Cox proportional hazards models are regularly used to analyse time-to-event data in prospective epidemiological cohort and case-cohort studies (Supplementary Text). Case-cohort studies are similar to cohort studies, with the exception that full covariate information is only collected for those individuals who develop the disease over follow-up and a randomly selected subgroup of the initial cohort, referred to as the subcohort (see Supplementary Figure S1). In contrast to traditional epidemiology, logistic regression is often used in genome-wide association studies (GWAS) of cohort and case-cohort data to assess the associations of single-nucleotide polymorphisms (SNPs) and disease outcomes, ignoring the time-to-event information in prospective studies (Supplementary Text).1, 2 The reasons for this include the faster computational time of the logistic regression model, the lack of implementation of time-to-event analysis models within most GWAS software and that genetic studies are often combined in multi-study consortia using meta-analysis in which it is convenient to analyse both case–control studies and prospective studies in the same way.
