Integrating polygenic risk scores into clinical breast cancer models: Influence on prediction in diverse cohorts.

Abstract

10506 Background: Breast cancer (BC) is the second leading cause of cancer death in women worldwide. Periodic mammography screening has been shown to reduce breast cancer mortality by around 20% in average-risk women and several BC risk models are currently used to identify women at higher risk who can be targeted with increased or earlier screening. However, despite their broad use, these models display only moderate discrimination performance (AUC ranging between 0.51 to 0.68). Here we explore the potential of integrating PRS into a BC risk model using a testing dataset comprising over 175,000 women of diverse ancestry in the UK Biobank and Women Health’s Initiative Cohorts. Methods: We built and validated novel ancestry-specific BC PRSs utilizing novel methodology that leverage multiple Genome Wide Association Studies (GWASs) and linkage disequilibrium maps. We assessed genetic ancestry from 5 continental level ancestry groups across all individuals and applied the PRSs to our testing dataset to assess the clinical implications of integrating genetic information to the Tyrer-Cuzick (TC) breast cancer clinical risk model. Results: The PRSs were well calibrated and displayed high risk stratification (Odds Ratios per Standard Deviation: 1.41 (1.28-1.62) to 1.76 (1.40-2.20; Table) which were comparable or better than benchmarking comparisons with previously published PRS panels. Across genetic ancestries, integrating PRS led to a relative increase of women identified at high lifetime BC risk (20% or higher) from between 1.7 fold (East-Asian ancestry) to around 3 fold (European ancestry). The Net Reclassification Improvement compared to the TC-only model was around 12% for East-Asian (0.001-0.254), European (0.111-0.133), and Admixed American ancestries (0.002-0.228), 17% (0.079-0.266) for South-Asians, and 5% (0.008-0.094) for African ancestry individuals (Table). Conclusions: Our results demonstrate that optimizing PRSs for genetic ancestries and integrating them into BC risk models can lead to a significant improvement in risk stratification. This can enable more targeted use of enhanced screening and prevention strategies improving their cost/benefit ratio.[Table: see text]