Independent validation of polygenic risk scores for overall and triple-negative breast cancer among high-risk African American women

Background : Polygenic risk scores (PRSs) are emerging tools for stratifying breast cancer (BC) risk. We developed PRSs of overall BC and triple-negative BC (TNBC) risk in women of African ancestry, using case-control data from studies unselected for family history. PRS could be part of comprehensive breast cancer risk assessment for women at higher risk due to cancer family history. Here we aimed to evaluate the performance of these PRSs in women with elevated cancer risk. Methods : We included samples from African American women who underwent multigene hereditary genetic testing between 2016 and 2024 and who tested negative for pathogenic or likely pathogenic variants in known BC genes. Genotyping was performed using custom Axiom PMRA arrays (ThermoFisher) and imputation based on TOPMed reference data. We evaluated two PRS models for overall BC with 2,324,063 variants (“Overall BC Model 1”) and 175,173 variants (“Overall BC Model 2”), and three PRS models for TNBC with 554,959 variants (“TNBC Model 1”), 37,226 variants (“TNBC Model 2”), and 162 variants (“PRS-162”). The 313-variant PRS developed in women of European ancestry (PRS-313) served as a benchmark. Only variants with imputation R 2 ≥ 0.7 were included. Associations between PRSs and BC status were examined using multivariable logistic regression, adjusting for age at diagnosis or testing, the top ten principal components, and weighted family history of BC. Adjusted odds ratio (OR) per standard deviation and area under the curve (AUC) were calculated. Results : The study included 19,455 unaffected and 12,067 women with BC (2,311 TNBC). Mean age at BC diagnosis was 55.2 years in BC cases vs. 42.1 years at testing in unaffected women. More unaffected women had a family history of BC in first-degree (32.8% vs. 28.0%) and second-degree relatives (55.7% vs. 35.0%) than affected women. Over 95% of African PRS variants and 82% of the PRS-313 variants had an imputation R 2 ≥ 0.7. For overall BC, PRS-313 showed an AUC of 0.567 (95% confidence interval [CI]: 0.560, 0.574) and an OR of 1.28 (95% CI: 1.25, 1.31). In comparison, the African overall BC Models 1 and 2 performed better, with AUCs of 0.588 (95% CI: 0.580, 0.595) and 0.584 (95% CI: 0.576, 0.591) and ORs of 1.39 (95% CI: 1.35, 1.43) and 1.37 (95% CI: 1.34, 1.41), respectively. For TNBC, Models 1 and 2 yielded similar performance, with AUCs of 0.579 and 0.574, respectively. Notably, the TNBC PRS-162 showed very good performance with an AUC of 0.609 (95% CI: 0.596, 0.622) and an OR of 1.47 (95% CI: 1.40, 1.55). Conclusion : The PRSs demonstrated good performance among women with a strong family history of BC, reflecting real-world populations where early PRS testing is most relevant for those at elevated risk. The high predictive accuracy of the 162-variant TNBC PRS supports its potential as a cost-effective risk assessment tool to promote equitable care.