Results from an external validation analysis in the SWOG S8814 population, presented at the 2026 ASCO Annual Meeting (Abstract 107), suggest that a multimodal AI (MMAI) model may help guide adjuvant chemotherapy decisions in hormone receptor–positive, node-positive breast cancer by identifying patients most likely to benefit from treatment and those who may be candidates for de-escalation, such as subgroups with one to three positive nodes.

“MMAI is fast, scalable, and tissue-sparing,” commented presenting author Corey W. Speers, MD, PhD, the Merle M. Salter Endowed Chair of the Department of Radiation Oncology at the University of Alabama at Birmingham Heersink School of Medicine. “Because it used retained H&E [hematoxylin and eosin] pathology images plus clinical variables, it has the potential to provide personalized treatment information without consuming tissue and without the time, cost, or access barriers associated with some genomic assays.”

He continued, “The broader implication is that MMAI may help us move from broad clinicopathologic risk categorization toward a more individualized estimate of treatment benefit, especially in patients where chemotherapy decisions remain complex.”

Background and Model Methods

This U.S. Food and Drug Administration–cleared MMAI was developed using data from six phase III randomized trials and trained on more than 12,000 patients with early-stage invasive breast cancer across a broad range of clinical characteristics. The model integrates histopathology and clinical data, beginning with an H&E whole-slide image that is divided into patches and processed through an AI-derived tumor classifier and a pretrained, self-supervised model. An AI-derived image feature vector is then generated by the model and is fused together with clinical inputs of age, tumor size, and nodal status to generate a final prognostic risk score.

The MMAI had previously been validated for 10-year distant metastasis–free survival in the ABCSG-8 trial and for predicting chemotherapy benefit in node-negative disease in NSABP B-20.

The current study evaluated whether the same model could be generalized to a higher-risk population of postmenopausal patients with hormone receptor–positive, node-positive breast cancer.

Study Methods

The locked model was applied using the same previously validated risk-group cut points, which were developed primarily in node-negative patients. For clinical utility, intermediate- and high-risk categories were combined into a single non–low-risk group to enable a binary assessment of chemotherapy benefit.

Dr. Speers noted that “SWOG S8814 provides an ideal data set to evaluate this question.” The phase III trial enrolled nearly 1,500 postmenopausal patients with hormone receptor–positive, node-positive breast cancer and randomized them to tamoxifen alone, chemotherapy followed by tamoxifen, or concurrent chemotherapy and tamoxifen. The primary comparison for the current analysis was chemotherapy followed by tamoxifen vs tamoxifen alone.

The digital pathology–evaluable cohort included 413 patients. Overall, 32% were classified as MMAI low risk and 68% as non–low risk. Disease-free survival, the primary endpoint of SWOG S8814, and overall survival were evaluated using univariable and multivariable Cox proportional hazards models, while differential chemotherapy benefit was assessed by comparing relative risk reductions across MMAI-defined risk groups.

Key Findings

Patients with MMAI low-risk disease had higher 10-year disease-free and overall survival rates than those classified as non–low risk (74.1% vs 50.2% and 86.9% vs 63.2%, respectively; both log-rank P < .001). After adjustment for age, tumor size, and nodal burden, the MMAI risk score appeared significantly associated with disease-free (hazard ratio [HR] per 10-unit score increase = 1.43) and overall survival (HR per 10-unit score increase = 1.61; both P < .001), and non–low-risk patients had significantly poorer disease-free (HR = 2.15; P = .002) and overall survival (HR = 2.15; P = .01).

The 10-year disease-free survival appeared similar between the chemotherapy followed by tamoxifen and tamoxifen-alone arms (74.5% vs 73.9%; log-rank P = .69) among low-risk patients. In non–low-risk patients, 10-year disease-free survival was higher with chemotherapy followed by tamoxifen (55.4% vs 42.3%; log-rank P = .04), corresponding to a 22.6% relative reduction in the risk of disease recurrence.

Among patients with one to three positive nodes (n = 253)—the most clinically relevant subgroup, per Dr. Speers—the low-risk cohort showed no significant improvement in 10-year disease-free survival in the chemotherapy followed by tamoxifen arm (78.9% vs 77.4%; log-rank P = .97), corresponding to a 1.8% relative reduction in the risk of disease recurrence, whereas those with non–low-risk disease showed a 26.3% relative risk reduction (10-year disease-free survival: 67.5% vs 55.9%; log-rank P = .14).

“The subgroup analysis is limited by a low number of events…, but the direction and the magnitude of effect are consistent with broader hypotheses that even among patients with limited nodal burden, MMAI may distinguish those with minimal chemotherapy benefit from those with more substantial benefit,” Dr. Speers remarked. “Clinically, this is where the model could be most useful: not simply telling us the node-positive patients that are at high risk but helping refine which node-positive patients are most likely to benefit from chemotherapy.”

DISCLOSURES: SWOG S8814 was funded by the National Cancer Institute, with additional support from The Hope Foundation, a Biobank grant, and Artera. For full disclosures of the study authors, visit coi.asco.org.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.