Clinical Research, Evidence-Based Practice and Statistics
What this domain tests
This domain tests whether you can read a trial abstract critically: identify the design, the endpoint, the effect size and its precision, and decide whether the result applies to your patient. Calculations of absolute benefit and diagnostic test performance are common.
Core rules
- Phases: I establishes safety and dose; II screens activity; III compares against a standard; IV is post-marketing surveillance.
- Endpoints: overall survival is the gold standard; progression-free and disease-free survival are surrogate endpoints subject to ascertainment bias; response rate is a screening endpoint only.
- Effect measures: hazard ratio for time-to-event data with 95% confidence intervals; a confidence interval excluding 1 corresponds to statistical significance at the 0.05 level. Relative risk reduction must be converted to absolute risk reduction before claiming clinical relevance.
- Analysis populations: intention-to-treat preserves randomisation and is conservative; per-protocol is susceptible to attrition bias. Non-inferiority trials require a pre-specified margin and both intention-to-treat and per-protocol analyses.
- Diagnostic statistics: sensitivity, specificity, positive and negative predictive values. Predictive values depend on prevalence, so a test's performance in a screening population can differ markedly from a tertiary clinic.
- Quality frameworks: CONSORT for trials, PRISMA for systematic reviews, GRADE for certainty of evidence; RECIST 1.1 for response and CTCAE for toxicity grading.
Worked example
A randomised trial reports a hazard ratio for progression-free survival of 0.75 (95% CI 0.62–0.91), p = 0.003, with five-year survival of 60% versus 50%. Reasoning: the hazard ratio shows a 25% relative reduction in the hazard of progression and the confidence interval excludes 1, so the result is statistically significant. Absolute benefit is 10 percentage points, giving a number needed to treat of 1 ÷ 0.10 = 10 patients to prevent one event at five years. That absolute figure, not the relative reduction, frames the conversation with the patient.
Traps
Equating a statistically significant hazard ratio with a clinically meaningful benefit; reading a subgroup result as definitive without interaction testing; comparing predictive values across populations with different prevalence.