For each disease, the GPS with the best discriminative capacity in the testing dataset was calculated in the testing dataset of 288,278 participants using genotyped and imputed variants using the Hail software package.36 The proportion of the population and of diseased individuals with a given magnitude of increased risk was determined by comparing progressively more extreme tails of the distribution to the remainder of the population in a logistic regression model predicting disease status and adjusted for age, gender, four principal components of ancestry, and genotyping array. Individuals were next binned into 100 groupings according to percentile of the GPS and unadjusted prevalence of disease within each bin determined. We next compared the observed risk gradient across percentile bins to that which would be predicted by the GPS. For each individual, the predicted probability of disease was calculated using a logistic regression model with only the genome-wide polygenic score (GPS) as a predictor. The predicted prevalence of disease within each percentile bin of the GPS distribution was calculated as the average predicted probability of all individuals within that bin. The
