We calculated a range of metrics to assess the predictive performance of the PRS. To measure the overall prediction accuracy, we calculated (i) the proportion of variation in the T2D case-control status explained by the PRS on the liability scale [38], after accounting for a basic set of covariates including age, sex, top 10 PCs of the genetic data, and study site (in the eMERGE analysis only); (ii) the area under the receiver operating characteristic (ROC) curve (AUC) for the covariates-only model (age, sex, top 10 PCs and study site), the PRS-only model, the PRS adjusting for the covariates, and the PRS combined with covariates; (iii) the odds ratio (OR) per standard deviation (OR/SD) change in the PRS, adjusting for the basic covariates. To quantify the discrimination capability at the extreme tail of the PRS, we identified individuals at the top 2%, 5%, or 10% of the PRS distribution, and calculated OR of these high-risk individuals versus the rest of the samples, adjusting for the covariates. We further calculated the sensitivity, specificity, positive predictive value (PPV; the proportion of identified
