BB2025 Chap.9 Survival, Epidemiology and Evidence Synthesis
Survival, Epidemiology and Evidence Synthesis
Time-to-event outcomes preserve partial follow-up
Survival analysis combines whether an event occurred with when it occurred. Censoring records that follow-up is known only to a point, allowing event-free participants to contribute available information. Kaplan-Meier estimates survival probabilities across event times. Cox regression estimates hazard contrasts under a proportionality condition.
Neither a censor mark nor a model removes bias when departure depends on changing prognosis.
Synthesis follows appraisal
A systematic review uses a reproducible question, search, eligibility process, appraisal and synthesis. Meta-analysis can combine compatible estimates, but a common numerical scale does not make designs or estimands exchangeable.
Heterogeneity should be interpreted in terms of population, intervention, outcome, follow-up and bias rather than used only as a threshold for selecting a model.
Match evidence to the question
Descriptive epidemiology maps who, where and when, while analytical designs compare groups to examine explanations.
Qualitative research addresses experience, process and meaning through purposive sampling and transparent interpretation. It should be appraised for fit, richness, analytic credibility and grounding rather than against statistical sampling rules designed for prevalence estimation.
What this chapter covers
- 01
Censoring and risk sets
- 02
Kaplan-Meier survival estimates
- 03
Hazard ratios and proportionality
- 04
Descriptive and analytical epidemiology
- 05
Systematic review methods
- 06
Meta-analysis and pooled estimands
- 07
Clinical and statistical heterogeneity
- 08
Qualitative design and credibility
Worked example · free
Decide whether three studies can be pooled
- 3Compare populations, designs and estimands.
- 3Appraise precision and bias separately.
- 3Choose a synthesis and justify its target.
Key terms
- Censoring
- Incomplete event-time information known up to the last observed follow-up without the event recorded.
- Risk Set
- Participants still under observation and event-free immediately before a specified event time.
- Hazard Ratio
- A relative comparison of instantaneous event rates under a fitted time-to-event model.
- Systematic Review
- A reproducible method for identifying, appraising and synthesising evidence for a defined question.
- Meta-analysis
- A statistical synthesis that combines compatible study estimates into a defined pooled quantity.
- Heterogeneity
- Variation across studies in effects, populations, interventions, outcomes, methods or bias processes.
Survival, Epidemiology and Evidence Synthesis FAQ
What does censoring mean in a survival analysis?
It means the exact event time is unknown beyond the participant's last event-free observation. The participant still contributes information up to that point. Censoring is not the same as never experiencing the event, and it can bias inference if leaving follow-up is related to prognosis after the available model information is considered.
Is a hazard ratio the same as a risk ratio?
No. A hazard ratio compares instantaneous event rates among people still at risk under the fitted model, while a risk ratio compares cumulative event probabilities by a specified time. The values can differ substantially, especially when hazards vary over time. Check proportional hazards and report survival probabilities for clinically meaningful time points.
When should studies not be meta-analysed together?
Avoid one pooled estimate when populations, interventions, comparators, outcomes, follow-up or design estimands do not represent a coherent target. Standardising scales cannot remove those differences. Separate synthesis, subgroup reasoning or structured narrative comparison may be more honest, with each study's precision and risk of bias kept visible.
Exam move
Draw a survival timeline with events, censor marks and a changing risk set, then explain why a censored participant still contributes information. Contrast hazard ratio with fixed-time risk in words. For synthesis, make an appraisal grid covering population, intervention, comparator, outcome, follow-up, design and likely bias direction before looking at pooled output.
Practise one qualitative appraisal by asking whether the method fits the question and whether interpretations remain grounded in data and disconfirming material. Read a Kaplan-Meier curve from left to right and record the risk-set size whenever a visual separation appears. Then ask whether late divergence rests on enough participants.
For reviews, write the pooled target in one sentence; if that sentence cannot name a coherent population of studies and effect, replace the average with structured comparison. Compare one fixed-time risk difference with a hazard ratio from the same imagined study and explain why the quantities need not move identically when effects change over follow-up.
Build a miniature evidence table with study estimate, precision, follow-up and dominant bias. Predict the pooled direction before seeing any model. If one large study controls the answer, explain whether its population and methods justify that influence.
Working through Survival, Epidemiology and Evidence Synthesis in BB2025? Sia is AskSia’s AI Statistics tutor — ask any BB2025 Survival, Epidemiology and Evidence Synthesis question and get a clear, step-by-step explanation grounded in how BB2025 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
