A Causal Multi-modal AI Model Stratifies Residual Risk and Identifies Candidates for Treatment Escalation in Node-Positive HR+/HER2- Early Breast Cancer

PURPOSE Prognostic biomarkers that estimate residual recurrence risk after standard adjuvant therapy to guide further treatment escalation are an unmet clinical need. This study used Ataraxis Breast CTX, a causal multi-modal AI model integrating clinical variables and histopathology, to stratify residual risk in clinically defined high-risk patients. We aimed to identify patients who have excellent outcomes on standard-of-care chemoendocrine therapy and patients who may benefit from additional therapies such as everolimus or CDK4/6 inhibitors. METHODS CTX combines clinical variables with image features from H&E images to produce estimates of recurrence risk (CTX-prognostic) and chemotherapy benefit (CTX-benefit). Prospective-retrospective validation was performed in the phase III UNIRAD trial to assess CTX-prognostic's performance in the subset of patients who had not received neoadjuvant chemotherapy, who had received adjuvant chemotherapy, and who had H&E slides available (n = 556). The primary endpoint in this study was disease-free survival (DFS). As an exploratory analysis, we evaluated CTX-benefit's ability to predict benefit from adjuvant everolimus. RESULTS CTX-prognostic effectively stratified patients into high- and low-risk groups. Across all patients, 5-year DFS was significantly higher in the low-risk group (93%) than in the high-risk group (80%). Additionally, treatment benefit from everolimus differed significantly by CTX-benefit score in an exploratory multivariable analysis adjusting for age, tumor size, nodal status, grade, and menopausal status (interaction p = 0.01). CONCLUSION CTX identified patients with good outcomes under chemoendocrine therapy alone and patients with residual risk who may benefit from treatment escalation. Additionally, CTX was found to be predictive of everolimus benefit. These results suggest that CTX may effectively stratify clinically high-risk HR+/HER2- patients by residual risk and may inform selection of adjuvant escalation strategies.