The three CAD GRSs (46K, 1KGCAD, FDR202) were generated previously using an n = 3000 derivation subset of the UKB (included in the larger n = 11,995 subset employed here);21 briefly: (i) the 46K score was derived by LD thinning of the Metabochip summary statistics;46 (ii) the 1KGCAD was derived by LD thinning of the 1000Genomes CAD summary statistics;47 and (iii) the FDR202 score was from the 1000Genomes CAD summary statistics, consisting of SNPs with associations at false discovery rate < 0.05. The AF GRS was derived from a GWAS of AF48 using a pruning and thresholding approach. For the remaining GRSs, we used published summary statistics to generate a range of scores based on different r2 thresholds with PLINK49 LD thinning (–indep-pairwise), and selected one optimal model (in terms of the largest magnitude hazard ratio), resulting in one representative GRS for each set of summary statistics.
