Generalizability of Polygenic Risk Scores for Breast Cancer in the Multiethnic eMERGE Study

Background The majority of polygenic risk scores (PRS) for breast cancer have been developed and validated using cohorts of European ancestry (EA). Less is known about the generalizability of these PRS in other ancestral groups. Methods The Electronic Medical Records and Genomics (eMERGE) network cohort dataset was used to evaluate the performance of seven previously developed PRS (three EA-based PRSs, and four non-EA based PRSs) in three major ancestral groups. Each PRS was separately evaluated in EA (cases: 3939; controls: 28840), African ancestry (AA) (cases: 121; controls: 1173) and self-reported LatinX ancestry (LA) (cases: 92; controls: 1363) women. We assessed the association between breast cancer risk and each PRS, adjusting forage, study site, breast cancer family history, and first three ancestry informative principal components. Results EA-based PRSs were significantly associated with breast cancer risk in EA women per one SD increase (odds ratio [OR] = 1.45, 95% confidence interval [CI] = 1.40–1.51), and LA women (OR = 1.41, 95% CI = 1.13–1.77), but not AA women (OR = 1.13, 95% CI = 0.92–1.40). There was no statistically significant association for the non-EA PRSs in all ancestry groups, LA including an LA-based PRS and an AA-based PRS. Conclusion We evaluated EA-derived PRS for estimating breast cancer risk using the eMERGE dataset and found they generalized well in LA women but not in AA women. For non-EA based PRSs, we did not replicate previously reported associations for the respective ancestries in the eMERGE cohort. Our results highlight the need to improve representation of diverse population groups, particularly AA women, in research cohorts.