Epidemiology II — Trials, Bias, Causation and Screening

Trials

• RCT: random sequence → allocation concealment (before assignment, prevents selection bias) → blinding (after assignment, prevents performance and detection bias) → ITT analysis.
• Only randomisation controls unknown confounders.
• ARR = CER − EER; RRR = ARR/CER; NNT = 1/ARR. Report absolute effects.
• Field trial = healthy individuals (Salk 1954). Cluster/community trial = groups; design effect = 1 + (m − 1)ICC.
• Crossover = own control with washout. Factorial = combinations. Stepped‑wedge = staggered cluster roll‑out.
• Phases: I (safety, dose), II (efficacy signal), III (confirmatory RCT), IV (post‑marketing; PvPI). CTRI registration; NDCT Rules 2019; CONSORT 2025.

Bias, confounding, causation

• Random error ↓ with sample size; bias does not.
• Selection biases: Berkson's, volunteer, non‑response, healthy worker, attrition, Neyman. Information: recall, interviewer, observer, misclassification. Hawthorne effect.
• Confounder: linked to exposure, causes outcome, not a mediator. Control: randomisation, restriction, matching / stratification, standardisation, regression.
• Effect modification is a finding to report, not a bias to remove.
• Bradford Hill: temporality (essential), strength, consistency, dose–response, plausibility, coherence, experiment, specificity (weakest), analogy.

Evidence synthesis

• EBM: evidence + expertise + patient values. Ask (PICO) → Acquire → Appraise → Apply → Assess.
• Forest plot: squares (weight), lines (CI), diamond (pooled); I² heterogeneity (25/50/75%). Funnel plot for publication bias.
• GRADE: RCTs start high, observational low; certainty ≠ strength of recommendation. PRISMA 2020; PROSPERO.

Screening

• Screening sorts apparently healthy people; diagnosis confirms. Types: mass, high‑risk, multiphasic, opportunistic.
• Wilson–Jungner (1968) + Andermann (2008): important problem, early stage, suitable test, effective treatment, facilities, agreed policy, cost balance, continuing process, equity, evaluation.
• Se = TP/(TP+FN); Sp = TN/(TN+FP); PPV = TP/(TP+FP); NPV = TN/(TN+FN). ↓ prevalence → ↓ PPV, ↑ NPV.
• LR+ = Se/(1 − Sp); LR− = (1 − Se)/Sp. Post‑test odds = pre‑test odds × LR.
• ROC: Se vs 1 − Sp; AUC 0.5–1; Youden J = Se + Sp − 1 (≠ AUC).
• Serial → ↑Sp, ↓Se; parallel → ↑Se, ↓Sp.
• Lead‑time, length, overdiagnosis, volunteer biases → judge by disease‑specific mortality in invited populations.