Two presentations at the 2025 San Antonio Breast Cancer Symposium (SABCS) highlighted new artificial intelligence (AI) tools and systems for improving distant recurrence risk stratification among patients with early-stage breast cancer.
The first tested multimodal AI models with imaging, clinical, and molecular data to predict both early and late breast cancer recurrence in patients from the TAILORx trial. The second was the external validation study for a multimodal-multitask deep learning algorithm to estimate late distant recurrence risk patients with hormone receptor–positive early breast cancer from the TAILORx trial.
Both studies—which help to improve the accuracy of individualized prognostic assessments beyond standard methods for patients with early breast cancer—were the result of a major multiyear collaboration between the ECOG-ACRIN Cancer Research Group and Caris Life Sciences to transform recurrence risk assessment with the use of AI.
“This public-private partnership represents a methodological, logistical, and collaborative integration of datasets from the historically impactful TAILORx trial to further extend the benefits for breast cancer patients," said Peter J. O'Dwyer, MD, the Group Co-Chair of ECOG-ACRIN. “The advance in personalized medicine afforded in this work, in turn, helps to advance the potential of AI to refine treatment and improve outcomes.”
“By integrating imaging, clinical data, and molecular profiling, we are advancing beyond single-dimension diagnostics to deliver a more precise and comprehensive understanding of recurrence risk in breast cancer,” said George W. Sledge, Jr, MD, the Executive Vice President and Chief Medical Officer of Caris. “The development of these models underscores the transformative power of multimodal AI and machine learning in precision oncology.”
Multimodal AI Models
The first study was presented by Joseph A. Sparano, MD, Chief of the Division of Hematology and Oncology at the Mount Sinai Tisch Cancer Center (Abstract GS1-08).
Researchers analyzed AI models using combinations of imaging, clinical, and molecular data to determine the optimal multimodal model for predicting early and late breast cancer recurrence in patients with early breast cancer, as compared with the Oncotype DX recurrence score alone. Samples included in the study came from patients in the TAILORx trial.
“Although the TAILORx trial was the first randomized trial to establish the role of the 21-gene recurrence score to guide chemotherapy use in early breast cancer, our goal was to take one step further in personalizing cancer therapy by developing a new diagnostic test using tumor specimens derived from the trial,” said Dr. Sparano.
A combination of clinical, molecular, and histopathological integrated features in the multimodal model provided the greatest prognostic information and the strongest performance overall (concordance index [C-index] for overall distant recurrence = 0.705; hazard ratio [HR] for high vs low risk = 3.6; P < .001; C-index for late distant recurrence = 0.656; HR = 2.84; P < .001), and will be validated further going forward.
Additionally, the molecular-only models significantly strengthened the prognostic accuracy for early distant recurrence through 5 years (C-index = 0.757), while histopathologic data in imaging-only models drove improvements in accuracy for late distant recurrence (C-index = 0.637; HR = 1.9; P = .001).
“We found that the expanded gene panel was a strong predictor of early recurrence within 5 years after diagnosis, the pathomic imaging was a strong predictor of late recurrence after 5 years, and when combined, a test which added both features to the prognostic information provided by clinicopathologic factors was the strongest predictor of distant recurrence out to 15 years,” he said.
Clarity BCR
The second study, exploring the use of the deep-learning algorithm Clarity BCR, was presented by Eleftherios Mamounas, MD, MPH, Medical Director of the Breast Program at AdventHealth Cancer Institute (Abstract RF3-07).
Clarity BCR was initially developed and validated in the NSABP B-42 trial to stratify patients by risk and determine which postmenopausal women with hormone receptor–positive early breast cancer would gain the most benefit from extended endocrine therapy. Researchers then conducted an external validation study of Clarity BCR in the TAILORx translational cohort of patients who had completed at least 4.5 years of endocrine therapy and remained disease-free at 5 years.
The algorithm integrates image features from whole-slide images, captured using a Pramana SpectralHT scanner, and clinical variables to come up with a continuous risk score and a dichotomized risk group.
In the overall TAILORx cohort (n = 6,516), 17.4% of patients were classified by Clarity BCR as high risk and 82.6% as low risk; in the external validation cohort (n = 4,469), 17.2% were high risk and 82.8% were low risk.
The C-index was 0.59 for Clarity BCR and 0.54 for Oncotype DX for prognostic discrimination.
For the entire follow-up period, high-risk patients showed significantly worse distant recurrence outcomes (HR = 1.73; 95% confidence interval [CI] = 1.40–2.15; P < .001). The distant recurrence–free interval at 15 years was 83.7% in the high-risk group and 90.8% in the low-risk group.
Multivariable analysis confirmed the independent prognostic value of Clarity BCR (HR = 1.32; 95% CI = 1.02–1.72; P = .037) and the C-index for distant recurrence–free interval was 0.59 for Clarity BCR and 0.60 for Oncotype DX.
The researchers believe these findings support the potential clinical utility of Clarity BCR to guide long-term treatment decisions in patients with hormone receptor–positive breast cancer.
Disclosure: For full disclosures of the study authors, visit abstractsonline.com.
