The differences in power between the Cox and logistic regression models were reasonably similar across the RAFs for each of the combinations of cumulative disease incidence and censoring strategy (Supplementary Figures S2–S4). In the secondary simulations, where the survey and random follow-up censoring strategies were applied to the event time distributions from the complete follow-up model, there was less divergence between the HRs and ORs as the rate of censoring increased (Supplementary Table S4). This decreased divergence, when censoring was introduced, led to a small increase in the greater power of the Cox model for the larger cumulative disease incidences (Supplementary Table S4).
