Abstract PS2-08-28: Artificial Intelligence-Based Histopathology Model Predicts Recurrence Risk in Older Patients with Early HR+/HER2− Breast Cancer: Results from the Basel University Hospital Cohort

Abstract Background: Older patients with hormone receptor-positive (HR+)/HER2-negative (HER2−) early-stage breast cancer (BC) face frequent uncertainty in treatment decisions, as tools like Oncotype DX (ODX) are less validated in this age group. Ataraxis Breast RISK (ATX) is a novel artificial intelligence (AI) model that integrates digitized H50 years (n=41). Reclassification was assessed within the ODX-intermediate-risk group (n=167). Pooled results from five external cohorts (totaling 611 patients ≥70 years) were also referenced for external context. Results: Among USB patients ≥70 years, ATX achieved a C-index of 0.79 (95% CI: 0.51-1.00) and an HR of 2.12 (95% CI: 1.18-3.78), indicating strong predictive accuracy. In contrast, ODX performance was limited (C-index: 0.55; HR: 1.76, p=0.13). In patients 50 years, ATX also demonstrated robust prognostic value (HR: 4.36, p=0.03).Stratifying patients ≥70 by a 10% 5-year recurrence threshold, ATX separated a high-risk group (n=31) with substantially worse DFI from a low-risk group (n=9), supporting its potential clinical utility.Within the ODX-intermediate-risk group (n=167), ATX reclassified 128 patients (77%) to low risk and 39 (23%) to high risk. Although the separation in DFI curves was less pronounced (HR: 1.61, p=0.393; C-index: 0.69), ATX offered definitive risk categorization, addressing the ambiguity inherent to ODX intermediate scores. Across 5 external validation cohorts including 611 patients ≥70 years, ATX maintained strong performance with a pooled C-index of 0.74 (95% CI: 0.70-0.81) and HR of 1.82 (95% CI: 1.51-2.20), reinforcing its generalizability in older BC populations. Conclusions: ATX provides high prognostic accuracy in older HR+/HER2− BC patients and outperforms ODX in the USB cohort. It consistently stratifies patients across age groups and resolves the uncertainty associated with intermediate genomic risk scores. These findings highlight ATX's potential to support more individualized, evidence-based CHT decisions in elderly patients—overlooked population in BC risk modeling. Citation Format: E. D. Chiru, L. Sojak, J. Witowski, K. Zeng, C. Kurzeder, S. Muenst, M. Vetter. Artificial Intelligence-Based Histopathology Model Predicts Recurrence Risk in Older Patients with Early HR+/HER2− Breast Cancer: Results from the Basel University Hospital Cohort 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 PS2-08-28.