We repeated our simulations using a sample size of 612 cases and 3366 controls, generating survival times from the exponential model using the simulated prognosis trait as the log hazard, and testing association using Cox regression (Supplementary Tables 13 and 14). Adjusted and unadjusted analyses had similar overall properties, suggesting that our approach could be applicable in this setting. Together our results suggest that the paradoxical association of MUC5B with increased survival could indeed be due to index event bias, and that the risk allele of MUC5B is in fact associated with decreased survival.
