[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。

设计 → 合法测量

• RCT / cohort
  • 用：
    • RR
    • RD
    • NNT
    • 也可能有 AF / PAF[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• Case-control
  • 用：
    • OR
    • 公式：$$OR=\frac{ad}{bc}$$[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境
• Cross-sectional
  • 用：
    • prevalence ratio[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• Diagnostic test
  • 用：
    • sensitivity
    • specificity
    • PPV
    • NPV[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• SR/MA
  • 通常不强调计算，重点是读图和解释[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

最重要原则

• 先命名设计，再命名测量，再计算
• 因为“选对测量”本身就有分。[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").

6）最核心知识块 3：$2\times 2$ 表

PUBH5010 的很多东西其实都围着 $2\times 2$ 转。[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[12]Source: asksia-bible-pubh5010-bilingual.pdf- CONTENTS The whole course, in one ordered book 整门课,浓缩成一本有序的书 Thirteen teaching weeks - one decision-and-appraisal toolkit 十三个教学周 →一套决策与 appraisal 工具箱 TL;DR. The book follows the exam's own logic - build the 2x2 (W1-3: name the design, count frequency, pick the measure, then say what it means), name the bias (W4-6: selection, confounding & standardisation, measurement error and its direction), judge the study (W7-8: the critical-appraisal pipeline and RCTs), then synthesise & screen (W9-13: screening & diagnostic tests, systematic reviews and meta-analysis, Bradford Hill causation) - before turning it all into marks with the glossary, practice bank and exam decoder. TL;DR. 本书遵循考试自身的逻辑 -- 建 2×2 (W1-3:说出设计、计算频率、选指标,再说出它意味着什么)、说出偏倚 (W4-6: selection、confounding & standardisation、测量误差及其方向)、评判研究(W7-8: critical-appraisal 流程 与 RCTs),然后综合 & 筛查(W9-13: screening & 诊断检验、systematic reviews 与 meta-analysis、Bradford Hill causation) -- 最后借助术语表、习题库与考试解码器把这一切转化成分数。 Ch Topic Core ideas Build the 2x2 . Weeks 1-3 Study types the design tree · ecological / cross-sectional / case-control / cohort / RCT · well- vs → poorly-defined study base 2 Measures of frequency ratio / proportion / rate · point & period prevalence · cumulative incidence & incidence rate · person-time → 3 Measures of association RR, RD, OR · which measure for which design · attributable & population-attributable fractions → 4 Interpreting in plain words the lay-sentence stems . the 'two times higher' trap . RRR . NNT framed for a layperson → Name the bias . Weeks 4-6 5 Selection bias selection of cases / controls . healthy-worker . loss to follow-up . how it distorts the measure 6 Confounding, EM & standardisation the 3 criteria . direction of bias . effect modification vs confounding . direct & indirect standardisation → 7 Measurement error differential vs non-differential . bias toward the null · exposure vs outcome misclassification · sens/spec of an instrument → Judge the study . Weeks 7-8 8 Critical appraisal the Section-C pipeline . selection - measurement - confounding - chance . what flaw matters most 9 Randomised controlled trials randomisation & blinding · intention-to-treat vs per-protocol . efficacy vs effectiveness

为什么 $2\times 2$ 这么重要

• 因为它是以下指标的共同起点：
  • RR
  • RD
  • OR
  • NNT
  • AF
  • PAF[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

你在考试里要养成的习惯

• 看到情境题，马上做这几步：
  • 标出谁是 exposure
  • 标出谁是 outcome
  • 把人数放进 $a,b,c,d$
  • 再从设计决定该抽哪根箭头、算哪个指标[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

7）最核心知识块 4：关联测量 Measures of Association

RR：Risk Ratio

• 用于 cohort / RCT[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• 解释重点：
  • $RR=2$ 表示：
    • twice the risk
    • 或 100% higher risk
  • 绝不能说：two times higher[1]Source: asksia-bible-pubh5010-bilingual.pdfstandardisation。这些都是不受版权保护的范本,平实陈述。每个范例都使用 AskSia 自拟的数字,每个考试风格的情 景都经过改写和重新编号 -- 我们绝不复制任何往届试卷的具体题干或数据。书面状态(开卷、提供公式表、允许计算 器、禁止AI、考试门槛)均已引述并遵守。请在你的 Canvas 站点上核实。 AskSia Library · PUBH5010 · 双语 Bilingual THE BLUEPRINT - THE EXAM BLUEPRINT 60-85% FINAL . OPEN BOOK . EXAM HURDLE Where every mark lives 每一分藏在哪里 An open-book final - formula sheet supplied, so judgement and speed win it 一场 open-book 期末 -- 已提供公式表,所以靠判断与速度取胜 TL;DR. The final is worth 60% or 85% (whichever scheme is greater), it is open book, and the course supplies a "Helpful Formulae" sheet - so the formulas are not the bottleneck. Its make-or-break skills are (1) which measure / decision applies, (2) interpreting a number in plain words (never "two times higher" for RR = 2), and (3) a critical-appraisal pipeline. And you must clear the exam hurdle: ≥50% on the exam itself, or you fail the unit. TL;DR. 期末占 60% 或 85%(两套方案取分高者),它是 open book,而且课程提供一张“Helpful Formulae”表 -- 所以 公式不是瓶颈。它的成败技能是(1)该用哪个指标/该做哪个决策、(2)用大白话解读一个数字(对 RR=2 永远不要说“two times higher"),以及(3) critical-appraisal 流程。而且你必须过 exam hurdle:考试本身≥50%,否则整门课挂科。 60-85% FINAL EXAM WEIGHT 期末考权重 2h+10 WRITING + READING MIN 书写 + 阅读 分钟 OPEN BOOK . FORMULAE SUPPLIED 开卷 · 提供公式表 ≥50% EXAM HURDLE TO PASS 通过所需的考试 HURDLE The four assessment pieces 四个考核组成部分 Component Weight When Format Final exam 60- Exam 85% period In-person, open book; 2 h + 10 min reading; 3 sections A/B/C, separate booklets; non-prog calculator OK; no AI; HURDLE ≥50% Mid-sem assignment 25% / - Wk 6 0%[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

RD：Risk Difference

• 是绝对差异[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境
• 常见说法：
  • “每 100 个暴露者多出多少个病例”[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境

OR：Odds Ratio

• 用于 case-control[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• 解释时要说 odds，不是 risk[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境

NNT

• 关键公式：
  • $$NNT=\frac{1}{RD}$$
  • 若题目给定某人数规模 $k$，还要乘上 $k$
  • 向上取整[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境
• 超高频陷阱：
  • NNT 分母用 RD，不是 RR[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• 解释时要说明：
  • “多少人需要接受干预，才能预防 1 个病例”
  • 要加时间范围[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境

RRR

• $$RRR = 1 - RR$$[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)

AF / PAF

• 文件有提到：
  • AF(暴露者)可由 RD 与暴露组风险推得
  • PAF 小计算会考“still occur / would not occur”这种措辞陷阱[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。
• 这里你最要注意的是：
  • 小计算不仅考会不会代数，还考你有没有看清题目动词[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

8）最核心知识块 5：通俗解释 Lay interpretation

这是这门课一个非常特别、而且真的会单独给分的点。[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").

评分喜欢的表达方式

• 你必须把结果说成：
  • 谁（exposed）
  • 相比谁（comparison group）
  • 结局怎样
  • 差多少[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").

安全表达模板

• RR 型：
  • “暴露组的风险是对照组的 $X$ 倍”
  • 或“暴露组风险高 $(X-1)\times100%$”[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• RD 型：
  • “每 100 个暴露者中，额外多出 $Y$ 个病例”[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境
• NNT 型：
  • “需要给 $N$ 个人干预，才能在该时间范围内避免 1 个病例”[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境

头号陷阱：RR = 2 怎么说

• 正确：
  • twice the risk
  • 100% higher risk
• 错误：
  • two times higher
• 原因：
  • “higher”说的是“多出来的部分”
  • “two times higher”字面意思会变成 3 倍风险[1]Source: asksia-bible-pubh5010-bilingual.pdfstandardisation。这些都是不受版权保护的范本,平实陈述。每个范例都使用 AskSia 自拟的数字,每个考试风格的情 景都经过改写和重新编号 -- 我们绝不复制任何往届试卷的具体题干或数据。书面状态(开卷、提供公式表、允许计算 器、禁止AI、考试门槛)均已引述并遵守。请在你的 Canvas 站点上核实。 AskSia Library · PUBH5010 · 双语 Bilingual THE BLUEPRINT - THE EXAM BLUEPRINT 60-85% FINAL . OPEN BOOK . EXAM HURDLE Where every mark lives 每一分藏在哪里 An open-book final - formula sheet supplied, so judgement and speed win it 一场 open-book 期末 -- 已提供公式表,所以靠判断与速度取胜 TL;DR. The final is worth 60% or 85% (whichever scheme is greater), it is open book, and the course supplies a "Helpful Formulae" sheet - so the formulas are not the bottleneck. Its make-or-break skills are (1) which measure / decision applies, (2) interpreting a number in plain words (never "two times higher" for RR = 2), and (3) a critical-appraisal pipeline. And you must clear the exam hurdle: ≥50% on the exam itself, or you fail the unit. TL;DR. 期末占 60% 或 85%(两套方案取分高者),它是 open book,而且课程提供一张“Helpful Formulae”表 -- 所以 公式不是瓶颈。它的成败技能是(1)该用哪个指标/该做哪个决策、(2)用大白话解读一个数字(对 RR=2 永远不要说“two times higher"),以及(3) critical-appraisal 流程。而且你必须过 exam hurdle:考试本身≥50%,否则整门课挂科。 60-85% FINAL EXAM WEIGHT 期末考权重 2h+10 WRITING + READING MIN 书写 + 阅读 分钟 OPEN BOOK . FORMULAE SUPPLIED 开卷 · 提供公式表 ≥50% EXAM HURDLE TO PASS 通过所需的考试 HURDLE The four assessment pieces 四个考核组成部分 Component Weight When Format Final exam 60- Exam 85% period In-person, open book; 2 h + 10 min reading; 3 sections A/B/C, separate booklets; non-prog calculator OK; no AI; HURDLE ≥50% Mid-sem assignment 25% / - Wk 6 0%[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

9）最核心知识块 6：Bias 和 direction

这是 Section C 的灵魂，也是 open-book 最容易决定分数高低的部分。[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

记住：考试不奖励你只报 bias 名字

• 你必须回答：
  • 这个偏倚在这里可能吗？
  • 是 differential 还是 non-differential？
  • 它把估计往哪个方向推？
  • 影响大不大？[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

10）Selection bias

Section C 一定要会讲 selection。[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[12]Source: asksia-bible-pubh5010-bilingual.pdf- CONTENTS The whole course, in one ordered book 整门课,浓缩成一本有序的书 Thirteen teaching weeks - one decision-and-appraisal toolkit 十三个教学周 →一套决策与 appraisal 工具箱 TL;DR. The book follows the exam's own logic - build the 2x2 (W1-3: name the design, count frequency, pick the measure, then say what it means), name the bias (W4-6: selection, confounding & standardisation, measurement error and its direction), judge the study (W7-8: the critical-appraisal pipeline and RCTs), then synthesise & screen (W9-13: screening & diagnostic tests, systematic reviews and meta-analysis, Bradford Hill causation) - before turning it all into marks with the glossary, practice bank and exam decoder. TL;DR. 本书遵循考试自身的逻辑 -- 建 2×2 (W1-3:说出设计、计算频率、选指标,再说出它意味着什么)、说出偏倚 (W4-6: selection、confounding & standardisation、测量误差及其方向)、评判研究(W7-8: critical-appraisal 流程 与 RCTs),然后综合 & 筛查(W9-13: screening & 诊断检验、systematic reviews 与 meta-analysis、Bradford Hill causation) -- 最后借助术语表、习题库与考试解码器把这一切转化成分数。 Ch Topic Core ideas Build the 2x2 . Weeks 1-3 Study types the design tree · ecological / cross-sectional / case-control / cohort / RCT · well- vs → poorly-defined study base 2 Measures of frequency ratio / proportion / rate · point & period prevalence · cumulative incidence & incidence rate · person-time → 3 Measures of association RR, RD, OR · which measure for which design · attributable & population-attributable fractions → 4 Interpreting in plain words the lay-sentence stems . the 'two times higher' trap . RRR . NNT framed for a layperson → Name the bias . Weeks 4-6 5 Selection bias selection of cases / controls . healthy-worker . loss to follow-up . how it distorts the measure 6 Confounding, EM & standardisation the 3 criteria . direction of bias . effect modification vs confounding . direct & indirect standardisation → 7 Measurement error differential vs non-differential . bias toward the null · exposure vs outcome misclassification · sens/spec of an instrument → Judge the study . Weeks 7-8 8 Critical appraisal the Section-C pipeline . selection - measurement - confounding - chance . what flaw matters most 9 Randomised controlled trials randomisation & blinding · intention-to-treat vs per-protocol . efficacy vs effectiveness

你要检查什么

• case / control 或 exposed / unexposed 是否来自同一个 study base
• non-response / loss to follow-up 有多大
• 损失比例是否大到足以影响结果[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

表达时要写的点

• 损失了多少百分比
• 这个比例算大还是小
• 丢失的人是谁
• 如果不知道丢的是谁，就要老实说：
  • 方向无法判断[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

高频句子

• “若无信息说明哪些人丢失，则 direction cannot be determined.”[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

11）Confounding

这是全课最核心之一。[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[12]Source: asksia-bible-pubh5010-bilingual.pdf- CONTENTS The whole course, in one ordered book 整门课,浓缩成一本有序的书 Thirteen teaching weeks - one decision-and-appraisal toolkit 十三个教学周 →一套决策与 appraisal 工具箱 TL;DR. The book follows the exam's own logic - build the 2x2 (W1-3: name the design, count frequency, pick the measure, then say what it means), name the bias (W4-6: selection, confounding & standardisation, measurement error and its direction), judge the study (W7-8: the critical-appraisal pipeline and RCTs), then synthesise & screen (W9-13: screening & diagnostic tests, systematic reviews and meta-analysis, Bradford Hill causation) - before turning it all into marks with the glossary, practice bank and exam decoder. TL;DR. 本书遵循考试自身的逻辑 -- 建 2×2 (W1-3:说出设计、计算频率、选指标,再说出它意味着什么)、说出偏倚 (W4-6: selection、confounding & standardisation、测量误差及其方向)、评判研究(W7-8: critical-appraisal 流程 与 RCTs),然后综合 & 筛查(W9-13: screening & 诊断检验、systematic reviews 与 meta-analysis、Bradford Hill causation) -- 最后借助术语表、习题库与考试解码器把这一切转化成分数。 Ch Topic Core ideas Build the 2x2 . Weeks 1-3 Study types the design tree · ecological / cross-sectional / case-control / cohort / RCT · well- vs → poorly-defined study base 2 Measures of frequency ratio / proportion / rate · point & period prevalence · cumulative incidence & incidence rate · person-time → 3 Measures of association RR, RD, OR · which measure for which design · attributable & population-attributable fractions → 4 Interpreting in plain words the lay-sentence stems . the 'two times higher' trap . RRR . NNT framed for a layperson → Name the bias . Weeks 4-6 5 Selection bias selection of cases / controls . healthy-worker . loss to follow-up . how it distorts the measure 6 Confounding, EM & standardisation the 3 criteria . direction of bias . effect modification vs confounding . direct & indirect standardisation → 7 Measurement error differential vs non-differential . bias toward the null · exposure vs outcome misclassification · sens/spec of an instrument → Judge the study . Weeks 7-8 8 Critical appraisal the Section-C pipeline . selection - measurement - confounding - chance . what flaw matters most 9 Randomised controlled trials randomisation & blinding · intention-to-treat vs per-protocol . efficacy vs effectiveness[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

Confounder 的 3 条标准

• 文件明确提到你必须能给出这三条：[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
  • 是 outcome 的独立危险因素
  • 不在 causal pathway 上
  • 在暴露组和未暴露组之间分布不均

Confounding 的方向怎么判

• 文件给了核心逻辑：
  • 方向 = 该因素对 outcome 的影响 × 它在各暴露组的分布[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

必会两个方向例子

• 若 confounder：
  • 提高 outcome
  • 且在暴露组更常见
  • 则估计会被往上推[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)
• 若 confounder：
  • 提高 outcome
  • 且在未暴露组更常见
  • 则估计会被往下推[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

Effect modification 和 confounding 不一样

• Effect modification
  • 各 strata 的结果真的不同
  • 这是“真实差异”，不要合并
  • 应分层报告[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• 这点考试很容易拿来区分概念。

12）Measurement error / Misclassification

这是另一个 Section C 高频。[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[12]Source: asksia-bible-pubh5010-bilingual.pdf- CONTENTS The whole course, in one ordered book 整门课,浓缩成一本有序的书 Thirteen teaching weeks - one decision-and-appraisal toolkit 十三个教学周 →一套决策与 appraisal 工具箱 TL;DR. The book follows the exam's own logic - build the 2x2 (W1-3: name the design, count frequency, pick the measure, then say what it means), name the bias (W4-6: selection, confounding & standardisation, measurement error and its direction), judge the study (W7-8: the critical-appraisal pipeline and RCTs), then synthesise & screen (W9-13: screening & diagnostic tests, systematic reviews and meta-analysis, Bradford Hill causation) - before turning it all into marks with the glossary, practice bank and exam decoder. TL;DR. 本书遵循考试自身的逻辑 -- 建 2×2 (W1-3:说出设计、计算频率、选指标,再说出它意味着什么)、说出偏倚 (W4-6: selection、confounding & standardisation、测量误差及其方向)、评判研究(W7-8: critical-appraisal 流程 与 RCTs),然后综合 & 筛查(W9-13: screening & 诊断检验、systematic reviews 与 meta-analysis、Bradford Hill causation) -- 最后借助术语表、习题库与考试解码器把这一切转化成分数。 Ch Topic Core ideas Build the 2x2 . Weeks 1-3 Study types the design tree · ecological / cross-sectional / case-control / cohort / RCT · well- vs → poorly-defined study base 2 Measures of frequency ratio / proportion / rate · point & period prevalence · cumulative incidence & incidence rate · person-time → 3 Measures of association RR, RD, OR · which measure for which design · attributable & population-attributable fractions → 4 Interpreting in plain words the lay-sentence stems . the 'two times higher' trap . RRR . NNT framed for a layperson → Name the bias . Weeks 4-6 5 Selection bias selection of cases / controls . healthy-worker . loss to follow-up . how it distorts the measure 6 Confounding, EM & standardisation the 3 criteria . direction of bias . effect modification vs confounding . direct & indirect standardisation → 7 Measurement error differential vs non-differential . bias toward the null · exposure vs outcome misclassification · sens/spec of an instrument → Judge the study . Weeks 7-8 8 Critical appraisal the Section-C pipeline . selection - measurement - confounding - chance . what flaw matters most 9 Randomised controlled trials randomisation & blinding · intention-to-treat vs per-protocol . efficacy vs effectiveness

你要会区分

• Non-differential misclassification
• Differential misclassification

Non-differential 的方向

• 文件反复强调：
  • non-differential → toward the null[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• 如果测得 $RR>1$，而存在非差异性暴露误差：
  • 说明测得值被往 1 拉
  • 真效应其实更强[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

Differential 的方向

• 可以是：
  • up
  • down
  • either direction[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

Recall bias

• 属于典型 differential exposure measurement problem
• 方向可双向[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

Outcome measurement

• 客观标准（reference standard）是优点
• 受暴露信息影响的主观结局报告，是 differential bias 风险[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

13）Critical appraisal：Section C 作答流水线

这是你最需要背模板的地方。[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

标准流水线

• 1. Hypothesis & study type
  • 先说研究问题
  • 叫出设计
  • 点出 exposure、outcome、study base[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• 2. Selection
  • 是否同一 study base
  • non-response / loss to follow-up 是否明显
  • direction 是否可判断[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• 3. Exposure measurement
  • 是否可能 recall bias
  • 是否 differential
  • direction 如何[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• 4. Outcome measurement
  • 是客观还是主观
  • 有无 reference standard
  • 暴露知情是否影响结局判定[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• 5. Confounding
  • 三条件是否满足
  • 方向
  • 是否被设计或分析控制[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• 6. Effect modification
  • 分层结果是否不同
  • 若不同，不要 pool[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• 7. Chance & causation
  • CI 是否跨 null
  • 再结合 Bradford Hill 看因果[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

Section C 最重要的写作原则

• 不要泛泛列清单
• 要写成：
  • 这个偏倚在本研究里是否可能
  • 大小如何
  • 往哪边偏
  • 为什么[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

14）RCT：一整块必须掌握

文件专门给了 RCT 回忆清单。[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

为什么 RCT 是 gold standard

• 因为 randomisation 在期望意义上能平衡已知和未知 confounders
• 但前提是样本量足够[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

四个常考保护机制

• Randomisation
  • 主要防 confounding[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• Allocation concealment
  • 防入组时的 selection bias[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• Blinding
  • 防 measurement / performance bias[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• Placebo
  • 让 blinding 更可行[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

ITT vs per-protocol

• ITT 更好
• 因为按分配组分析，保留 randomisation
• 若把不依从者剔除，会破坏 randomisation，带来 selection bias / confounding[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

Loss to follow-up

• 要写：
  • 损失比例
  • 是否差异性
  • 方向是否可知[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

RCT 常用测量

• RR
• RRR
• RD
• NNT
• 注意：
  • 绝不能从 RR 算 NNT[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

15）Screening / Diagnostic tests

这是后半课程的重要内容。[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

诊断 $2\times 2$ 的方向必须搞清楚

• 文件说得非常明确：
  • 行是 test result
  • 列是真实状态 / reference standard[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

四个核心指标

• Sensitivity
  • 在患病者中，谁测出阳性
• Specificity
  • 在未病者中，谁测出阴性
• PPV
  • 测出阳性的人里，真有病的比例
• NPV
  • 测出阴性的人里，真没病的比例[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

你必须懂的区别

• Sensitivity / specificity
  • 是检验本身属性[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• PPV / NPV
  • 会随 prevalence 改变[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

高频考点

• 把 row / column 方向弄对，本身就是高频得分点[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

16）Systematic Review / Meta-analysis / Forest plot

这是课程后段的重要综合内容。[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

SR 和 MA 的区别

• Systematic review
  • 是方法
  • 有 protocol
  • 有系统检索
  • 有明确纳入排除标准
  • 有 risk-of-bias appraisal
  • 可重复[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS
• Meta-analysis
  • 是可选的统计合并步骤
  • 把多个研究估计值做加权平均[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

森林图怎么读

• 每个 box
  • 一个研究估计值
  • 大小表示权重
• 横线
  • 置信区间
• diamond
  • pooled estimate
• 竖线
  • null line，通常 RR/OR = 1[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

你考试里要会说什么

• pooled CI 是否跨 1
• 是否存在 heterogeneity
• 是否可能有 publication bias[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

高频提醒

• 不是所有 SR 都有 MA
• 研究差异太大时，不应强行 pool[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

17）Bradford Hill causation

文件强调这是“因果判断指南”，不是机械打勾表。[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

关键条目

• strength
• consistency
• specificity
• temporality
• biological gradient
• plausibility
• coherence
• experiment
• analogy[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

最重要的一条

• temporality
  • 原因必须先于结果发生
  • 文件说它几乎是近乎必要的一条[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

核心态度

• association 不等于 causation[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

18）必须背的定义/句子清单

下面这些，是你特别值得背成“考场模板”的：

研究设计

• Cohort
  • 按暴露分组并随访，比较发病[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• Case-control
  • 按结局选人，再回顾暴露[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• Cross-sectional
  • 单时间点观察，无随访[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• Ecological
  • 以群体为分析单位[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.
• RCT
  • 按随机分配干预后随访[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。

Confounding

• 一个变量若：
  • 是 outcome 的独立危险因素
  • 不在因果路径上
  • 在暴露组间分布不均
• 则可构成 confounding[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

Effect modification

• 不同 strata 的效应不同
• 这是“真实差异”，应分层报告，不应合并[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

Non-differential misclassification

• 两组中误差程度相同
• 常使估计偏向 null[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

ITT

• 按原随机分配组分析，以保留 randomisation 的优势[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

Systematic review

• 按 protocol 对全部证据做系统、可重复的检索与评价[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

Meta-analysis

• 在 systematic review 内，对研究结果做统计合并[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

19）最重要的公式清单

这门课虽然不是公式课，但你还是要知道“公式对应什么场景”。

OR

• $$OR=\frac{ad}{bc}$$
• 用于 case-control[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。

RRR

• $$RRR=1-RR$$[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)

NNT

• $$NNT=\frac{1}{RD}$$
• 向上取整[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境

诊断指标

• 具体公式在课程提供的 formula sheet 上
• 你更重要的是要搞清：
  • sens/spec 看列
  • PPV/NPV 看行[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

Forest plot 判读

• 如果 pooled CI 跨过 1
  • 证据仍不确定 / 不显著[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

20）全课最常见失误清单

这个部分你一定要在考前反复看。

失误 1：设计认错

• cohort 和 RCT 混
• cross-sectional 和 cohort 混
• ecological 和 cross-sectional 混[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.

失误 2：测量选错

• case-control 算 RR
• cohort / RCT 算 OR[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。

失误 3：只给数字，不解释

• 文件明确说：
  • 开卷考试里，过程和解释都单独给分[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").

失误 4：RR = 2 说成 “two times higher”

• 这是文件点名的经典扣分陷阱[1]Source: asksia-bible-pubh5010-bilingual.pdfstandardisation。这些都是不受版权保护的范本,平实陈述。每个范例都使用 AskSia 自拟的数字,每个考试风格的情 景都经过改写和重新编号 -- 我们绝不复制任何往届试卷的具体题干或数据。书面状态(开卷、提供公式表、允许计算 器、禁止AI、考试门槛)均已引述并遵守。请在你的 Canvas 站点上核实。 AskSia Library · PUBH5010 · 双语 Bilingual THE BLUEPRINT - THE EXAM BLUEPRINT 60-85% FINAL . OPEN BOOK . EXAM HURDLE Where every mark lives 每一分藏在哪里 An open-book final - formula sheet supplied, so judgement and speed win it 一场 open-book 期末 -- 已提供公式表,所以靠判断与速度取胜 TL;DR. The final is worth 60% or 85% (whichever scheme is greater), it is open book, and the course supplies a "Helpful Formulae" sheet - so the formulas are not the bottleneck. Its make-or-break skills are (1) which measure / decision applies, (2) interpreting a number in plain words (never "two times higher" for RR = 2), and (3) a critical-appraisal pipeline. And you must clear the exam hurdle: ≥50% on the exam itself, or you fail the unit. TL;DR. 期末占 60% 或 85%(两套方案取分高者),它是 open book,而且课程提供一张“Helpful Formulae”表 -- 所以 公式不是瓶颈。它的成败技能是(1)该用哪个指标/该做哪个决策、(2)用大白话解读一个数字(对 RR=2 永远不要说“two times higher"),以及(3) critical-appraisal 流程。而且你必须过 exam hurdle:考试本身≥50%,否则整门课挂科。 60-85% FINAL EXAM WEIGHT 期末考权重 2h+10 WRITING + READING MIN 书写 + 阅读 分钟 OPEN BOOK . FORMULAE SUPPLIED 开卷 · 提供公式表 ≥50% EXAM HURDLE TO PASS 通过所需的考试 HURDLE The four assessment pieces 四个考核组成部分 Component Weight When Format Final exam 60- Exam 85% period In-person, open book; 2 h + 10 min reading; 3 sections A/B/C, separate booklets; non-prog calculator OK; no AI; HURDLE ≥50% Mid-sem assignment 25% / - Wk 6 0%[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境

失误 5：NNT 用 RR 算

• 错
• 必须用 RD[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

失误 6：bias 只报名字，不说方向

• 会严重丢分[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

失误 7：Section C 写成泛泛而谈

• 必须结合本研究，写 likely + direction + size[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

失误 8：把 effect modification 当成 confounding

• effect modification 是真实分层差异，不是噪音[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

失误 9：诊断题把 row / column 搞反

• 这是 screening 部分的经典错误[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。

失误 10：SR 和 MA 不区分

• SR 是方法
• MA 是可选的统计合并[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

21）如果你现在时间不够，优先级怎么排

按文件内容，我建议你这样排：

第一优先级：一定拿住

• 研究设计分类[10]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. Section A is fast recognition - match a cue phrase to one of the six designs, then a couple of short calcs (RRR, NNT, ITT). Section B is one scenario worked end-to-end. name+justify - 2×2 - right measure + lay interpret - NNT - mini-calcs. Pick the measure from the design, show working, and close with a lay sentence. TL;DR. Section A 是快速识别 -- 把一句线索短语匹配到六种设计之一,然后做几道简短计算(RRR、NNT、ITT)。Section B 是把一个情景从头到尾做完:命名+论证→ 2×2→正确测量→通俗解读→ NNT→小计算。从设计出发挑测量、展示过 程,并以一句通俗的话收尾。 A Section A - identify the design + short calcs ASection A -- 辨认设计 +短计算 AskSia Library · PUBH5010 · 双语 Bilingual A1 NAME THE 6 STUDY TYPES [12: 1 word each] Prompt (paraphrased, our exposure-outcome = 'energy drinks - insomnia'). Name the design behind each vignette (more than one may share a type): Prompt(已改写,我们的暴露一结局=‘能量饮料→失眠 energy drinks → insomnia')。说出每个小情景背后的设计 (可能不止一个共享同一类型): Model-answer skeleton - the structure that scores. 范答骨架 -- 能得分的结构。 · "20 regions, each with region-level rates of both" - Ecological (group unit). “20个地区,各有两者的地区级率”→ Ecological(群体单位)。 - “按自报摄入量分组,随访8年,比较 incidence"→ Cohort。 “随机分配到新 vs 常规摄入,加以随访”→ RCT(randomisation 是与 cohort 唯一的区别)。 · "interviewed once, asked if they've ever had insomnia" - Cross-sectional (no follow-up). “访谈一次,询问其是否 曾经失眠”→ Cross-sectional(无随访)。 “失眠患者vs 隔壁诊所的患者,询问其既往摄入”→ Case-control(界定不良的基础人群)。 “30 个研究合并成一个合并估计”→ Systematic review 与 meta-analysis。 What earns the marks. one correct label per vignette. Use the study-design tree: group vs individual unit, follow-up or not, randomised or not, grouped by exposure or by outcome. 得分要点。每个小情景给出一个正确标签。使用研究设计树(study-design tree):群体还是个体单位、是否随访、是 否随机化、按暴露分组还是按结局分组。 Trap. cohort vs RCT (randomisation only); cross-sectional vs cohort (no follow-up = cross-sectional); ecological vs cross-sectional (group vs individual unit); forgetting to label the case-control base well- or poorly-defined. Trap. cohort vs RCT(仅看是否随机化); cross-sectional vs cohort (无随访 = cross-sectional); ecological vs cross-sectional (群体 vs 个体单位);忘记标注 case-control 的研究基础(base)是定义良好还是定义不良。 AskSia Library · PUBH5010 · 双语 Bilingual [4: show working] A2 % RISK REDUCTION FROM AN RCT Prompt (our numbers). A daily-supplement arm has 12/600 events; usual-care 22/580. What is the % reduction in risk? Prompt(我们的数字)。每日补充剂组有12/600例事件;常规护理组 22/580。风险降低的百分比是多少? Model-answer skeleton - the structure that scores.[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
• 设计 → 测量 对应关系[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• RR / OR / RD / NNT 的解释[5]Source: asksia-bible-pubh5010-bilingual.pdfMigraine + Migraine - Total Beanville (exp) Decaftown (unexp) Total (c) Choose & compute the measure [5]: cohort - risk ratio. (c) 选择 & 计算测量 [5]: cohort →risk ratio。 Ie = 200 Iue = 6 = 0. 430, RR = 0. 590 = 1. 37 0. 430 (if it were a case-control, compute OR = ad/bc instead - the choice is itself marked). (d) Lay interpretation [5]: "People in Beanville had a 37% higher risk of migraine than people in Decaftown" (or "1. 37 times the risk"). (d) 通俗解读[5]:“Beanville 的人发生偏头痛的风险比 Decaftown 的人高 37%”(或“1. 37 倍的风险 (1. 37 times the risk)") . (e) NNT-style harm-prevented [5]: (e) NNT 式的避免危害(harm-prevented) [5]: RD = |0. 590 - 0. 430| = 0. 160, NNT1 1 == 6. 25, NNT30 = 0. 160 30 0. 160 = = 187. 5 ~188 (round UP; "188 would need to switch to low coffee to avoid 30 migraines"). ! Which-measure & rounding traps in Section B Section B 中的选指标与取整陷阱 Using OR for a cohort (or RR for a case-control) loses the choice mark. For NNT: use RD (not RR) in the denominator, multiply by k, and round UP (you cannot treat a fraction of a person). For the lay sentence, never say "1. 37 times higher" loosely - "1. 37 times the risk" or "37% higher". 对一个 cohort 用 OR(或对一个 case-control 用 RR)会丢掉选择分。算 NNT 时:分母用 RD(不是 RR),乘以 k,并向上取 整(你不能治疗一个人的零头)。写通俗句时,绝不松散地说“1. 37 times higher” -- 应说“1. 37 times the risk”或“高 37%"o AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 36 + 40 MARKS . SPEED PLUG-INS + THE LONG STUDY Section B mini-calcs & Section C critical appraisal Section B 小计算与 Section C critical appraisal TL;DR. Section B ends with 1-mark 'no working' mini-calcs (a-e) - rapid plug-ins where wording (PAF "still occur" vs "would not occur") is the trap. Section C hands you a long study and asks you to appraise it: comment on selection . exposure . outcome . confounding - each as strengths + weaknesses, and for every flaw say likely here?, differential?, and which direction. - TL;DR. Section B 以1分的“无需过程”小计算(a-e)收尾 -- 快速代入,其中措辞(PAF 的“still occur 仍会发生” vs “would not occur 不会发生”)是陷阱。Section C 给你一项长研究并要求你评价它:就 selection · exposure · outcome · confounding 各自评论 -- 每项都要谈优点+缺点,并且对每个缺陷都要说:在此处可能吗 ?、 是否差异性 ?、 朝哪个方 向? 。[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境
• RR=2 的表达陷阱[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境
• bias direction 基本规则[3]Source: asksia-bible-pubh5010-bilingual.pdfAIHW / ABS AIHW/ABS 数据源 Key AU population-health data (registries, surveillance, vital stats) - judge completeness & validity. AskSia Library · PUBH5010 · 双语 Bilingual Term (EN) 中文 One-line meaning SMR / SIR (O/E) 标准化死 Observed+Expected×100 亡/发病 ⽐ from indirect standardisation; >100 = more than expected. Effect modification 效应修饰 (交互) Real, biological: strata differ from each other - do NOT pool; report by stratum. ✓ How to spend a glossary term in the exam 在考试里如何用好一个术语 Never just name it. Define - apply - because. e. g. "This is non-differential misclassification (define); the faulty meter mislabels exposure equally in both groups (apply); so the RR is biased toward the null - and since the measured RR>1, the true effect is even larger (because). " The three-move sentence, with the direction, is what the rubric pays for. 绝不只是叫个名。定义→应用→因为。例如:“这是 non-differential misclassification(定义);那台故障的仪表在两组 中同等地误标暴露(应用);因此 RR 被偏向无效值 -- 而既然测得的 RR>1,真实效应其实更大(因为)。”这套带方向的三步句 式,正是评分细则付钱的东西。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK ALL CHAPTERS . EXAM REHEARSAL Practice bank: rehearsing the fixed A/B/C exam 习题库:演练固定的 A/B/C 考卷 Every card = paraphrased prompt - scoring skeleton - marking note - trap 每张卡片=改写后的题干→得分骨架→评分注记→ 陷阱 TL;DR. The PUBH5010 final has the same three-section skeleton every year (2017-2025): Section A = name the six study types + short calcs; Section B = one scenario worked end-to-end (type & justify - 2x2 - the right measure - lay interpretation - NNT - mini-calcs); Section C = a long study critically appraised (selection . exposure . outcome . confounding). It is open book and a formula sheet is supplied - so the win is not formulas but which-measure-when, lay interpretation, and a bias-direction argument. Every card below is one rep.[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

第二优先级：Section C 大分块

• selection
• exposure measurement
• outcome measurement
• confounding
• effect modification
• chance / causation[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

第三优先级：后段综合题

• screening / diagnostic accuracy[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• SR / MA / forest plot[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS
• Bradford Hill[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

22）最实用的考前冲刺法

文件本身就在强调：open-book 考试要练“翻页速度”和“决策速度”。[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

你可以这样做：

第一步：做一个“解码器”页

• 把这几行写在最前面：
  • cohort / RCT → RR, RD, NNT
  • case-control → OR
  • cross-sectional → prevalence ratio
  • diagnostic → sens/spec/PPV/NPV[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

第二步：做一个通俗表达页

• 写上：
  • RR: “$X$ times the risk” / “$(X-1)\times100%$ higher”
  • RR=2: “twice the risk” / “100% higher”
  • 不要写 “two times higher”[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

第三步：做一个 bias-direction 页

• 写上：
  • non-differential → toward null
  • differential → either direction
  • confounding 方向 = effect on outcome × distribution across exposure groups[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

第四步：做一个 Section C 模板页

• 顺序固定：
  • study type
  • selection
  • exposure measurement
  • outcome measurement
  • confounding
  • effect modification
  • chance / causation[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.

第五步：练 3 件事

• 看到题干，10 秒内认设计
• 30 秒内写出该用哪个 measure
• 每算完一个数，立刻补一句 lay interpretation[2]Source: asksia-bible-pubh5010-bilingual.pdfOpen book means the slow part is deciding - so that is what we drill Open book 意味着慢的环节是做决定 -- 所以这正是我们要练的 TL;DR. This is not a formula dump - PUBH5010 is open book and the course already hands you a "Helpful Formulae" sheet, so copying equations onto a crib sheet wins nothing. What the exam actually rewards is which measure / which bias / which direction, the one plain-English sentence a layperson would understand, and a critical-appraisal pipeline for the long study. This book is a bank of those decision rules, worked interpretations and appraisal checklists, each tied to the fixed A / B / C exam structure. The same pages serve you three ways across the thirteen teaching weeks. TL;DR. 这不是公式堆砌 -- PUBH5010 是 open book,而且课程已经给你一张“Helpful Formulae”表,所以把方程抄到小 抄上毫无意义。考试真正给分的,是选哪个指标/哪种偏倚/哪个方向、一个外行能听懂的一句大白话,以及针对那篇长研 究的critical-appraisal 流程。这本书就是这些决策规则、做好的解读与 appraisal 清单的资料库,每一条都绑定在固定的A / B/C 考试结构上。同一批页面在十三个教学周里以三种方式为你服务。 A 1 . LEARN 1 . LEARN 1 . LEARN You haven't done the week's lecture yet. Read a chapter top to bottom. Each idea opens with a plain definition, lands a diagram or a decision table, then a worked example with our own numbers that shows how to reason - build the 2×2, pick the measure, then say what it means. Meet study types, confounding, bias direction and screening here cold. 本周的课你还没上。从头到尾读 一整章。每个概念先以一个平实 的定义(definition)开场,接 着是一张图或一张决策表,然后 是一个用我们自己数字的范例 (worked example),展示如何 推理 -- 构建 2×2、挑测量,再 说它意味着什么。在这里第一次 接触研究类型、confounding、 偏倚方向和 screening。 B 2 . REVISE 2 · REVISE 2 . REVISE You've done the week. Use the grids and the chapter-end recall checklists to self-test: can you name the six study types from a cue, give the three confounding criteria, predict the direction of non-differential error, recite Wilson & Jungner? Tab this book so the right page opens in seconds - open-book speed is a skill you practise. 本周的课你已上完。用各张网格 表和章末的回忆清单 (recall checklists) 自测:你能从一句 线索说出六种研究类型吗?能给 出三条 confounding 判据吗?能 预测非差异性误差的方向吗?能 背出 Wilson & Jungner 吗?给 这本书贴标签,让对的那一页几 秒内就能翻到 -- 开卷的翻阅速 度是一项你要练习的技能。 C 3 . APPLY 3 . APPLY 3 . APPLY You're sitting the paper. Run the which-measure / which-bias decoder (Ch 14) on every prompt: read the cue -> name the design or measure - write the because. With the formulae supplied and a calculator allowed, your edge is decision speed and clean interpretation, not recall. 你正在考这张卷子。对每道题跑 哪个测量 / 哪种偏倚解码器 (which-measure / which- bias decoder) (Ch 14): 读线 索→命名设计或测量→ 写出 because。在公式已提供、允许 使用计算器的情况下,你的优势 是决策速度和干净的解读,而不 是记忆。 AskSia Library · PUBH5010 · 双语 Bilingual ! Read this first: the assessment shape, the hurdle, and the open-book rule 先读这一段:考核形态、hurdle 与 open-book 规则 PUBH5010 is assessed by four pieces: a mid-semester assignment (25% or 0%, a critical appraisal of a paper), online quizzes (10%, best 10 of 12 counted), in-class tutorial tasks (5%, only attemptable if you attend), and the final exam (60% or 85%, whichever scheme is greater). The final is in person, open book, 2 hours writing + 10 minutes reading, in three sections (A / B / C) with separate answer booklets. A non-programmable calculator, pens, rulers and highlighters are permitted; no electronic aids and no Al. A There is an exam hurdle: you must score ≥50% on the exam itself to pass the unit - even a combined mark of 50%+ fails if the exam is below 50%. Always confirm current weights, dates and conditions on your own Canvas site, as details shift between cohorts. PUBH5010 由四个部分评估:一份学期中作业(25% 或0%,对一篇论文的批判性评价)、 在线测验 (online quizzes)(10%,取12 次中最好的10 次计分)、课堂辅导任务(in-class tutorial tasks) (5%,只 有出席才能完成)、以及期末考试(60% 或85%,取两种方案中较高者)。期末为现场、开卷、书写 2小时+阅读10 分钟,分三个 section (A/B/C),答题册分开。允许使用非编程计算器、笔、尺和荧光笔;不得使用电子辅助工具, 禁止 AI。△ 设有考试门槛(hurdle):你必须在考试本身上拿到≥50% 才能通过本单元 -- 即便合计达到50% 以上, 只要考试低于50% 仍不及格。请务必在你自己的 Canvas 站点上确认当前的权重、日期和条件,因为细节会因 cohort 不同而变动。 i How this book was built - the two-layer rule 这本书是怎么搭出来的 -- 两层规则 The method canon here is standard, widely-published introductory epidemiology - Webb, Bain & Page (Essential Epidemiology), the standard study-design and measure definitions, Wilson & Jungner screening principles, the Bradford Hill criteria and Mantel-Haenszel / standardisation. These are non-copyrightable canon, stated plainly. Every worked example uses AskSia-invented numbers, and every exam-style scenario is paraphrased and re-numbered - we never reproduce a past-paper's specific stem or data. Book status quoted and honoured (open book, formula sheet supplied, calculator allowed, no AI, exam hurdle). Verify on your Canvas site. 这里的方法学范本是标准的、广为出版的流行病学入门内容 -- Webb, Bain & Page (Essential Epidemiology)、标准 的研究设计与测量定义、Wilson & Jungner 筛查原则、Bradford Hill 准则以及 Mantel-Haenszel /[6]Source: asksia-bible-pubh5010-bilingual.pdfTL;DR. PUBH5010 期末每年(2017-2025)都是同样的三段骨架:Section A = 说出六种研究类型+简短计算;Section B = 一个情景从头到尾做完(类型 &论证→2×2→正确的测量→通俗解读→NNT→小计算);Section C = 对一项长研究 做批判性评价(selection · exposure · outcome · confounding)。它是开卷且提供公式表 -- 所以得分不在公式,而在何 时用哪个测量、通俗解读,以及偏倚方向的论证。下面每张卡片都是一次练习。 ★ What the exam asks here - the format you are rehearsing 这里考试考什么 -- 你正在演练的格式 The 60% (or 85%) final is in-person, open book, 2 hours writing + 10 min reading, with a 'Helpful Formulae' sheet supplied and a non-programmable calculator - no electronic aids, no Al. There is an EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit, even if your combined mark is ≥50%. Three sections in separate booklets: A (24) . B (36) . C (40) - budget your time by marks (~1 min per mark, leaving reading time for the long Section C stem). 这场 60%(或 85%)期末考是线下、开卷、2小时书写+10 分钟阅读,提供一张‘Helpful Formulae' 公式表和一个不可 编程计算器 -- 不许电子辅助,不许AI。有一道 EXAM HURDLE:即便你的综合分≥50%,也必须在考试本身得分≥50%才 能通过该单元。三节分装在不同答题册:A(24) · B(36) · C(40) -- 按分值分配时间(约1分钟/分,为冗长的 Section C 题干留出阅读时间)。 P. 1 How to read each card - the marking model P. 1如何读每张卡片 -- 评分模型 Because the formula sheet is in front of you, markers cannot reward you for knowing a formula - they reward choosing the right one, showing the working, and interpreting the number in plain words. A "5-mark calculate" item is almost always 1 mark for picking the measure, 3 for correct lines of working, 1 for the answer - so an unexplained right number can still drop marks. Every card gives you the skeleton the rubric looks for, then the trap that zeroes a careless answer. 因为公式表就在你面前,评分者无法因你记住一个公式而给分 -- 他们奖励的是选对那一个、展示运算过程,以及用平实的话 解读这个数字。一道“5分计算”题几乎总是1分给挑对测量、3分给正确的运算步骤、1分给答案 -- 所以一个没有解释的正 确数字仍可能丢分。每张卡片都给你评分标准(rubric)所寻找的骨架,再给出那个会让粗心答案归零的陷阱(trap)。 1 Name the design / measure first. Markers reward the labelled choice - say cohort - risk ratio, case-control - odds ratio by name before you compute. AskSia Library · PUBH5010 · 双语 Bilingual 先叫出设计/测度。评分者奖励有标注的选择 -- 在计算之前,先点名说出 cohort → risk ratio、case-control → odds ratio. 2 Show every line of working. Open book means the method is marked, not the recall; an answer with no working caps your marks even if it is right. 展示每一步运算。开卷意味着评的是方法而非记忆;一个没有过程的答案,即便正确也会给你的分封顶。 3 Interpret in a lay sentence. A number with no plain-English reading loses the interpretation marks - use the stem in P. 2. 用一句通俗话解释。一个数字若没有通俗英文的读法,会丢掉解释分 -- 用 P. 2 里的句式。 4 For bias, give direction. Never just name a bias; say likely here?, differential or not?, and which way it pushes the estimate (up / down / either / toward null). 对 bias,要给方向。绝不只叫出一个 bias 的名字;要说在此是否可能 ?、 是否差异性?,以及它把估计值推向哪一方(上/下/任 一/ 偏向无效值)。 5 Count answers against marks. A "[3]" comment wants three scoring points; a 1-mark mini-calc wants the number only (no working). 把答案点数对着分值数。一个标着“[3]”的评论要三个得分点;一个1分小计算只要那个数字(不要过程)。 P. 2 The lay-interpretation stems - memorise the verbs P. 2外行解读的题干 -- 背熟那些动词 Section B always asks you to "interpret in plain words a layperson understands. " The marks live in a complete sentence with four parts: exposed group - how much - outcome - comparison group. Drop the comparison group and you lose marks. Section B 总是要求你“用外行能懂的平实话语解读”。分数藏在一个含四部分的完整句子里:暴露组→多大程度→结局→ 比较组。漏掉比较组就会丢分。 Measure Value (our number) Correct lay sentence RR > 1 RR = 1. 37 "The exposed had a 37% higher risk of the outcome than the unexposed" (or "1. 37 times the risk").[7]Source: asksia-bible-pubh5010-bilingual.pdfquizzes 整张试卷的引擎浓缩成一张图:构建 暴露 × 结局 的 2×2,然后每一个测度都从单元格 a,b,c,d 辐射而出 -- 各组的风险、cohort 或 RCT 的risk ratio (RR) 5 risk difference (RD), case-control #x] odds ratio (OR)。Section B 的情境本质上只有一个要求: 把它画出来、挑对那支箭、说清那个数字意味着什 么。 AskSia Library . PUBH5010 . XXia Bilingual Exposure scope = the 13-week spine: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . measurement error . critical appraisal . RCTs . outbreaks . systematic reviews & meta-analysis . screening. Always confirm on Canvas. 两套评分方案 -- 取分高者:(5% tutorials + 5%-10% quizzes + 25% assignment + 60%-65% exam) = ( 试占 85%-90%)。无论哪种,考试都占主导。可考范围= 13 HET: study types . measures of frequency . measures of association . selection bias . confounding, effect modification & standardisation . 测量误差 · critical appraisal · RCTs · 暴发疫情 · systematic reviews & meta-analysis . screening. the 以 Canvas 为准。 The fixed A / B / C skeleton 固定的 A/ B/ C 骨架 Section What it asks Marks A - short Name 6 study types from cue vignettes + short calcs (% risk reduction, NNT) + RCT-concept Qs (ITT, blinding) B - scenario One narrative - name & justify the design, draw the 2×2, calculate the right measure, interpret in lay words, NNT, then 1-mark "no-working" mini- calcs C - long study A realistic cohort or case-control - critically appraise: selection . exposure measurement · outcome measurement · confounding (direction) ✓ The one habit that wins this exam 赢下这场考试的那一个习惯 For every prompt, name the concept first, then write the because. "Selected on the outcome" - case- control - odds ratio; "divided by exposure, followed forward" - cohort - risk ratio; "a factor linked to both exposure and outcome" - confounding (check 3 criteria + direction); "error the same in both groups" + non-differential - bias toward the null; "is this test any good" - sensitivity / specificity / PPV. 对每道题,先命名概念,再写出because。“按结局选 取”→ case-control → odds ratio;“按暴露分组、 向前随访”→ cohort → risk ratio;“一个与暴露和结 局都相关的因素”→ confounding(核对 3条判据+ 方向);“两组中同样的误差”→非差异性→偏向 null (bias toward the null);“这个检验好不好”→ sensitivity / specificity / PPV. AskSia Library · PUBH5010 · 双语 Bilingual ≈ ★ The documented marker trap - never "two times higher" 有据可查的评分陷阱- -永远不要说“two times higher” For RR = 2, the correct lay sentence is "the exposed group had twice the risk / a 100% higher risk" - not "two times higher" (which would imply three times the risk). "X times higher" adds the baseline back on; "X times the risk" does not. This is a marked-down error every year. Ch 4 drills the safe stems; the open- book win is having them ready to copy. 对于 RR =2,正确的通俗句子是“暴露组有 twice the risk (两倍风险)/风险高 100%” -- 而不是“two times higher"(那会暗示风险是三倍)。“X times higher”把基线又加回去了;“X times the risk”则没 有。这是每年都被扣分的错误。Ch 4 反复操练安全句 式;开卷的优势在于把它们备好可直接照抄。 AskSia Library · PUBH5010 · 双语 Bilingual CONTENTS

23）给你一个“最终回忆清单”

你考前最后一轮，就背这个：

• 设计 → 测量
  • cohort/RCT → RR, RD, NNT
  • case-control → OR
  • cross-sectional → prevalence ratio
  • diagnostic → sensitivity / specificity / PPV / NPV[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• 安全解释
  • “$X$ times the risk”
  • “$(X-1)\times100%$ higher”
  • RR=2 = twice the risk
  • 不说 two times higher[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• Bias direction
  • non-differential → toward null
  • differential → either direction
  • confounder direction = effect × distribution[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一[14]Source: asksia-bible-pubh5010-bilingual.pdf一 标出 2×2 数据所在的位置,以便你能立刻画出。 · Underline the verb in mini-calcs (still occur vs prevented; treat vs prevent). 在小计算里给动词划线(still occur vs prevented;treat vs prevent). · Triage the high-mark Section-C parts first. 优先处理高分值的 Section-C 部分。 对每个情境,在页边叫出设计 → 它锁定下游的每一 个测度。 ● 标出 2×2 数据所在的位置,以便你能立刻画出。 ● 在小计算里给动词划线(still occur vs prevented; treat vs prevent). ● 优先处理高分值的 Section-C 部分。 i What to put on your open-book reference 在你的 open-book 参考资料上放什么 · The decoder D. 1 (design - measure) - the formula sheet won't pick for you. 解码器 D. 1(设计→测度) -- 公式表不会替你挑。 · Lay stems + the RR=2 trap (D. 2). 通俗句式 + RR=2 陷阱(D. 2)。 · The appraisal pipeline (D. 3) + the 3 confounding criteria + the direction logic. 评价流水线(D. 3)+ 3 条 confounding 标准 + 方向 逻辑。 · Bias-direction one-liners: non-differential - null; exposure-misclassification - RR toward 1 (true value more extreme). Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 · A worked 2×2 showing RR/RD/NNT/AF/PAF in one pass. 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. 解码器 D. 1(设计 →测度) -- 公式表不会替你挑。 ● 通俗句式+ RR=2 陷阱(D. 2)。 ● 评价流水线(D. 3)+3 条 confounding 标准 + 方向 逻辑。 ● Bias 方向一句话:非差异性→ 无效值;暴露错分 → RR 偏向 1(真值更极端)。 ● 一个算好的 2×2,一次走完展示 RR/RD/NNT/AF/PAF. AskSia Library · PUBH5010 · 双语 Bilingual ★ Final recall checklist - the whole unit on one breath 最终回忆清单 -一口气过完整门课 · Design - measure: cohort/RCT-RR,RD,NNT; case-control-OR; cross-sectional-prevalence ratio; diagnostic-sens/spec/PPV/NPV. 设计→ 测度:cohort/RCT→RR,RD,NNT;case-control→OR;cross-sectional→prevalence ratio;诊断 ->sens/spec/PPV/NPV. 安全地通俗表述:“X times the risk”或“(X-1)×100% higher";RR=2= twice the risk,绝不说“two times higher”。 · Bias direction: non-differential++toward null (+ sign); differential-either way; confounder = effectxdistribution.
• Section C 流水线
  • selection
  • exposure
  • outcome
  • confounding
  • effect modification
  • chance / causation[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
• RCT
  • randomisation
  • allocation concealment
  • blinding
  • placebo
  • ITT > per-protocol[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• Diagnostic
  • sens/spec 看列
  • PPV/NPV 看行[9]Source: asksia-bible-pubh5010-bilingual.pdf★ Recall checklist - RCTs 回忆清单 -- RCTs · Why gold standard: randomisation balances known & unknown confounders in expectation (needs adequate n; check Table 1). 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 · Four distinct safeguards: randomisation (confounding) · allocation concealment (selection bias at enrolment) . blinding (measurement/performance bias) · placebo (enables blinding). Comment on concealment and blinding separately. 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 ITT > per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 I Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 · Phases I-IV (safety -> efficacy - RCT - post-marketing); CONSORT flow diagram + CASP RCT checklist for appraisal. I Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图 +CASP RCT 清单。 · Measures: RR & RRR (ratio), RD (absolute), NNT = 1/RD then xk, round up - never NNT from RR. 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR算 NNT。 ● 为何是金标准:randomisation 在期望意义上均衡已知与未知的 confounder(需足够的 n;查 Table 1)。 ● 四种各不相同的防护:randomisation(confounding)· allocation concealment(入组时的 selection bias)· blinding(测量/实施 bias)· placebo(使 blinding 成为可能)。分别评论 concealment 与 blinding。 · ITT >per-protocol:在被分配到的组里分析;剔除不依从者会破坏 randomisation → selection bias/confounding。 不依从 → bias 偏向无效值(ITT 低估)。 ● Loss to follow-up:陈述损失的 %,判断量级,说明方向是否可知;差异性损失才是危险所在。 ● Phases I-IV(safety → efficacy →RCT → 上市后);评价用 CONSORT 流程图+ CASP RCT 清单。 ● 测度:RR 与 RRR(比值)、RD(绝对)、NNT=1/RD再 xk,向上取整 -- 绝不从 RR 算 NNT。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 11 . SCREENING & TEST EVALUATION - WEEK 11 . SCREENING & TEST EVALUATION CH 9 . WEBB, BAIN & PAGE . DR RACHEL FARBER The diagnostic 2x2: columns are accuracy, rows are predictive value 诊断 2×2:列是准确度,行是预测值 Sensitivity / specificity / PPV / NPV - and which one to quote sensitivity / specificity / PPV / NPV -- 以及该引用哪一个 TL;DR. Lay the test result down the rows and the truth (the reference standard) across the columns. Sensitivity and specificity read down the columns - they are fixed properties of the test (of the diseased, who tests +; of the well, who tests -). PPV and NPV read across the rows - they answer the patient's question ("I tested +, do I have it?") and they move with prevalence. Getting columns vs rows the right way round is the single most-marked skill here. TL;DR. 把检验结果放在行(rows)上,把真相(金标准 reference standard) 放在列(columns) 上。Sensitivity 和 specificity 是沿列读取的 -- 它们是检验的固定属性(在患病者中,谁测出+;在健康者中,谁测出 一)。PPV 和 NPV 是沿 行读取的 -- 它们回答患者的问题(“我测出+,我真得病了吗?”),并且会随 prevalence 变化。把列与行的方向搞对,是 这里最常被打分的单项技能。 ★ What the exam asks here 这里考试考什么 Test evaluation feeds the open-book final (60%/85%, 2 h + 10 min reading, "Helpful Formulae" sheet supplied, no Al, ≥50% exam hurdle). The signature task is a Section B diagnostic-accuracy sub-question: build the 2×2 from sens/spec and a prevalence, calculate sens/spec/PPV/NPV, then interpret one in plain words; sometimes an apparent-prevalence 1-mark mini-calc in B(9). Because the formulae are given, marks turn on which measure answers the question (a test property vs a post-test probability) and a clean lay sentence. Budget ~1. 2 min/mark. 诊断试验的评价喂给开卷期末考(60%/85%,2 小时+10分钟阅读,提供“Helpful Formulae”公式表,禁用 AI,≥50% 考试 hurdle)。标志性任务是一道 Section B 诊断准确性子问:由 sens/spec 与一个 prevalence 构建 2×2,计算 sens/spec/PPV/NPV,然后用大白话解释其中一个;有时 B(9)会出一道1分的表观患病率(apparent-prevalence)小计 算。因为公式是给定的,分数取决于哪个测度回答该问题(是一项试验属性还是一个验后概率)以及一句干净的通俗句子。 按约 1. 2 分钟/分配时。
• SR/MA
  • SR 是方法
  • MA 是 pooling[15]Source: asksia-bible-pubh5010-bilingual.pdfWEEK 10-11 . SYNTHESIS & CAUSATION WEEK 10-11 . SYNTHESIS & CAUSATION CH 16 . WEBB, BAIN & PAGE Systematic reviews, meta-analysis & reading a forest plot systematic reviews、meta-analysis 与读懂 forest plot From a pile of studies to one pooled estimate - and how to read it 从一堆研究到一个 pooled estimate -- 以及如何读它 TL;DR. A systematic review (SR) is the method - a protocol-driven, reproducible search and appraisal of all the evidence on one question. A meta-analysis (MA) is the optional quantitative step inside it: a weighted average of the study estimates, shown as a forest plot. SR/MA of RCTs sits at the top of the evidence hierarchy - but only if it is reproducible, checks heterogeneity, and watches for publication bias. TL;DR. Systematic review (SR)是方法 -- 对某一问题的全部证据进行的、由方案驱动、可重复的检索与评价。Meta- analysis (MA)是其内部可选的定量步骤:对各研究估计值的加权平均,以 forest plot 呈现。RCT 的 SR/MA 位于证据等级 的顶端 -- 但前提是它可重复、检查 heterogeneity,并警惕 publication bias。 ★ What the exam asks here 这里考试考什么 SR/MA seeds Section A (name the design when a stem says "N studies were identified and combined" - systematic review & meta-analysis) and feeds the appraisal sub-parts of Section C (read a forest plot: does the pooled CI cross 1?, is there heterogeneity?, is there publication bias?). The exam is open book with a Helpful Formulae sheet supplied and no AI, so the marks are in the SR-vs-MA distinction and the worded read-out of the plot, not in any calculation. Remember the hurdle: ≥50% on the exam itself or you fail the unit. SR/MA 在 Section A 埋题(当题干说“识别并合并了N个研究”时叫出该设计→ systematic review 与 meta- analysis),并喂给 Section C 的评价子问(读一张 forest plot:合并 CI 是否跨过1 ?、 是否存在 heterogeneity ?、 是否存 在 publication bias?)。考试是开卷并提供一张 Helpful Formulae 公式表且禁用 AI,所以分数在于 SR 与 MA 的区分和 对图的文字读出,而不在任何计算。牢记hurdle:考试本身须≥50% 否则该单元不及格。 10. 1 Systematic review vs narrative review vs meta-analysis 10. 1Systematic review vs narrative review vs meta-analysis Definition. A narrative (traditional) review is an expert's informal summary - no protocol, no documented search, prone to the author's selection. A systematic review follows a pre-registered protocol: a focused PICO question, a comprehensive search, explicit inclusion/exclusion criteria, a risk-of-bias appraisal of every included study, and a transparent synthesis - so it is reproducible. A meta-analysis is the statistical pooling step within an SR: it is not always done (you should not pool apples and oranges). Definition. 叙述性(传统)综述(narrative review)是专家的非正式总结 -- 无方案、无记录在案的检索,易受作者取舍影 响。Systematic review 遵循预先注册的方案(protocol):一个聚焦的 PICO 问题、全面检索、明确的纳入/排除 (inclusion/exclusion) 标准、对每项纳入研究的 risk-of-bias 评价,以及透明的综合 -- 因此它可重复。Meta-analysis 是 SR 内部的统计合并步骤:并非总要做(不应把苹果和橘子合并)。 Feature Narrative review Systematic review (± meta-analysis) Question Broad, informal Focused - PICO AskSia Library · PUBH5010 · 双语 Bilingual Feature Narrative review Systematic review (± meta-analysis) Search Unstated, selective Comprehensive & documented (PRISMA flow) Study selection[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS
• Bradford Hill
  • temporality 最关键[16]Source: asksia-cheatsheet-pubh5010.pdfW10 . CAUSATION SR = a structured, reproducible synthesis (PICO question, protocol, comprehensive search, explicit in/exclusion, risk-of-bias appraisal of included studies). MA = the quantitative pooling step - a weighted average of study estimates. Forest plot: each box = a study estimate (size = weight) + CI line; the diamond = the pooled estimate; the vertical line = the null (RR/OR=1). Read off: which studies favour exposure/control, do CIs cross 1, is the pooled CI significant. Heterogeneity via 12 and the Q- test (fixed- vs random-effects); publication bias via funnel-plot asymmetry (small negative studies missing). SR/MA of RCTs sits at the top of the evidence hierarchy; PRISMA documents the study flow. Bradford Hill (a guide to causal judgement, not a tick- box): strength . consistency . specificity . temporality (cause precedes effect - the near-necessary one) . biological gradient (dose-response) . plausibility . coherence . experiment (does removing exposure cut risk?) . analogy. Association # causation; the RCT is the strongest causal design. SR # MA: the systematic review is the method (the structured search + appraisal); the meta-analysis is the pooling step that may or may not follow. Always check heterogeneity before pooling - if studies are too dissimilar, a pooled estimate misleads. A wide pooled CI crossing the null line means the combined evidence is still inconclusive. OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 open-book companion . confirm on the current PUBH5010 unit outline . @ 2026 independent . not affiliated with the University of Sydney Bias PUBH5010 Epidemiology Methods and Uses SYDNEY SCHOOL OF PUBLIC HEALTH . USYD . POSTGRADUATE OPEN-BOOK COMPANION Sem 1 2026 . SIDE 1 OF 2 Exam hurdle · 60-85% final SIDE 1/2 WHICH-MEASURE-WHEN . Measures of frequency . The 2x2 + RR/OR/RD/NNT/AF/PAF . Lay-interpretation stems + the RR=2 trap . Study-design classifier . Hypothesis types OPEN BOOK . FORMULA SHEET SUPPLIED Compiled by AskSia . mapped to the PUBH5010 syllabus . asksia. ai/cheatsheet/usyd-pubh5010 0 . How to Use This READ FIRST * The PUBH5010 final is OPEN BOOK (2 h writing + 10 min reading) and the course supplies a "Helpful Formulae" sheet (prevalence, incidence, RR, RD, OR, NNT, AF/PAF, sens/spec/PPV/NPV). So the formulas are not the bottleneck - speed and judgement are . No electronic aids, no AI; a non-programmable calculator, pens, rulers & highlighters are permitted. & EXAM HURDLE: you must score ≥50% on the exam itself to pass the unit - even a combined mark ≥50% fails if the exam is below 50%. The exam is worth 60% or 85% of the unit (whichever scheme is greater), so it dominates the grade either way. This sheet carries what the formula sheet can't: which measure to use when, the lay-interpretation stems markers reward, the bias-direction logic, and the Section-C appraisal pipeline (Side 2). Treat it as a decision deck you flip through in the reading time. Side 1 = the quantitative core (frequency, the 2×2, which-measure, interpretation, the design classifier). Side 2 = the biases + appraisal (selection, measurement, confounding/EM, diagnostics, screening, RCTs, SR/MA, Bradford Hill, the pipeline). SIA > The fixed exam skeleton repeats every year (2017-25): A = name 6 study types + short calcs; B = one scenario ++ type, 2x2, the right measure, lay-interpret, NNT; C = a long study - critical appraisal. Budget by marks = A 24 / B 36 / C 40 -+ ~1. 2 min/mark. 1 . Measures of Frequency W2 . SECTION B + MINI-CALCS

24）如果你只想知道“哪里最容易拿高分”

• 最容易拉开分差的不是算术，而是：
  • 设计认对[13]Source: asksia-bible-pubh5010-bilingual.pdf→ Synthesise & screen . Weeks 9-13 10 Screening, SR/MA & causation diagnostic 2×2 · sens/spec/PPV & prevalence . Wilson & Jungner . lead/length-time . forest plots . Bradford Hill → AskSia Library . PUBH5010 . XXia Bilingual → → Ch Topic Core ideas Walk in ready 11 Glossary every term, bilingual, one line each - built for fast open-book lookup → 12 Practice bank the recurring A/B/C prompt types, drilled with model reasoning skeletons → 13 Exam decoder the which-measure / which-bias decoder . timing by marks . the open-book lookup → plan i Why this order - it mirrors the A/B/C paper 为什么是这个顺序 -- 它映射 A/B/C 考卷 Epidemiology is a pipeline you walk, so the book reads as one. First build the 2x2 (Ch 1-4): the question every Section A and the opening of every Section B asks is "what design is this, and what measure does it license?"-and then the plain-words interpretation that carries marks. Next name the bias (Ch 5-7): selection, confounding and measurement error are the heart of Section C, and the examiner wants the direction, not just the name. Then judge the study (Ch 8-9): the appraisal pipeline that organises a long answer, plus RCTs. Finally synthesise & screen (Ch 10): diagnostic tests, evidence synthesis and causal inference. Chapters 4 (interpretation) and 6-7 (confounding and bias direction) are where the open-book marks are won and lost - slow down there. 流行病学是一条你逐步走过的流水线,所以本书也照此来读。首先构建 2×2 (Ch 1-4):每道 Section A 题、以及每道 Section B 的开头都在问“这是什么设计,它许可什么测量?” -- 然后是那句能得分的平实解读。其次命名偏倚(Ch 5-7):selection、confounding 和测量误差是 Section C 的核心,考官想要方向,不只是名称。然后评判研究(Ch 8-9):组织长答案的评价流水线,外加 RCT。最后综合 & 筛查(Ch 10):诊断检验、证据综合和因果推断。第4章 (解读)和第 6-7章(confounding 和偏倚方向)是开卷分数得失之处 -- 在那里放慢速度。 AskSia Library · PUBH5010 · 双语 Bilingual WEEK 1 . STUDY TYPES WEEK 1 . STUDY TYPES CH 1-2 . WEBB, BAIN & PAGE Study types: the decision tree that opens every paper 研究类型:打开每一份试卷的决策树 Descriptive vs analytic, and who decided who got the exposure Descriptive vs analytic,以及由谁决定谁获得 exposure TL;DR. Every exam scenario starts with one move: name the study design. Two questions settle it - (1) does the study compare an exposure with an outcome (analytic) or just count/describe (descriptive)? and (2) how were people sorted - by exposure (RCT if randomised, cohort if not), by outcome (case-control), at one point in time (cross-sectional), or only as a group (ecological)? The design then fixes the measure you are allowed to compute: cohort/RCT - RR; case-control - OR. TL;DR. 每道考试情境都从同一个动作开始:说出研究设计。两个问题就能定下来 -- (1)该研究是把一个 exposure 与一个 outcome 做比较(analytic),还是只是计数/描述(descriptive)?以及(2)人是怎么分组的 -- 按 exposure(随机化即 RCT,未随机化即 cohort)、按 outcome (case-control)、在某一时间点(cross-sectional),还是只作为群体 (ecological)? 设计随即定下你被允许计算的指标:cohort/RCT → RR; case-control →OR。
  • 测量选对[4]Source: asksia-bible-pubh5010-bilingual.pdf● A:设计树 -- 群体 vs 个体单位、有无随访 ?、 是否随机化 ?、 按暴露还是按结局分组。RRR = 1-RR;ITT 让每个人留 在其被分配到的组。 ● B 选哪个测度:RCT/cohort → RR, RD, NNT;case-control → OR = ad/bc;cross-sectional → prevalence ratio. ● 通俗句:暴露者→多多少→结局→比较组。RR=2 ⇒ ‘twice the risk',绝不说‘two times higher'。 ● NNT:1/RD,xk,向上取整;说明时间范围。 ● 小计算:PAF ‘still'=(1-PAF);AF(暴露者)= RD/le;prevalence=已有+新发。 ● C 流水线:选择(量化损失)→暴露/结局(差异性 vs 非差异性+方向)→ confounding(3 条标准;方向来自效应×分 布)。 ● Confounder 方向:抬高结局+在未暴露者中更常见→RR 偏向下;抬高结局+在暴露者中更常见→偏向上。 AskSia Library · PUBH5010 · 双语 Bilingual EXAM MORNING . THE DECODER EXAM MORNING . THE DECODER OPEN-BOOK COMPANION . A / B / C The exam-morning decoder: which measure, and how to say it 考试当天的解码器:选哪个指标,以及怎么说它 The formula sheet is supplied - bring the decisions, not the formulas 公式表已提供––带上决策,而不是公式 TL;DR. The final is open book and a "Helpful Formulae" sheet is supplied, so the arithmetic is free - the marks live in which measure the design lets you compute, how to phrase the result for a layperson, and a clean critical-appraisal pass. This page is the one-glance decoder: read the study - pick the measure - write the safe lay sentence. TL;DR. 期末为开卷且提供“Helpful Formulae”公式表,所以算术是白送的 -- 分数藏在哪个测量是该设计许可你计算的、如 何把结果向外行措辞,以及一遍干净的批判性评价(critical-appraisal)。本页是一目了然的解码器:读研究→挑测量→写 安全的通俗句子。 ★ What the exam asks here 这里考试考什么 The PUBH5010 final is worth 60% (or 85%) of the unit, is open book, runs 2 hours + 10 min reading, and is the same A / B / C shape every year: A = name the study types + short calcs; B = one scenario - type + justify, 2x2, the right measure, the lay interpretation, NNT, and the 1-mark mini-calcs; C = a long study - critical appraisal. A non- programmable calculator is allowed; no Al, no electronic aids. A Remember the HURDLE: you must score ≥50% on the exam itself to pass the unit - a combined mark ≥50% still fails if the exam is below 50%. PUBH5010 期末考占该单元的60%(或85%),是开卷,历时2小时+10分钟阅读,且每年都是同一个A/B/ C 形态:A = 叫出研究类型+短计算;B=一个情境→类型+论证、2×2、正确测度、通俗解释、NNT,以及那些1分小计算;C= 一个长篇研究→批判性评价。允许一个不可编程计算器;不许 AI,不许电子辅助。△ 牢记 HURDLE:你必须在考试本身得 分≥50%才能通过该单元- -综合分≥50% 但考试低于50% 仍然不及格。 D. 1 The which-measure-when decoder D. 1何时用哪个指标的解码器 Read the study, then read off the measure. Name the design first (Week 1 tree); the legitimate measure follows automatically. Computing an RR for a case-control study (or an OR for a cohort) is the single most common B(3) error. 读研究,然后读出测量。先命名设计(Week 1树);合法的测量随即自动得出。给 case-control 研究算 RR(或给 cohort 算OR)是最常见的单项 B(3)错误。 If the study is . . . Use this measure Watch for RCT / cohort (groups by exposure, followed, incidence) RR & RD - NNT (AF, PAF too) Cohort # RCT (randomisation only); do NOT report the OR; NNT = 1/|RD| round UP Case-control (selected on outcome, exposure asked back)
  • lay interpretation 说对[11]Source: asksia-bible-pubh5010-bilingual.pdfRR = 2 RR = 2. 0 "Twice the risk" / "100% higher risk" - NEVER 'two times higher' (see trap). RR < 1 RR = 0. 74 "The exposed had a 26% lower risk (RRR = 1-RR) than the unexposed" - a protective association. OR (case- control) OR = 3. 0 "Cases had 3 times the odds of exposure compared with controls" (odds, not risk - OR approximates RR only when the outcome is rare). RD RD = 0. 16 "16 extra cases per 100 exposed, over the study period, compared with the unexposed" (absolute, per-head). NNT NNT = 7 "7 people need the intervention to prevent one case" (round UP; state the time-frame). AskSia Library · PUBH5010 · 双语 Bilingual ! The documented marker trap: "two times higher" 有据可查的评分陷阱:“two times higher" For RR = 2, writing "two times higher risk" is marked wrong. "Higher" means the excess (RR-1), so "two times higher" literally claims RR = 3. Say "twice the risk" or "100% higher risk". Only "N times the risk" (RR) and "(RR-1) x100% higher" are safe - never mix the two. 对于 RR = 2,写“风险高两倍(two times higher risk)”会被判错。“higher”指的是超出部分(RR-1),所以“high two times”字面意思是 RR = 3。应说“twice the risk(两倍的风险)”或“100% higher risk(高 100%)”。只有“N times the risk"(RR)和“(RR-1)×100% higher”是安全的 -- 绝不混用两者。 ✓ Open-book speed kit Open-book 速查工具包 Tab your supplied formula sheet by section. Pre-mark the which-measure rule (RCT/cohort-RR & RD & NNT; case- control-OR; cross-sectional-prevalence ratio) and the bias-direction rules next to it. In the exam you are racing the clock, not your memory - the cards on the next two pages drill the moves you must make fast. 给你那张提供的公式表按节加标签。预先标出选哪个测度的法则(RCT/cohort→RR 与 RD 与 NNT;case- control→OR;cross-sectional→prevalence ratio)以及紧挨着的bias 方向法则。在考试里你是在和时钟赛跑,不是和记忆 赛跑 -- 接下来两页的卡片会把你必须快速做出的动作反复演练。 AskSia Library · PUBH5010 · 双语 Bilingual REVISION . A/B/C PRACTICE BANK - REVISION . A/B/C PRACTICE BANK 24 + 36 MARKS . THE BREAD AND BUTTER Section A & B drills: name the design, then work one scenario Section A 与 B 训练:先说出设计,再做一个情境
  • bias direction 讲清[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一
  • Section C 写得有逻辑，不泛泛而谈[8]Source: asksia-bible-pubh5010-bilingual.pdfSection C 给你一项长研究并要求你评价它。每次都按同一流水线、同一顺序走,对每种偏倚说出三件事:在此处可能吗? → 差异性还是非差异性?→ 朝哪个方向(up / down / either)? 1 Hypothesis & study type. Efficacy / causality / accuracy / magnitude? Name the design (Week 1 tree) - the legitimate measure (decoder D. 1). State the study factor, outcome factor, and study base in person / place / time. 假设与研究类型。Efficacy / causality / accuracy / magnitude?叫出设计(Week 1 树)→合法的测度(解码器 D. 1)。以人 / 地/时陈述研究因素、结局因素和研究基础人群。 2 Selection. Same study base for cases/exposed and controls/unexposed? Quantify non-response / loss (e. g. 25% of cases lost = big enough to matter; 3% = too small); if no info on who was lost, the direction cannot be determined - say so. 选择。病例/暴露者与对照/未暴露者是否来自同一研究基础人群?量化无应答 /损失(例如 25% 的病例丢失=大到要紧;3%= 太小);若没有关于谁丢失的信息,则方向无法判定 -- 就这么说。 - 3 Measurement - exposure. Recall bias (differential - either direction) vs general error (non-differential - toward the null, then add the sign of the estimate); blinding, validation, reliability. This is the heaviest sub- question (up to ~15 marks). 测量 -- 暴露。Recall bias(差异性→任一方向)对一般误差(非差异性→偏向无效值,再加上估计值的符号);blinding、 validation、reliability。这是分值最重的子问(最高约15分)。 4 Measurement - outcome. Objective reference standard (a strength) vs subjective reporting swayed by exposure knowledge (- differential, either direction). 测量 -- 结局。客观参考标准(一个长处)对被暴露知情左右的主观报告(→差异性,任一方向)。 5 Confounding. Do the 3 criteria hold (independent risk factor; NOT on the causal pathway; unequally distributed)? Direction = effect on outcome x distribution across exposure groups - up / down. Controlled by design or analysis? Confounding。那3条标准是否成立(独立的风险因素;不在因果通路上;分布不均)?方向=对结局的效应 ×在各暴露组间的分 布→上/下。是否由设计或分析加以控制? 6 Effect modification. Stratum-specific estimates differ? Then do not pool - report by stratum (it is real, not a nuisance). Effect modification。各层别的估计是否不同?那就不要合并 -- 分层报告(它是真实的,不是个累赘)。 7 Chance & causation. CI excluding the null = significant; then weigh Bradford Hill (temporality is the key one). Finally, say which flaws matter MOST (a marked learning outcome). 机遇与因果。CI 排除无效值 =显著;然后权衡 Bradford Hill(temporality 是关键那条)。最后,说清哪些缺陷最要紧(一个被计分 的学习目标)。 ! Generic lists lose marks - commit to likely + direction + size 泛泛罗列会丢分 -- 要落到 likely+ 方向+大小 Naming a bias is not enough. For each one the marker wants: is it likely in THIS study?, which direction would it push the estimate, and - across the whole appraisal - which flaw matters most. Two reasoning chains land on "down" for different reasons: non-differential exposure error with a measured RR > 1 (toward the null), and a confounder that raises the outcome but is commoner in the unexposed. Show the chain, not just the word. 叫出一个 bias的名字还不够。对每一个,评分者要的是:它在本研究中是否可能 ?、 它会把估计值推向哪个方向,以及 --- 纵观整个评价 -- 哪个缺陷最要紧。两条推理链会出于不同理由都落到“下偏”:一是非差异性暴露误差且测得 RR>1(偏 向无效值),二是一个抬高结局但在未暴露者中更常见的 confounder。展示推理链,而不仅是那个词。 AskSia Library · PUBH5010 · 双语 Bilingual D. 4 The clock & your open-book reference D. 4时间分配 & 你的 open-book 参考资料 You have 10 min reading + 2 hours writing for 100 marks ~ ~ 1. 2 min per mark. Budget by marks and protect Section C - it is 40 marks and the easiest to run out of time on. Answer booklets are separate per section. 你有 10 分钟阅读+2 小时书写,应对 100分~ 每分约1. 2 分钟。按分数分配时间并保护 Section C -- 它有 40分,也最 容易时间不够。每个 section 的答题册是分开的。 24mk SECTION A . ~ 29 MIN Section A · 约 29 分钟 36 mk SECTION B . ~ 43 MIN Section B · 约 43 分钟 40 mk SECTION C . ~ 48 MIN Section C · 约 48 分钟 ✓ Reading time (the free 10 min) Reading time (免费的10分钟) 对每个情境,在页边叫出设计→ 它锁定下游的每一 个测度。 一

25）我建议你下一步这样做

你现在可以直接回我一个字母，我马上继续帮你拆：

• A：我先给你整理 六种研究设计 + 怎么秒认
• B：我先给你整理 RR / OR / RD / NNT 全部公式 + 解释模板
• C：我先给你整理 Section C 批判性评价万能答题模板
• D：我先给你整理 screening / sensitivity / specificity / PPV / NPV
• E：我先给你整理 SR / MA / forest plot / Bradford Hill
• F：我直接给你做一份 PUBH5010 final 冲刺背诵版清单

如果你愿意，我下一条可以直接给你做：

• “PUBH5010 期末高频公式 + 定义 + 万能答题句式”一页版
• 或者
• “Section C 40分长题满分模板”

小提醒：如果你想，我也可以把你这门课的复习重点整理成之后能继续接着用的固定清单。