Abstract RF3-03: Evaluation of a digital pathology based multimodal artificial intelligence model for prognosis and prediction of chemotherapy benefit in node-negative, hormone receptor-positive breast cancer patients: analysis of the NSABP B-20 trial

Abstract Background: Accurate risk stratification for distantmetastasis (DM) in hormone receptor-positive (HR+) early-stage breast cancer(EBC) is important in guiding decisions on selection of adjuvant therapies,particularly use of chemotherapy (CT). Genomic assays have improveddecision-making in these patients; however, global accessibility can be limiteddue to cost and logistical constraints. The Multimodal Artificial Intelligence(MMAI) breast cancer model validated for prognosis in ABCSG-08 integrateshistopathological images and clinical data to stratify risk of DM as apotential lower-cost alternative. Here, we evaluated the MMAI algorithm in thelandmark NSABP B-20 Trial - a randomized phase III trial of tamoxifen (TAM)alone, TAM plus methotrexate and fluorouracil (MFT) or TAM pluscyclophosphamide, methotrexate, and fluorouracil (CMFT) in women with nodenegative (N0) HR+ EBC. Methods: This study included patients from B-20 withavailable digitized H.001) and MMAI risk group (high vs low sHR 95% CI =3.97 2.57-6.16, p.001; intermediate vs low 2.78 1.47-5.23, p=.002). Inthe entire cohort, no statistically significant interaction effect betweentreatment arms and either continuous or dichotomized MMAI was found. However, the interaction was significant (p=.01) in patients aged ≥ 50: In MMAI high/intermediate-risk patients (32%), addition of CT was associated with a 52% relative 10-year DM riskreduction (10% in CMFT vs 21% in TAM), compared to (10-year DM rate: 7% in CMFTvs. 5% in TAM) in low-risk patients. The predictive interaction of dichotomizedMMAI was not significant in patients aged 50; chemotherapy added benefit inboth low and high/intermediate MMAI groups. Conclusion: The MMAI demonstrated strong independentprognostic performance in all patients with HR+ N0 EBC from the NSABP B-20 trialbut was not predictive of chemotherapy benefit in the entire population.However, in an exploratory analysis by age, in patients aged ≥ 50 the MMAI risk groups (high/intermediate vs. low) werepredictive for benefit of CMF. These findings supportthe potential use of MMAI as a lower cost, non-tissue consumptive alternativeto genomic testing for guiding CT decisions in older HR+ N0 EBC patients. Keywords: Artificial intelligence, breast cancer,HR-positive, early-stage, distant metastasis, multi-modal model, prognosticbiomarker, chemotherapy predictive Support: U10 CA180868, -180822; U24 CA196067 Citation Format: C. E. Geyer, M. Filipits, N. Harbeck, N. Harbeck, J. Zhang6, P. Rastogi, A. Piehler, T. Freeman, M. Balic, W. Zwerink, H. Kreipe, D. Hlauschek, S. Anderson, J. R. Griffin, D. Kates-Harbeck, M. Gnant, N. Wolmark. Evaluation of a digital pathology based multimodal artificial intelligence model for prognosis and prediction of chemotherapy benefit in node-negative, hormone receptor-positive breast cancer patients: analysis of the NSABP B-20 trial 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 RF3-03.