AI Model Outperforms Oncotype DX in Predicting Recurrence Risk for Common Breast Cancer Subtype
San Antonio, TX – A novel artificial intelligence (AI) model demonstrates superior accuracy in predicting recurrence risk for the moast common subtype of breast cancer – hormone receptor (HR)-positive, HER2-negative – compared to the widely used Oncotype DX (ODX) 21-gene recurrence score. These promising results, presented at the San Antonio Breast Cancer Symposium (SABCS) held December 9-12, 2025, suggest a potential paradigm shift in risk assessment and treatment planning for a significant proportion of breast cancer patients.
HR-positive, HER2-negative breast cancer accounts for the majority of breast cancer diagnoses. A critical challenge in managing this subtype is the ample risk of late recurrence – events occurring more than five years after initial diagnosis, representing at least 50% of all recurrences. While the Oncotype DX assay is a standard tool for assessing distant recurrence risk and predicting chemotherapy benefit, its ability to accurately forecast recurrence beyond the five-year mark is limited.
“Our goal was to develop a new diagnostic test that provides a more extensive estimation of recurrence risk, specifically addressing the challenge of late recurrence,” explained Dr. Joseph A.Sparano, chief of the Division of Hematology and Oncology at the Mount Sinai Tisch Cancer Center. “We leveraged data from the TAILORx trial to create an AI model that integrates digitized pathology images, molecular characteristics, and clinical data to provide a more accurate prognosis, extending out to 15 years.”
The research team utilized a comprehensive dataset comprising digitized tissue images and molecular RNA expression data from 4,462 tumor samples, alongside corresponding clinical data from participants in the TAILORx study. this data was used to train and validate multiple risk models.The performance of these models was rigorously compared to the oncotype DX results used in the TAILORx trial, utilizing the concordance index (C-index) – a statistical measure of a test’s ability to correctly rank recurrence risk. A C-index of 0.5 indicates performance no better than chance, while a C-index of 1 represents perfect prediction.
The resulting multimodal model, designated ICM+ (integrating pathomic imaging (I), clinical (C), and expanded molecular (M+) data), demonstrated a statistically significant improvement over Oncotype DX in predicting 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) within the training and cross-validation set of 2,806 patients. This superior performance was further validated in an self-reliant holdout set of 1,621 patients, showing a C-index of 0.733 for overall recurrence versus 0.631 for ODX, and 0.705 versus 0.527 for late distant recurrence.
These findings suggest that ICM+ has the potential to become a new standard in recurrence risk assessment for women with HR-positive, HER2-negative, node-negative breast cancer – a subtype representing approximately half of all breast cancer cases in the United States.
“this study highlights the transformative potential of AI in developing more accurate diagnostic tests,ultimately leading to more individualized treatment decisions,” stated Dr. Sparano. He further emphasized the accessibility advantages of AI-based pathology tools. “Current molecular assays require specialized instrumentation and expertise. AI-driven pathomic tools can utilize digitized slides routinely generated in clinical practice, captured with readily available scanners or even smartphones, and analyzed centrally at a minimal cost.”
The researchers acknowledge a limitation of the study: it was not designed to assess the predictive value of the model for chemotherapy benefit or the duration of adjuvant endocrine therapy beyond five years.
This research represents a significant step