91828 Chap.7 Randomisation, Blinding and Bias
Randomisation, Blinding and Bias
Random sequence generation, allocation concealment and blinding protect different stages of a trial. This chapter follows them chronologically and shows how failures can create selection, performance or detection bias. Rather than scoring labels mechanically, it uses actor, mechanism and outcome language to decide whether knowledge of assignment or unequal care could materially move a specific result.
What this chapter covers
- 01
Random sequence generation
- 02
Allocation concealment
- 03
Participant blinding
- 04
Clinician blinding
- 05
Outcome-assessor blinding
- 06
Bias mechanisms
Worked example · free
Separate sequence from concealment
- +1A computer-generated sequence is a credible random method if implemented as described.
- +1Unsealed accessible envelopes do not conceal upcoming assignments.
- +1Recruiters could alter enrolment according to prognosis, disrupting group comparability.
- +1Random generation is therefore a strength, but implementation leaves selection-bias concern.
Key terms
- Random Sequence
- A random sequence creates unpredictable assignments intended to balance prognostic factors between groups.
- Allocation Concealment
- Allocation concealment hides the upcoming assignment until enrolment is irrevocable.
- Performance Bias
- Performance bias arises when groups receive different care or behave differently beyond the intended intervention.
- Detection Bias
- Detection bias arises when outcome measurement differs because assessors know assignment or expectations.
Randomisation, Blinding and Bias FAQ
Is alternation a random method?
No. Alternating assignments, birth dates and record numbers are predictable. Recruiters can anticipate the next group and change who enters. A genuinely random sequence needs secure implementation so the next assignment remains unknown until enrolment is complete.
What if participant blinding is impossible?
The trial can still blind outcome assessors or analysts, use objective measures, standardise contact and co-interventions, and measure adherence and expectations. Appraise which outcome is susceptible and how the remaining knowledge could change behaviour or measurement.
Does double blind tell me enough?
No. Reports use the term inconsistently. Identify whether participants, clinicians, outcome assessors and analysts were masked, how masking was maintained and whether the outcome was vulnerable. The actor and mechanism matter more than the label.
Can baseline imbalance prove randomisation failed?
Not by itself. Chance can produce imbalance even with a valid sequence, especially in small samples. Inspect the allocation method and concealment directly, decide whether the imbalance is prognostically important, and check whether any adjusted analysis was prespecified rather than testing each baseline row for significance.
Which outcomes are most vulnerable to no blinding?
Subjective symptoms, behavioural outcomes and clinician-judged endpoints are often more susceptible to expectations than independently verified events. Vulnerability also depends on co-interventions and assessor discretion. Judge each outcome separately and seek blinded adjudication, objective measures or standardised care where participant masking is impossible.
How does selective reporting create bias?
Choosing outcomes, time points or analyses after results are known changes which evidence readers see. Compare the trial registration or protocol with the report, note unexplained changes and consider whether omitted or newly emphasised results could make the intervention look more favourable or certain.
Assessment move
Draw a trial timeline from eligibility and recruitment through allocation, treatment delivery, follow-up, outcome assessment and analysis. Place each safeguard at the point where it operates. Random sequence generation creates unpredictable assignments; allocation concealment prevents foreknowledge before enrolment; blinding limits behaviour or judgement after assignment.
Use a different colour for each so the terms cannot collapse into one idea. Test sequence methods by sorting them into unpredictable and predictable categories, then explain how a recruiter could exploit each predictable rule. Read envelope or central-system descriptions closely because implementation can subvert a sound sequence.
For blinding, construct an actor table covering participants, clinicians, outcome assessors and analysts. For each actor, state what they know, what behaviour could change and which endpoint is susceptible. Compare a subjective symptom score with independently verified mortality so the judgement remains outcome-specific. Add co-interventions, adherence and expectations to the performance pathway.
When blinding is impossible, propose realistic protections such as equal contact, standardised care, blinded adjudication or objective measurement rather than declaring the trial invalid. Practise baseline-table interpretation without testing randomisation by p-values. Identify large prognostic imbalances, ask whether chance is plausible and check whether adjusted analysis was prespecified.
Compare registration with the final report to detect selective outcomes and time points. For every risk-of-bias paragraph, use an actor, mechanism, affected outcome and confidence consequence. State uncertainty about direction when the report does not support one. In meta-analysis, note whether the studies carrying most weight share the same bias pathway and whether sensitivity analysis changes the pooled effect.
For AT2, convert each major threat into a protocol safeguard located on the visual abstract. Finish by appraising a well-described open-label trial and a poorly described trial labelled double blind. The exercise reinforces that transparent mechanisms, not prestige words, determine confidence.
