H&E to Oncotype DX: Predicting recurrence risk in HR+/HER2- breast cancer

• High Predictive Accuracy: Superior performance with AUCs of 0.89 and 0.86. • Generalizability: Validated on a public multi-center dataset (TCGA) and an institutional dataset (OSU). • Enhanced Interpretability: Identifies key biomarkers like LVI and Comedonecrosis. • Cost and Time Efficiency: Reduces need for costly and time-consuming genomic assays. The Oncotype DX® Recurrence Score (ODX) is a clinically validated prognostic assay for hormone receptor-positive early-stage breast cancer. Despite its utility, widespread adoption remains limited due to high costs and lengthy turnaround times. In this study, we propose AnchorMIL, a novel framework that predicts ODX scores directly from H&E stained whole-slide images, with the potential to decrease dependence on molecular profiling. AnchorMIL employs an anchored regression-classification mechanism to predict both continuous risk scores and binary risk predictions. On the TCGA-BRCA and OSU cohort, AnchorMIL achieved AUCs of 0.89 and 0.86, respectively, Beyond predictive accuracy, AnchorMIL demonstrates promising generalizability, and its interpretability reveals biologically meaningful features of aggressive tumor biology. The model captures complex prognostic interactions, and prioritizes dominant risk factors in high-risk cases. AnchorMIL offers a scalable, cost-effective tool for risk stratification, with the potential to reduce reliance on genomic assays, accelerate treatment decisions, and support equitable breast cancer care.