We have shown that there is increased power to detect SNP–disease associations using the Cox model instead of logistic regression for the cohort study design and the extent to which it has additional power depends on cumulative disease incidence and the length of follow-up. However, the increase in power using the Cox model was small when the cumulative disease incidence was low or when the non-cases had complete follow-up. We also observed that the ORs and HRs diverge as follow-up time and cumulative disease incidence increase (especially for large underlying effects), that the HRs were more precise, and logistic regression was more computationally efficient. These findings are in line with and extend those from previous studies with the emphasis in the current work being the application of these modelling approaches in genetic association studies.4, 5, 7, 8, 9
