Research, evidence-based medicine, statistics and scientific communication
This domain tests whether you can read a paper critically, extract the number that matters, and explain it to a committee in three minutes.
Study design and bias
Randomised controlled trial (strongest for causality), cohort (incidence, multiple outcomes), case-control (rare outcomes, recall bias), cross-sectional (prevalence), diagnostic accuracy study. Appraise randomisation and allocation concealment, blinding, completeness of follow-up, intention-to-treat analysis, and pre-registration against selective outcome reporting. Name the bias: selection, information/measurement, confounding, attrition, publication.
Effect measures
Relative risk (cohort), odds ratio (case-control; overstates when the outcome is common), hazard ratio (time-to-event). Absolute risk reduction = control event rate − intervention event rate; NNT = 1/ARR. Confidence intervals express precision; a wide CI means the estimate is compatible with both benefit and harm.
Inference
A p-value is the probability of data at least this extreme assuming the null hypothesis is true - not the probability that the null is true. Type I error is a false positive; type II error a false negative; power = 1 − β. Watch multiplicity in subgroup analyses, non-inferiority margins, and surrogate endpoints that have not been validated against patient-relevant outcomes.
Synthesis and reporting
Systematic review with a pre-specified search, risk-of-bias assessment, heterogeneity (I²), and publication bias assessment. Certainty grading (GRADE) separates evidence quality from recommendation strength. Reporting guidelines: CONSORT (trials), PRISMA (reviews), STROBE (observational).
Worked example
A trial reports mortality 10% with standard care versus 7% with the new drug, RR 0.70, 95% CI 0.52–0.94, p = 0.02, N = 2000. Reasoning: ARR = 3 percentage points, NNT ≈ 33; the CI excludes 1, so the result is statistically significant; but the absolute benefit is modest and cost per avoided death is the governance question. Presenting "70% risk reduction" is relative-only framing and misleads a committee.
Traps
- Confusing RR and OR, or reporting only the relative effect.
- Reading p = 0.04 as proof or p = 0.06 as proof of no effect.
- Ignoring the CI when interpreting a non-significant subgroup.
- Citing a surrogate endpoint as if it were a clinical benefit.