We used GWAS summary statistics that did not include the UKB for five stroke outcomes and 14 stroke-related phenotypes (Supplementary Table 1) to generate 19 GRSs associated with IS (Fig. 1). As expected, the 19 individual GRSs were correlated with each other in several distinct clusters: (i) any stroke (AS), IS, cardioembolic stroke (CES), large artery stroke (LAS), and small vessel stroke (SVS); (ii) the three CAD scores (1KGCAD, 46K, and FDR202); (iii) total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL), and high-density lipoprotein cholesterol (HDL); (iv) systolic BP (SBP) and diastolic BP (DBP); and (v) body mass index (BMI) and type 2 diabetes (T2D) (Fig. 2). From the 19 distinct GRSs, we constructed the metaGRS using elastic-net logistic regression with 10-fold cross-validation on the derivation set (Fig. 1; metaGRS; model weights are shown in Supplementary Fig. 1), and subsequently converted the model to a set of 3.2 million SNP weights, which are freely available (10.6084/m9.figshare.8202233).
