The absolute power increased as the underlying effect size increased for both modelling approaches (Table 2 and Supplementary Table S5). These power curves were steeper as the disease incidence and sampling fraction (which increases the sample size) increased for both models. However, for the case-cohort study design, we found a large decrease in power when using the Prentice-weighted Cox model compared with logistic regression. This loss of power worsened as the cumulative disease incidence increased and as the amount of censoring increased over the 20 years (Figure 2).
