The ECOG-ACRIN Cancer Research Group has announced a new AI-powered test designed to improve prediction of distant recurrence risk in patients with hormone receptor–positive, HER2-negative early-stage breast cancer. The test, called IICM+, was developed and independently validated using data and tumor specimens from participants in the landmark TAILORx trial, according to findings published in npj Breast Cancer.

Although the 21-gene Recurrence Score, Oncotype DX, is widely used to estimate recurrence risk and guide chemotherapy use in patients with hormone receptor–positive, HER2-negative, lymph node–negative early breast cancer, it is more effective at predicting recurrence during the first 5 years than later recurrence. More than half of distant recurrences in this breast cancer subtype occur more than 5 years after surgery, providing the rationale for developing a model capable of improving both early- and late-recurrence risk prediction.

“Powered by AI integrating clinical, molecular, and histopathology data, this new test provides more reliable prognostic information for breast cancer recurrence,” said lead author Joseph A. Sparano, MD, of the Icahn School of Medicine at Mount Sinai, Tisch Cancer Institute, New York.

Development and Validation of IICM+

The investigators developed IICM+ by combining digitized histopathology images with clinical characteristics, including patient age and tumor size and grade, as well as molecular information from an expanded 42-gene panel. AI and deep-learning methods were applied to digitized images of archived pathology slides from the TAILORx biorepository. All features were processed through a multimodal transformer with a fusion layer, transformer block, and attention pooling block to output a continuous risk score denoting a binary low or high risk. 

The model was developed using data and tumor specimens from 2,808 TAILORx participants with five-fold cross-validation. It was subsequently evaluated in an independent validation cohort of 1,621 participants whose data had not been included during model development. Both cohorts had a median follow-up exceeding 11 years.

"Our multimodal AI model integrates multiscale histopathology features with routine clinical variables and gene expression to provide robust prognostic stratification for distant recurrence," the study authors noted. 

TAILORx included patients with hormone receptor–positive, HER2-negative, lymph node–negative early-stage breast cancer and established the clinical utility of the 21-gene Recurrence Score for estimating recurrence risk and informing chemotherapy use. The investigators evaluated the ability of IICM+ to predict overall distant recurrence as well as early distant recurrence, occurring within 5 years, and late distant recurrence, occurring after 5 years.

Predicting Early and Late Distant Recurrence

In the independent validation cohort, IICM+ significantly outperformed the 21-gene Recurrence Score in distinguishing among patients with different risks of distant recurrence, based on the concordance index (C-index).

For overall distant recurrence, the C-index was 0.735 with IICM+ compared with 0.578 with the Recurrence Score (P < .001). For early distant recurrence, the C-indices were 0.791 and 0.722, respectively (P = .046). The difference was also observed for late distant recurrence, for which the C-index was 0.710 with IICM+ vs 0.514 with the Recurrence Score (P < .001).

The AI model also identified differences in recurrence risk within established Recurrence Score categories. Among patients with a Recurrence Score of 0 to 25, representing lower genomic risk, IICM+ identified a subset with a higher observed risk of distant recurrence. Conversely, among patients with Recurrence Scores of 26 to 100, representing higher genomic risk, the model identified a subset whose observed recurrence risk was lower than would have been suggested by the Recurrence Score alone.

George W. Sledge, Jr, MD, Executive Vice President and Chief Medical Officer of Caris Life Sciences, said the integration of clinical information, next-generation sequencing data, and imaging data through AI provided a new approach to obtaining prognostic information from breast tumors.

However, the findings do not establish IICM+ as a tool for selecting or modifying treatment. Further studies are needed to validate the model in other patient populations and determine whether using its predictions to guide treatment decisions can improve patient outcomes. Such studies are currently in development.

DISCLOSURES: The study was supported by the National Cancer Institute of the National Institutes of Health as well as the Breast Cancer Research Foundation, the Komen Foundation, and the Breast Cancer Research Stamp issued by the United States Postal Service. Several of the authors are employees of Caris Life Sciences, Inc. Drs. Sparano, Gray, and Wang received grant support from the National Cancer Institute. All other authors declared no competing financial or non-financial interests. 

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.