casecohortdesign

Case-cohort design is a type of prospective epidemiologic study in which a subcohort is selected at baseline and all cases that occur in the full cohort during follow-up are included for analysis. The design was introduced by Prentice in 1986 as a modification of the classical cohort approach to increase efficiency when studying multiple outcomes or costly biomarkers. The subcohort is a random sample of the full cohort, and cases arising outside the subcohort are added as they occur.

In analysis, exposure data are collected for all members of the subcohort and for all cases. Cases that are also members of the subcohort contribute exposure information in both components. Relative risks or hazard ratios are estimated with weighted methods, typically a Cox proportional hazards model using Prentice weights or related schemes. Variants include the Barlow method with robust variance estimators and Langholz–Goldstein two-phase weighting. These methods yield valid estimates under standard assumptions despite only observing exposures on a subset of non-cases.

Advantages include efficiency for biomarker research and the ability to study multiple outcomes without sampling each exposure. It accommodates time-to-event data and allows reuse of stored specimens. Limitations involve the need for correct subcohort sampling, potential biases if exposure data are not missing at random, and the complexity of the weighting scheme and variance estimation.

Applications are common in large prospective cohorts and biobanks, where expensive or limited-endpoint measurements can be obtained on a random subcohort plus all incident cases. The design remains a standard option in epidemiology for efficient estimation of relative risks in time-to-event analyses.