Overview
Early breast cancer is generally treated with surgery, and doctors use clinical information and genomic tests to estimate recurrence risk and decide whether patients need chemotherapy or other additional treatments. However, no comprehensive, real-world predictive model exists that combines all relevant clinical information to guide individualized treatment decisions after surgery.
Researchers used Flatiron Health’s US Breast Cancer Panoramic Database, inclusive of more than 985,000 patients with breast cancer, to analyze data from over 158,000 patients with hormone-receptor positive, HER2-negative early breast cancer who underwent surgery without neoadjuvant therapy between 2016-2023. They used machine learning to develop a predictive model that incorporates various clinical, demographic, and genomic factors, including age, tumor grade and stage, smoking history, socioeconomic factors, and standard of care biomarkers. The model successfully stratified patients into three risk groups (low, medium, high) with median recurrence-free survival ranging from not reached (low-risk) to 8.5 years (high-risk). The researchers also identified risk predictors specific to younger patients (age ≤45), where race and tumor cell proliferation markers became more important than in older patients.
Why this matters
This large-scale analysis provides valuable insights into which patients are at highest risk for recurrence after surgery. By accurately identifying high-risk patients in the real world, oncologists can make better-informed decisions about treatment intensity—escalating therapy for those most likely to relapse while potentially avoiding unnecessary treatment in lower-risk patients. These findings suggest real-world data models could complement existing clinical tools to personalize breast cancer treatment.