We visually examined the PRS distribution in each ancestral group of the eMERGE dataset before and after the adjustment, and evaluated the overall calibration of the adjusted PRS in the prediction model across eMERGE samples by comparing the predicted risk (i.e., the average predicted probability of being a case) and the observed risk (i.e., the proportion of cases) in each PRS decile. Lastly, to assess the impact of the adjustment on the tail discrimination of the trans-ancestry PRS, we compared the OR of individuals in the top percentiles of the adjusted PRS across ancestries by applying a single cutoff with the OR of individuals in the top percentiles of the raw PRS within each ancestry in the multi-ethnic eMERGE dataset.
