Summary
Develops cohort study design as the observational design closest to a randomized experiment, walking through two-cohort, single longitudinal, and virtual designs anchored by Framingham and the Nurses' Health Study. Distinguishes risk-based analyses for closed populations from rate-based analyses requiring person-time, Poisson regression, and Cox proportional hazards, and works through examples on discharge against medical advice, environmental tobacco smoke in pregnancy, and antipsychotics and falls. Covers exposure measurement in detail including dichotomous, ordinal, continuous, and compound scales, induction periods, time-varying exposure, and validation in dietary cohorts like the Swedish Mammography Cohort and EPIC.