As expected, the power to detect SNP–disease associations increased as the underlying HR increased for both Cox and logistic regression models (Table 1 and Supplementary Table S3). Power also increased as the cumulative disease incidence increased (Table 1 and Supplementary Table S3). The Cox model tended to have more power than logistic regression, and as the cumulative disease incidence increased this difference in power also increased (Figure 1). Furthermore, the increase in power for the Cox model was greater in magnitude as the rate of censoring over the 20 years of follow-up increased (compare Figures 1a–c), from very small differences in power for the complete follow-up model (<1%) to larger differences (up to 7%) for the random follow-up model.
