log-linearity and proportional hazards. The assumption of log-linearity is common in the additive logistic regression used in a typical GWAS. We validated the assumption of proportional hazards assumed by the Cox model by conducting the Schoenfeld test in the 22 prioritized SNPs. None of the SNPs has a Schoenfeld P value, which is the P value for Pearson product-moment correlation between the scaled Schoenfeld residuals and time, lower than 0.035 (multiple test correction threshold = 0.00227) in any of the 7 cohorts. Further, only 3 out of the 148 P values were less than 0.05, suggesting that the time proportionality assumption is unlikely to be violated in these associations (Supplementary Table 1). Similarly, the Schoenfeld test was conducted for all 22 SNP association models on the covariates in the ADGC and GERAD cohort (Supplementary Table 1). We also examined the effect sizes of our candidate SNPs in these cohorts and found consistent effect sizes (Supplementary Fig. 3) in the 3 retrospective case-control cohorts (ADGC, GERAD, EADI case-control) and 4 prospective cohorts (EADI-prospective, CHARGE FHS, CHS and Rotterdam).
