Fig. 6

Towards the clinical implementation of PRS-based disease prediction models. Comparison between PRS risk profiles and functional variant carriers in UKB Europeans. (A) Cumulative incidence of coronary artery disease among carriers with pathogenic or likely-pathogenic mutations in familial hypercholesterolemia genes vs. individuals in the top 22% of our PRS distribution from the ensemble method, where a percentile was chosen to match the risk up to age 70 with that of mutation carriers. The median 25%-75% of the PRS distribution serves as the population reference. (B) Percentage of coronary artery disease (CAD) cases diagnosed in individuals aged under 50, under 65, and across all age groups, occurring either in mutation carriers or in those at the top of the PRS distribution. Plots display the ratio of high risk PRS cases to mutation carrier cases in each age category. (C) Forest plot displays odds ratios for the observed risk in coronary artery disease. Horizontal lines represent 95% confidence intervals. Individuals were categorized into three groups based on their polygenic scores – low, intermediate, and high, corresponding to the lowest 25%, middle 25–75%, and highest 25% of the distribution, respectively. The intermediate PRS group served as the reference category for calculating odds ratios. (D) Diagnostic test effectiveness of PRS-based disease prediction models. Odds ratios are reported per ancestry group, defined as the ratio of the odds that an individual with a clinical condition will score above a specified risk threshold, to the odds of scoring below this threshold without the condition. Error bars represent 95% confidence intervals. (E) ‘High risk threshold’ illustrates the percentile cutoff for a specific condition, above which results are considered high-risk. Odds ratios were optimized by looking for the largest high risk group for which the DOR exceeded the pre-set threshold of 10 using cutoffs at the top 1%, 2%, 3%, 5%, 10%, and 20% percentiles.