Research Methodology, Biostatistics, and Evidence Appraisal
Overview
This domain tests whether you can read a paper critically: identify the design, spot bias, compute and interpret test performance, and judge whether the conclusion follows from the data.
Core rules
- Designs: randomized controlled trials for interventions, cohort studies for incidence and prognosis, case-control studies for rare outcomes, cross-sectional studies for prevalence, and diagnostic accuracy studies reported under STARD.
- Test performance: sensitivity = TP/(TP+FN), specificity = TN/(TN+FP), PPV = TP/(TP+FP), NPV = TN/(TN+FN). PPV and NPV depend on prevalence; sensitivity and specificity do not.
- Statistics: the p value is the probability of the observed data (or more extreme) given a true null hypothesis, not the probability that the null is true. Confidence intervals convey precision. Power and sample size depend on effect size, variance and alpha.
- Evidence synthesis: systematic reviews with meta-analysis pool effect estimates; I² quantifies heterogeneity; funnel plots assess publication bias; GRADE rates certainty of evidence.
- Bias: selection, information, confounding, and lead-time bias in screening studies.
Worked example
A new test has 90% sensitivity and 95% specificity. In a population with 1% prevalence, of 10,000 people, 100 have disease and 90 are detected; of 9,900 without disease, 495 are false positives. PPV = 90/(90+495), about 15%. Reasoning: even an excellent test yields mostly false positives at low prevalence, so confirmatory testing is mandatory before treatment decisions. The same test in a 30% prevalence clinic would give a PPV near 89%.
Traps
Quoting PPV without prevalence. Interpreting a p value as the probability the null is true. Confusing statistical with clinical significance. Ignoring lead-time bias in screening survival data.