LAUSR

Introduction: Recent studies suggest a substantial improvement in risk prediction of coronary artery disease (CAD) with polygenic risk scores (PRS). The degree to which this improvement is generalizable to other European (EUR) and non-EUR populations remains unclear. Hypothesis: We hypothesized that genome-wide PRSs trained and validated in the UK Biobank (UKBB) would perform best in external EUR populations and reduce the performance gap in prediction between EURs and African Americans (AA). Methods: We tested our hypothesis in ~91 000 EUR, AA, and Hispanic (HISP) Million Veteran Program (MVP) participants with CAD and ~225 000 non-cases as of July 2017. We created 5 separate weighted PRSs in all participants using: 1) 164 genome-wide significant SNPs for CAD (164SNPs), 2) 46 000 Metabochip SNPs (GRS46K), 3) 1.5 million (M) SNPs from standard pruning & thresholding of CARDIOGRAM+C4D (Khera_P&T), 4) 6.6M SNPs PRS created with LDpred (LDpred_Khera), and 5) 1.7M SNP PRS combining weights of 3 PRSs through m...