Machine Learning Models for Predicting Surgical Case Times in Breast Cancer Procedures

Accurate estimation of surgical case duration is essential to improve operating room efficiency and optimize hospital resources. This study analyzed 2,265 breast cancer surgeries performed at the Chilean National Cancer Institute between June 2019 and December 2024 for predicting surgical case duration. Several machine learning regressor models — such as linear regularized regressions, non-linear models, and ensemble models— were compared with the hospital’s current estimation method using RMSE, MAE, and the coefficient of determination. The best-performing model, XGBoost, reduced the MAE from 39.54 to 21.80 minutes, the RMSE from 49.72 to 28.92 minutes, and improved the coefficient of determination from -0.01 to 0.66. Feature importance analysis revealed that the surgeon’s time estimation —based on experience— was the second most influential predictor, followed by operational, procedure, team and oncologic features, underscoring the need to complement clinical expertise with data-driven insights. These findings demonstrate the potential of ML models to enhance surgical time prediction, supporting more reliable operating room scheduling and improved resource utilization in oncology surgery.