We applied genomic control to the three GWAS files prior to meta-analysis. Inflation due to stratification was estimated using LD-score regression, which can differentiate inflation due to population stratification from that due to real signal. The intercept was used to correct the standard errors (SEs) of the estimated effect sizes as follows: SEGC=LDSCintercept∗SE2. The intercepts were b0= 1.005 (SE=0.007) for ICC, b0=1.004 (SE=0.007) for 23andMe, and b0=1.022 (SE=0.008) for UK-Biobank. We then performed a fixed effects meta-analysis based on effect sizes (log odds ratios [OR]) and standard errors in METAL53. We applied the conventional p-value threshold of 5e-08 as indication of genome-wide significance. The meta-analysis was performed on 11,733,371 SNPs that passed quality control. The combined sample size of the meta-analysis was 184,765 individuals, although the sample size varied per SNP due to differential missingness across samples.
