AI Model Predicts Late Distant Recurrence Risk in HR-Positive Breast Cancer

According to results from a study published in Cancer Research Communications, a multimodal, multitask deep learning model that integrates digitized hematoxylin and eosin (H&E) whole-slide pathology images with clinicopathologic variables predicted late distant recurrence (DR) risk in hormone receptor (HR)-positive early breast cancer.¹ Caris Life Sciences disclosed the findings, developed with the NSABP Foundation/NRG Oncology and the ECOG-ACRIN Cancer Research Group.²

HR-positive disease accounts for approximately 70% to 80% of breast cancer diagnoses and carries a risk of recurrence that can persist beyond the initial 5 years of endocrine therapy (ET). Extended ET may lower that risk but adds years of side effects, and identifying which patients stand to benefit has remained a challenge.

Study Design

The model was developed using banked tumor specimens from 2271 patients enrolled in the NSABP B-42 trial (NCT00382070) with 5-fold cross-validation, then externally validated in 4300 banked specimens from patients in the TAILORx trial (NCT00310180) who remained disease-free 5 years after diagnosis. The model generates risk predictions for late DR in both node-positive and node-negative HR-positive breast cancer, and prognostic performance was assessed using hazard ratios and absolute risk differences; exploratory analyses examined extended letrozole therapy (ELT) benefit across model-defined risk groups.

Efficacy Results

In the NSABP B-42 cohort, the model identified patients with substantially different outcomes, with a 10-year absolute distant recurrence risk difference of nearly 8% between high- and low-risk groups. External validation in TAILORx confirmed the model’s prognostic performance; it independently predicted late DR risk even after adjusting for established clinical risk factors and the Oncotype DX Recurrence Score.

Exploratory analyses further suggested that patients classified as high risk experienced greater absolute benefit from ELT than those classified as low risk, supporting the model’s potential to inform discussions about extending endocrine therapy beyond 5 years.

“This study underscores the transformative potential of artificial intelligence to extract clinically meaningful insights from routinely collected data. By integrating AI-powered analysis of standard pathology images with clinical variables, this approach offers a promising path toward more precise risk stratification and may help inform individualized decisions regarding extended endocrine therapy for patients with hormone receptor-positive breast cancer,” said George W. Sledge, MD, chief medical officer, Caris Life Sciences, in a news release.¹

Clinical Context and Limitations

Existing genomic assays provide prognostic insight in HR-positive breast cancer but can be limited by cost, accessibility, and turnaround time; the investigators noted that AI-based analysis of routinely available pathology slides and clinical data could offer a scalable alternative or complement to those tools. Caris launched a related commercial test, Caris MI Clarity, in May 2026, designed to report both early and late DR risk (years 0-5 and 5-15) for postmenopausal patients with HR-positive, HER2-negative, node-negative early breast cancer at diagnosis, along with chemotherapy and extended endocrine therapy decision support.

The model was first previewed at the 2025 San Antonio Breast Cancer Symposium, where it was reported to have been trained in the NSABP B-42 trial and validated in TAILORx.³ The publication in Cancer Research Communications represents the peer-reviewed report of that work.

REFERENCES

1. Mamounas EP, Chen M, Sparano JA, et al. Development and validation of a multimodal-multitask deep learning approach for estimating late distant recurrence risk in HR-positive early breast cancer. Cancer Res Commun. 2026;6(7):1825-1835. doi:10.1158/2767-9764.CRC-26-0362

2. Caris Life Sciences, ECOG-ACRIN and NRG Oncology study demonstrates multimodal AI approach to predict late distant recurrence risk in HR+ early breast cancer. News release. Caris Life Sciences. August 6, 2026. Accessed August 7, 2026. https://tinyurl.com/4t88uyua

3. A multimodal-multitask deep learning model trained in NSABP B-42 and validated in TAILORx for late distant recurrence risk in HR+ early breast cancer. Caris Life Sciences. Presented at the 2025 San Antonio Breast Cancer Symposium; December 10, 2025. Accessed August 7, 2026. https://tinyurl.com/mr2dn2mb