One of the major goals of the PGS is to identify individuals at high or low risk. Therefore, we dichotomized PGSs as high or low risk using different thresholds. To determine the best thresholds, we defined high-risk groups as the top 5% to 50% (in increments of 5%) of samples with the highest PGS, and we defined low-risk groups as the bottom 5% to 45% (in increments of 5%) of samples with the lowest PGS. Then each high-risk or low-risk group was compared with the remaining samples; for example, the top 5% of samples were compared with the remaining 95%. The PGS distributions were determined in the screening dataset (COGA, SAGE, and OZALC combined), the AOU dataset, and the IB dataset separately. COGA and OZALC were familial cohorts, and some SAGE samples were related. Therefore, we fit generalized linear mixed models using generalized estimating equations; specifically, we included a random intercept to adjust for the family relationships. Unrelated samples from the AOU and IB datasets were included; thus, we used logistic regression. Sex, the first 10 principal components of genetic
