Large prospective cohorts like the Canadian Partnership for Tomorrow Project (CPTP) follow participants longitudinally and capture incident cases of disease. In two-phase studies, researchers select a subset of the complete cohort based on observed outcomes and covariates and measure additional, possibly expensive, variables. In the CPTP, blood samples are collected and stored when participants enrol. During follow-up a small fraction of the cohort will present with rare diseases, and researchers often select a matched case-control sample and obtain biological information from their blood samples, seeking to discover biomarkers that predict disease. This presentation will review methods for analyzing rare events based on two-phase studies and discuss areas needing development, including study design, methods for assessing genomic and other risk factors, and predictive modelling. An illustration will be based on predicting leukaemia risk in healthy individuals.

Background Reading:

Case-cohort analysis: Samuelsen, S. O., ÅNestad, H., Skrondal, A. (2007). Stratified Case-Cohort Analysis of General Cohort Sampling Designs. Scandinavian J of Statistics 34, 103-119

Predicting leukaemia: Abelson, S., Collord, G., Ng, S. W. K., et al. (2018). Prediction of acute myeloid leukaemia risk in healthy individuals. Nature 559, 400-404