Our simulations showed reduced type-1 error rates for our procedure compared with an unadjusted analysis, except in the case of strong negative genetic correlation between incidence and prognosis or no non-genetic correlation. Power is slightly reduced overall, but may be considerably increased for some individual SNPs. Again our approach performed more poorly under strong negative genetic correlation, which is arguably less likely than positive correlation. We simulated genetic architectures that were typical of complex diseases32 while allowing a high degree of index event bias. In smaller studies, however, such as our IPF survival GWAS, our adjustment may have high variance resulting in more severely reduced power.
