As reported in The Lancet Oncology, Shamai et al have developed an artifical intelligence (AI) model based on digital histopathology slide images and clinical features to predict the Oncotype DX 21-gene recurrence score (RS) in patients with hormone receptor–positive, HER2-negative invasive breast cancer.
As stated by the investigators, “Genomic assays such as Oncotype DX have transformed adjuvant treatment selection for hormone receptor–positive, HER2-negative, early breast cancer, but remain inaccessible to many patients because of high cost and logistical barriers.”
Study Details
The model was pretrained for predicting RS on 171,189 histopathology slides. In addition, it was fine-tuned and validated using data from the TAILORx trial (n = 8,284; 5,877 in the training set and 2,407 in the test set) and externally validated in six independent patient cohorts (Carmel, Haemek, and Sheba medical centers in Israel; University of Chicago Medical Center; Australian Breast Cancer Tissue Bank; and Cancer Genome Atlas Breast Invasive Carcinoma project; n = 5,497).
Key Findings
Among 2,407 patients in the TAILORx test set, the AI model classified 1,097 (45.6%) patients as low risk, 1,021 (42.4%) as intermediate risk, and 289 (12.0%) as high risk. In identifying high genomic risk (RS ≥ 26), the receiver operating characteristic AUC was 0.898 (95% confidence interval [CI] = 0.879–0.913).
AI-based risk stratification was prognostic for recurrence-free interval (hazard ratio [HR] = 2.61, 95% CI = 1.68–4.04), distant recurrence–free interval (HR = 2.88, 95% CI = 1.73–4.79), and disease-free survival (HR = 1.32, 95% CI = 0.92–1.89).
Chemotherapy benefit in disease-free survival was predicted in premenopausal patients classified as high risk by the AI model (HR = 0.63, 95% CI = 0.46–0.86) but not in postmenopausal patients classified as low risk (HR = 0.94, 95% CI = 0.78–1.12). A total of 151 (31.3%) clinically high-risk postmenopausal women on MINDACT criteria were reclassified as low risk with no chemotherapy benefit.
Analysis in the six external validation cohorts showed AUCs for RS ≥ 26 ranging from 0.858 to 0.903.
The investigators concluded: “These findings show that AI applied to routine histopathology can serve as a practical and scalable tool for guiding chemotherapy decisions in hormone receptor–positive, HER2-negative, early breast cancer. This approach has the potential to reduce unnecessary chemotherapy and broaden access to precision oncology, particularly in resource-limited settings where genomic testing remains unavailable or unaffordable.”
Gil Shamai, PhD, of Taub Faculty of Computer Science, Technion–Israel Institute of Technology, Haifa, Israel, is the corresponding author for The Lancet Oncology article.
DISCLOSURE: The study was funded by the Zimin Institute for Artificial Intelligence Solutions in Healthcare, Israel Cancer Research Fund, and others. For full disclosures of the study authors, visit thelancet.com.
