Clinical history shapes the predictive value of polygenic risk

For most diseases, whether PRS improves prediction beyond clinical data remains unknown, as existing evidence is concentrated in a handful of conditions with established risk models. Using 900,000 participants from UK Biobank and FinnGen, we developed and validated models for predicting 150 diseases and all-cause mortality. Genetic information provided the largest predictive gains in metabolic, cardiovascular, autoimmune, and neurological disease, but added little for genitourinary and respiratory conditions. All significant PRS-clinical risk interactions were negative, with predictive power gains from genetic data larger at lower clinical risk, suggesting that PRS may reveal susceptibility not yet apparent clinically. For example, in atherosclerotic cardiovascular disease, PRS expanded the clinically identified high-risk subgroup by 40%, and this reclassified subgroup had 4.9-fold higher incidence during the follow-up than those remaining low-risk. Our interpretable clinical-genetic models generalized across three biobanks and are publicly available through PrognosIQ.