AI Model Enhances Recurrence Risk Prediction in HR-Positive, HER2-Negative Breast Cancer
A groundbreaking AI model, developed by integrating clinical, molecular, and histopathological data, has significantly improved recurrence risk stratification in hormone receptor (HR)-positive, HER2-negative breast cancer, as presented at the San Antonio Breast Cancer Symposium (SABCS) in December 2025. This subtype of breast cancer is the most common, with at least 50% of recurrences occurring more than five years after diagnosis, according to Joseph A. Sparano, MD, chief of the Division of Hematology and Oncology at the Mount Sinai Tisch Cancer Center.
The Oncotype DX (ODX) 21-gene recurrence score, a widely used test, provides prognostic information for distant recurrence and predictive information for chemotherapy benefit. However, its ability to forecast recurrence after the five-year mark is limited. To address this, Sparano and his team aimed to develop a new diagnostic test that offers better prognostic estimation of recurrence risk, including late recurrence risk, by studying tumor specimens from the TAILORx trial.
The AI model was designed to evaluate both the images of digitized slides used for routine pathologic assessment and the molecular and clinical characteristics of breast cancer. This comprehensive approach allows for better prognostic information about cancer recurrence risk out to 15 years, encompassing early recurrence within five years after diagnosis and late recurrence after five years.
The research team utilized digitized tissue images and molecular RNA expression data from 4,462 tumor samples, along with corresponding clinical data from TAILORx study participants. These data were employed to train and validate several risk models. The prognostic performances of these models were compared to the ODX results used in the trial to guide chemotherapy use, with the concordance index (C-index) used for assessment. The C-index measures the ability of a diagnostic test to correctly rank recurrence risk.
The multimodal model, ICM+, integrating pathomic imaging, clinical, and expanded molecular models, demonstrated superior performance compared to ODX for overall distant recurrence at 15 years (C-index 0.705 vs. 0.617) and late recurrence after 5 years (C-index 0.656 vs. 0.518) in the training/5-fold cross-validation set. This model also showed similar superior prognostic performance in a holdout validation set, outperforming ODX for overall recurrence and late distant recurrence.
The findings from this study will lead to the development of a new diagnostic test that more accurately estimates recurrence risk in women with HR-positive, HER2-negative, node-negative breast cancer, which accounts for about half of all breast cancers in the United States. Sparano emphasized the potential of AI in creating better diagnostic tests that may more accurately estimate recurrence risk and personalize treatment decisions.
While current molecular assays require sophisticated instrumentation and technical expertise, AI-based pathomic tools can be captured with scanners or smartphones, uploaded electronically, and analyzed centrally with minimal cost. However, Sparano noted that the study was not designed to predict chemotherapy benefit or the benefit of continuing adjuvant endocrine therapy beyond five years.
This research was a collaboration between the ECOG-ACRIN Cancer Research Group and Caris Life Sciences, supported by various organizations, including the Breast Cancer Research Foundation, the National Cancer Institute, and the U.S. Postal Service Breast Cancer Research Stamp Fund. Sparano's consulting and research support relationships with several pharmaceutical companies were also acknowledged.
