In summary, to minimise computational time in analysing genetic associations in cohort and case-cohort studies, while obtaining appropriate estimates of effect for SNPs of greater interest we propose a two-step procedure. Firstly, logistic regression is used to analyse all SNPs as an initial filtering process and secondly, Cox regression is fitted to those SNPs associated below a pre-defined P-value threshold to avoid inflated estimates of effect as well as to obtain appropriate P-values for these SNPs. However, logistic regression remains a practical alternative to Cox models in cohort and case-cohort studies, especially when the disease incidence is low.
