Evaluating the Survival Impact of Neoadjuvant Therapy and Development of Personalized Machine Learning Survival Predictive Model for Breast Cancer Patients Eligible for Breast‐Conserving Surgery

ABSTRACT The long‐term survival benefit of neoadjuvant therapy (NAT) in breast cancer patients eligible for breast‐conserving surgery (BCS) remains uncertain. This retrospective cohort study analyzed 94,677 BCS‐eligible patients from the SEER database (2010–2020), including 8565 who received NAT. After propensity score matching ( n = 5734 each), NAT significantly improved overall survival (OS) only in patients with triple‐negative (HR = 0.79), ER‐negative (HR = 0.80), and stage IIA (HR = 0.81) disease. No OS benefit was observed in HER2‐positive patients despite high response rates. To guide treatment decisions, two machine learning models using Random Survival Forest were developed to predict 5‐year OS, showing good discrimination (C‐index: 0.743 for BCS, 0.690 for NAT‐BCS). SHAP analysis identified age, stage, and breast subtype as key prognostic factors. Cross‐stratification based on predicted OS revealed that 8.9% of BCS patients could benefit from NAT, while 90.8% of NAT‐BCS patients might safely omit it. Patients whose treatment matched model recommendations had significantly better survival. These findings suggest that NAT provides limited survival benefit in BCS‐eligible patients, with the advantage concentrated in specific subgroups. Predictive modeling offers a clinically useful approach to personalize NAT use, potentially reducing unnecessary treatment while identifying those most likely to benefit.