Abstract 7398: Defining prostate cancer polygenic risk score categories: Reference populations and methods used impact risk interpretation

Abstract Background: Polygenic risk scores (PRS) could effectively identify individuals at higher risk of prostate cancer (PCa) and may have notable implications for screening. However, PRS risk classification depends on the comparison population. Objective: We investigated how different reference populations and methods for defining PRS risk categories affect interpretation. Methods: We evaluated a multi-ancestry PCa PRS in 6, 194 cases and 24, 693 controls from self-identified non-Hispanic White (WH), African American (AA), Japanese American (JA), and Latino American (LA) populations in the Multiethnic Cohort (MEC). The PRS was based on 451 variants identified in a prior PCa GWAS across four ancestries. MEC participants were categorized into PRS deciles using 1) PRS distribution of MEC controls within each population or across all populations combined; 2) MEC controls’ PRS distribution adjusted for the first 10 principal components (PCs); 3) PRS distribution from external reference populations within each matched continental ancestry or across all ancestries; and 4) individualized PRS deciles calculated as a linear combination of ancestral proportions and external reference populations’ decile cutoffs. Reference populations for methods 3 and 4 included 1, 890 EUR, AFR, EAS, and AMR individuals from 1KGP and the PAGE Global Reference Panel, chosen for their close genetic resemblance to respective populations based on PCA. Additionally, 41, 825 EUR, 18, 636 AFR, 1, 263 EAS, and 9, 888 AMR individuals from All of Us (AoU), defined by genetic ancestry, were also used as reference population for methods 3. Logistic regression models adjusted for age and the first 10 PCs were used to evaluate the association between PRS deciles and PCa risk.Results: Non-population-specific reference group cutoffs approaches consistently had higher PRS predictive ability than population-specific cutoffs for the top 90%-100% PRS decile across all methods and populations. For instance, the OR for AA men in the top 90%-100% PRS decile was 4.37 [3.49-5.48] and 5.29 [3.85-7.27] when using the PRS distribution of MEC controls within each population and across all populations combined, respectively. Similarly, the OR was 4.60 [3.35-6.30] and 5.32 [4.18-6.79] when using 1KGP/PAGE, and 4.73 [3.74-6.00] and 5.41 [4.14-7.07] when using AoU as reference population. There was high variability in cases and controls selected in the top 90%-100% PRS decile across methods. For instance, the top decile included 9, 447/1, 264, 485/440, and 750/872 cases/controls when using MEC controls, 1KGP/PAGE, and AoU, respectively, to define non-population-specific PRS decile cutoffs for AA men. Conclusion: PRS interpretation and individual risk categorization are highly dependent on reference populations and methods used to define risk categories, which have critical clinical implications when implementing PRS. Citation Format: Boya Guo, Ali Sahimi, Xin Sheng, Shuyan Cheng, Fei Chen, Christopher A. Haiman, David V. Conti, Burcu Darst. Defining prostate cancer polygenic risk score categories: Reference populations and methods used impact risk interpretation [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7398.