Development of Predictive Model of Surgical Case Durations Using Machine Learning Approach.

Area of Science:

• Healthcare Management
• Machine Learning in Medicine
• Surgical Operations Research

Background:

• Optimizing operating room (OR) utilization is crucial for hospital efficiency.
• Accurate surgical case duration prediction is key to effective OR management.
• Traditional estimation methods often lack the precision needed for advanced optimization.

Purpose of the Study:

• To develop and evaluate department-specific Random Forest models for predicting surgical case durations.
• To improve upon the accuracy of traditional estimation methods and general machine learning models.
• To identify key factors influencing surgical duration through feature importance analysis.

Main Methods:

• Applied Random Forest, XGBoost, Linear Regression, LightGBM, and CatBoost algorithms to a comprehensive surgical dataset.
• Assessed model performance using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-Squared (R²).
• Utilized SHAP analysis for feature importance interpretation.

Main Results:

• Department-specific Random Forest models achieved superior performance with MAE of 16.32, RMSE of 31.19, and R² of 0.92.
• Random Forest models significantly outperformed general models and conventional estimation techniques.
• Key predictors identified include morning operation timing, ICU ward assignment, operation codes, and surgeon ID.

Conclusions:

• Tailoring machine learning models to specific surgical departments substantially enhances prediction accuracy for surgical case durations.
• Department-specific Random Forest models offer a more reliable tool for optimizing surgical scheduling and OR management.
• Leveraging data-driven, customized models is vital for improving healthcare operational efficiency and outcomes.