AI-based prognosis and chemotherapy benefit in node-positive, HR+ breast cancer: a blinded external validation on SWOG-8814 trial

Abstract

Background The benefit from adjuvant chemotherapy varies among patients with postmenopausal, node-positive, hormone receptor-positive breast cancer. Genomic assays identify patients likely to benefit from chemotherapy but are expensive and tissue consuming. AI-enabled biomarkers may provide scalable alternatives. We performed a blinded external evaluation of Ataraxis Breast CTX, a causal multi-modal AI model that integrates H&E images with clinical variables, in the phase III SWOG-8814 trial. Methods The analytic cohort comprised 677 patients (46% of the 1477 enrolled) with digitized H&E slides. Two CTX scores were evaluated: a histology-only score and a multi-modal score integrating pathology features with clinical variables. CTX weights and thresholds were locked before evaluation and SWOG-8814 data were not used for retraining. Predictions were generated blinded to outcomes and provided to SWOG for validation. The primary endpoint was disease-free survival (DFS), matching the primary endpoint of the trial. Findings The histology-only score was prognostic for DFS (HR = 1.36, 95% CI = 1.22-1.51) and added prognostic information beyond clinical variables alone (p < 0.001). The multi-modal score showed a stronger association with DFS (HR = 1.60, 95% CI = 1.44-1.78, p < 0.001) and greater discrimination than the histology-only score (C-index 0.66 vs 0.59). In exploratory analyses, modeled as a continuous score, CTX-benefit showed a treatment-by-biomarker interaction of HR of 0.82 (0.64-1.04, p = 0.053) for CET versus ET. The chemotherapy effect on DFS was numerically larger in the high-benefit subgroup (n = 480, HR = 0.79, 95% CI = 0.59-1.05) compared to the low-benefit subgroup (n = 197, HR = 0.91, 95% CI = 0.49-1.68). Interpretation In this blinded validation, CTX was significantly associated with DFS, supporting its prognostic value. In exploratory analyses, CTX demonstrated suggestive evidence of a treatment-by-biomarker interaction for chemotherapy benefit. These findings support further prospective validation of CTX as a tissue-sparing tool for recurrence-risk assessment and treatment selection. Funding National Cancer Institute, The Hope Foundation for Cancer Research, and Ataraxis AI.