Abstract PS7-07: Predicting breast cancer risk for women veterans

Abstract

Abstract BackgroundAccurate breast cancer risk assessment allows personalized approach in breast cancer screening and/or prevention. Breast cancer risk prediction models have been developed. Independent prospective validation however was needed before broad clinical applications. The Department of Veterans Affairs (VA) Million Veteran Program (MVP) is one of the largest biobanks in the world. Important attributes include: large sample size (>830,000, April 20), national coverage, multi-ethnic representation, access to bio-specimens from which multi-omics measurements have been conducted, thousands of biomedical phenotypes derived from our baseline and life style surveys and electronic health records (EHR). About 9% of MVP registrants are women. Breast cancer risk prediction models were evaluated in a prospective cohort of women in MVP. MethodologyMore than 350000 MVP participants completed genotyping to date. Breast cancer diagnoses were captured from the EHR. Ethnicity was determined by a supervised learning algorithm, Harmonized Ancestry and Race/Ethnicity (HARE), that used genetically inferred ancestry to refine self-identified race/ethnicity. Clinical breast cancer risk prediction instruments were based on personal health, lifestyle, demographics, family history, and environmental exposure. Polygenic Risk Score used in this study is comprised of 313 single nucleotide polymorphisms (PRS313). Clinical risk models tested were: BPC3, Lit and Breast Cancer Risk Assessment Tool (BCRAT, a version of the Gail’s model). The BPC3 and Lit models were tested by deploying the iCARE (Individualized Cancer Absolute Risk Estimator) with or without incorporation of PRS313. The BCRAT was deployed in the R package “BCRA” (Breast Cancer Risk Assessment v. 2.1 by F. Zhang). The performance of the risk prediction models was assessed by Area under the Receiver Operating Characteristic Curve (AUC-ROC). The observed absolute risk over expected absolute risk was compared for each decile. Results35133 genetic females were identified without a prior breast cancer diagnosis. By HARE definition, 10717 were African Americans (AA), 317 non-Hispanic Asians (ASIAN), 19941 non-Hispanic whites (EU), 1803 Hispanic (HIS) and 1355 unassigned. The median age of this cohort at MVP entry was 55. 369 subjects developed incident breast cancers with a median follow up of 4 years. These new cases were identified from Cancer Registry (292 cases) and EHR search for compatible ICD plus CPT codes (77 cases). The breast cancer incidence rate was 2.6/1000/year. iCARE-lit model was tested in 2731 AA women (38 incident breast cancers), 9027 EU women (111 incident breast cancers), and 10254 non-AA women (122 incident breast cancers) that had comprehensive risk factor evaluation (<4 missing). The AUC with the iCARE-lit model alone is 0.569 for AA, 0.563 for EU, and 0.565 for non-AA. Incorporation of PRS313 into the iCARE-lit model improved the AUC to 0.573 for AA, 0.645 for EU and 0.638 for non-AA. Additionally, iCARE-lit plus PRS313 estimated that 15% (AA), 7.3% (EU) and 7.7% (non-AA) of the women would have a lifetime risk of >20%. The iCARE-BPC3 had similar performance as iCARE-lit. The AUC of BCRAT alone was 0.648 for EU (20550 with 179 incident breast cancers) and 0.661 for AA (10462 with 96 incident breast cancers). Incorporation of PRS313 to BCRAT improved the AUC to 0.682 (EU) and 0.658 (AA). BCRAT plus PRS313 estimated that 5.6% (AA) and 4.4% (EU) of the women would have a lifetime risk of >20%. DiscussionWe prospectively evaluated and validated the performance of breast cancer risk prediction models as reported from case control studies. Longer follow up of more women in MVP will provide better assessment esp. for AA women. Prospective validation of these risk prediction models provides strong rationale for further development of these models in the clinical setting. Citation Format: Shiuh-Wen Luoh, Cynthia Brandt, Sally Haskell, Lina Gao, Byung Park, Paul Spellman, Saiju Pyarajan, Million Veteran Program, Nallakkandi Rajeevan, Jessica Minnier. Predicting breast cancer risk for women veterans [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PS7-07.