AI-derived Mammographic Risk Score improves breast cancer risk prediction, increasing incidence estimates by 53.2% per quintile within Tyrer-Cuzick model, aiding especially younger and Black women.
Does an AI-derived mammographic risk score improve breast cancer risk prediction compared to established risk factor models in women?
An AI-derived mammographic risk score captures biologically relevant imaging features and improves breast cancer risk prediction beyond established clinical models, particularly for younger and Black women.
Abstract Backgrounds Mammographic risk score (MRS), an artificial intelligence (AI)-derived metric from digital mammograms, captures breast tissue characteristics that may reflect cumulative exposures influencing breast cancer development. However, its relationship with well-established risk factors remains unclear. Methods This study included 9,623 cancer-free women from the Joanne Knight Breast Health Cohort who completed a baseline risk questionnaire and had MRS from screening mammograms. From 2008 to 2024, 468 breast cancer were diagnosed. We used linear regression to estimate β coefficients and 95% confidence intervals (CIs) for associations of MRS with lifestyle, reproductive, and previously validated risk factor models. We cross-classified MRS risk quintiles with risk factor models quintiles to compare differences in estimated incidence. Results MRS was positively associated with the Rosner-Colditz (β = 0.199, 95% CI: 0.138 to 0.260) and Tyrer-Cuzick scores (β = 0.072, 95% CI: 0.055 to 0.089). Being postmenopausal before age 50 (β age50 post-menopause VS age50 pre-menopause = -0.526, 95% CI: -0.646 to -0.406) were associated with lower MRS, whereas a history of breast biopsy (β yes VS no = 0.373, 95% CI: 0.303 to 0.443), postmenopausal BMI (β = 0.005 per 1kg/m2, 95% CI: 0.002 to 0.009), and breast density (β extremely dense∼55% VS almost all fat∼ 5% =0.374, 95% CI: 0.272 to 0.476) were associated with higher MRS (P 0.001 to 0.022). MRS improved risk factors models: breast cancer incidence increased by 53.2% per quintile of MRS within the Tyrer-Cuzick model and 44.9% within the Rosner-Colditz model (both P 0.0001). Women reclassified as high risk by MRS were more often younger and Black women. Conclusions MRS captures biologically relevant imaging features shaped by reproductive and metabolic exposures and improves breast cancer risk prediction, particularly for younger and Black women. Incorporating MRS into clinical tools could enable earlier identification of high-risk women, guide risk-reduction strategies, and tailor screening guidelines. Citation Format: T. Lan. Breast Cancer Risk Factors and Artificial Intelligence-derived Mammographic Risk Score in Black and White 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 PS1-13-08.
Tuo Lan (2026) studied this question. AI-derived Mammographic Risk Score improves breast cancer risk prediction, increasing incidence estimates by 53.2% per quintile within Tyrer-Cuzick model, aiding especially younger and Black women.