Abstract
Risk-based approaches offer promise for enhancing early detection of prostate cancer. Polygenic risk scores (PGSs) have emerged as a potential approach for risk stratification, though their performance varies by population. We evaluated nine PGSs (four existing, five new) for predicting 5-year prostate cancer risk across three international population-based prospective cohorts: UK Biobank (UKB), the Australian QSkin Sun and Health Study (QSkin), and Melbourne Collaborative Cohort Study (MCCS). We analyzed UKB European-ancestry (n = 184,010), South-Asian-ancestry (n = 5,097), and African-ancestry (n = 3,193), QSkin European-ancestry (n = 6,791), and MCCS European-ancestry (n = 1,809) male participants. We estimated age-specific 5-year prostate cancer risks (from population data) and PGS-adjusted risks (age-specific risks multiplied by PGS-based relative risks). Predictive performance was assessed using discrimination (AUC) and calibration. PGS significantly enhanced 5-year risk prediction over age alone, particularly for European ancestry (AUC increase 0.05-0.12, p < 10(-6)). PGS performance was consistent across European-ancestry men in Australian and UK cohorts, and by pre-baseline prostate-specific antigen tests and family history in UKB. No single PGS outperformed others across all cohorts and ancestry groups. As an illustrative example for potential risk stratification, for a leading PGS in both Australian cohorts, we estimated the population-average 5-year risk at age 50 was reached 5 years earlier by individuals with 20% highest PGS451 and 5 years later by those with 20% lowest PGS451. In conclusion, rigorous analyses with consistent results from international cohorts support the potential of PGS to improve 5-year prostate cancer risk prediction. In the future, PGS may be improved further to enhance performance in diverse populations.