Biostatistics and Critical Appraisal
Domain 4 - Biostatistics and Critical Appraisal
This domain tests the numerical reasoning behind clinical decisions and the appraisal of published evidence.
Core rules
- From a 2×2 table: sensitivity = TP/(TP+FN), specificity = TN/(TN+FP), PPV = TP/(TP+FP), NPV = TN/(TN+FN). Only PPV and NPV change with prevalence; sensitivity and specificity are properties of the test.
- A p-value is the probability of the observed data (or more extreme) given the null - not the probability that the null is true. Report effect size and confidence intervals alongside.
- Type I error = alpha (false positive); type II = beta; power = 1 − beta. Shrinking the sample lowers power and widens the interval.
- Test choice: chi-square for proportions; t-test or ANOVA for means when assumptions hold; Mann–Whitney or Kruskal–Wallis when they do not; log-rank for survival.
- Appraisal hierarchy: allocation concealment, blinding, intention-to-treat, completeness of follow-up; for meta-analysis check heterogeneity (I²) and funnel-plot asymmetry; GRADE for certainty.
Worked example. A dengue rapid antigen test has 90% sensitivity and 95% specificity. In a clinic with 2% prevalence, per 1000 patients: TP = 18, FN = 2, FP = 49, TN = 931, so PPV = 18/67 ≈ 27%. In a dengue ward with 40% prevalence, PPV rises to roughly 92%. The same kit means different things in the two settings - the error is importing a tertiary-centre validation into primary care.
Traps
- Equating statistical significance with clinical importance; ignoring absolute risk reduction and NNT.
- Reading p = 0.06 as proof of "no effect".
- Applying parametric tests to skewed data without transformation.
- Accepting a pooled estimate without inspecting heterogeneity or publication bias.