A Data-Driven Approach: Investigating Prognostic Factors for Overall Survival in Breast Conserving Surgery (BCS) using Machine Learning

Abstract

Abstract Background Breast-conserving surgery (BCS) is a viable treatment for early-stage breast cancer, but post-operative recurrence is a significant concern linked to mortality. This study leverages Machine Learning and healthcare data to better identify patients at risk of recurrence. The goal is to assess how effectively the model predicts survival factors in breast cancer patients post-BCS. Methods This study retrospectively analyzed 1518 breast cancer patients, of whom 430 were excluded due to unknown post-surgery recurrence status from January 1993 to June 2021 using XGBoost model, optimized with grid search and 5-fold cross-validation. Feature importance was determined using the Shapley value technique, and data was collected with SPSS Statistics, Version 28.0, IBM. Results The machine learning model showed high effectiveness in predicting patient outcomes, with notable metrics like accuracy (0.947) and precision (0.897). Key findings emphasize the importance of clear surgical margins and reveal that demographic factors like age and race significantly affect prognosis, while luminal subtype and comorbidity are less influential. These insights are crucial for understanding disease recurrence in breast cancer patients after BCS and radiotherapy. Conclusion The XGBoost machine learning model demonstrated outstanding predictive performance for outcomes in breast cancer patients receiving BCS and radiotherapy. It confirmed the critical importance of clear surgical margins during initial surgery for prognosis. Demographic factors, especially age and race, were identified as significant predictors of patient outcomes.