Abstract PS3-01-08: Impact of allostatic load on breast cancer risk prediction among racially / ethnically diverse women

Abstract Background Current breast cancer risk assessment tools tend to underestimate risk of invasive breast cancer (IBC) among Black or Hispanic women and include clinical risk factors, such as family history of breast cancer, that are not readily available or under-reported in the electronic health record (EHR). Allostatic load (AL)—a measure of cumulative physiological stress which is comprised of standard laboratory results and vitals—has been associated with IBC incidence and mortality. Our objective was to determine whether components of AL derived from the EHR can improve IBC risk prediction. Methods We conducted a retrospective analysis of women undergoing screening mammography at Columbia University Irving Medical Center from 11/1/14 to 4/7/25. We extracted EHR data on the earliest index screening mammogram, age at index scan, race, ethnicity, body mass index (BMI), mammographic density (MD, 4-category BIRADS classification), breast biopsy results, and family history of breast cancer. Patients with a diagnosis of IBC or ductal carcinoma in situ prior to the index scan were excluded. The following laboratory data and vitals were collected for AL biometrics representing major regulatory systems: metabolic (albumin, alkaline phosphatase, glucose), renal (blood urea nitrogen, creatinine), immune (white blood cell count), and cardiovascular (pulse, systolic and diastolic blood pressure). Patients with at least five of these components available in the two years preceding the index scan were included, and multiple imputation by chained equations was used to impute missing AL data. For modeling, AL-derived data were used as individual values and as a composite score (0-9), with one point assigned for each biometric in the highest quartile except albumin which was the lowest quartile. The primary outcome measure was development of IBC, defined as histologically-confirmed IBC at least six months after the index scan. Multiple logistic regression models were constructed, and area under the receiver operating characteristic curve (AUC) values were compared using DeLong’s test to a baseline model constructed from established IBC risk factors readily extracted from the EHR. Results Among 27,155 evaluable patients, mean age was 59 years (SD 12) and 40% were Hispanic, 29% non-Hispanic White, and 14% non-Hispanic Black. A total of 135 (0.5%) patients had a subsequent IBC diagnosis. The baseline IBC risk prediction model of age, race / ethnicity, BMI, and MD achieved an AUC of 0.642; adding AL composite score or including breast biopsy results or family history did not significantly improve risk prediction. However, adding all nine individual AL-derived variables improved the AUC to 0.663 (p=0.011). Among non-Hispanic White women, the baseline model with AUC of 0.643 improved with the addition of breast biopsy results and family history (AUC of 0.656, p0.001) and the AL-derived variables (AUC of 0.691, p=0.037). Among non-Hispanic Black women, the baseline model had an AUC of 0.634, which improved with the addition of breast biopsy results and family history to an AUC of 0.649 (p0.001). Among Hispanic women, including AL-derived variables significantly improved the baseline model with an AUC of 0.611 to an AUC of 0.714 (p=0.002). Including the composite AL score did not improve prediction for any racial / ethnic group. Discussion We demonstrated that addition of AL-derived laboratory data and vitals extracted from the EHR significantly improved IBC risk prediction, particularly among Hispanic women, achieving AUCs comparable to existing well-validated breast cancer risk assessment tools. With further validation, incorporation of AL variables may facilitate efficient IBC risk prediction in the clinical setting to inform breast cancer screening and risk-reducing strategies among diverse women. Citation Format: J. B. Gibbons, V. Ro, J. E. McGuinness, R. Carlos, K. D. Crew. Impact of allostatic load on breast cancer risk prediction among racially / ethnically diverse women abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-01-08.