External validation of a MMAI model for prognosis and chemotherapy benefit prediction in postmenopausal, node-positive, hormone receptor–positive breast cancer patients: Analysis of SWOG S8814.

Abstract

107 Background: Despite effective endocrine therapy, the absolute benefit of adjuvant chemotherapy varies widely in node-positive HR+ disease, particularly among patients with 1-3 positive nodes. Prognostic tools for these patients often rely on genomic assays, which are costly, timely, or inaccessible in many clinical settings. We present a multimodal artificial intelligence (MMAI) model that integrates clinical and histopathological data to quickly stratify risk of distant metastasis and inform therapeutic decisions. Developed in six phase III randomized clinical and validated for chemotherapy benefit in N0 patients, MMAI offers an accessible alternative to genomic tools. Here, we validated MMAI for prognosis and prediction of chemotherapy benefit in SWOG S8814 – a randomized phase III trial of tamoxifen ± chemotherapy (CT) in postmenopausal women with node-positive (N+) HR+ breast cancer. Methods: Patients with digitized baseline H&E- stained diagnostic slides and clinical data (age, tumor size, nodal status) were analyzed (N = 413). The locked MMAI generated a continuous risk score and categorical risk groups (low, high). Associations with disease-free survival (DFS; 168 events) and overall survival (OS; 125 events) were assessed using univariable and multivariable Cox Proportional Hazard models. Hazard ratios (HRs) and 95% confidence intervals (CI) were estimated. Differential CT benefit was evaluated by estimating relative risk reduction by MMAI risk groups. Results: MMAI was prognostic for DFS (HR per SD 1.73, 95% CI 1.48–2.03; p < 0.001) and OS (HR per SD 1.93, 95% CI 1.60–2.32; p < 0.001), remaining significant after adjustment for age, tumor size, and nodal burden. In the subset of patients with 1–3 positive nodes (n = 253), MMAI identified differential CT benefit: high-risk patients (56% of the patients) demonstrated a 26.3% relative reduction in 10-year DFS risk with CAF-T+TAM versus TAM alone, while low-risk patients (44% of the patients) derived minimal benefit (1.8% relative reduction in 10-year DFS risk). Additionally, the addition of CT resulted in DFS HRs of 1.21 (95% CI: 0.51-2.83) and 0.85 (95% CI: 0.55-1.23) in low- and high-risk patients, respectively. Conclusions: In SWOG S8814, a locked MMAI model using routinely available pathology and clinical data independently stratified prognosis and identified node-positive HR+ patients most likely to benefit from adjuvant CT, with minimal benefit among MMAI low-risk patients with 1–3 nodes. These findings support the use of MMAI as a fast (hours instead of weeks), scalable, cost-effective, and non-tissue consumptive alternative to genomic testing to inform adjuvant decisions in HR+ N+ EBC patients.