Comparison of Machine Learning Algorithms for Predicting Spine Surgery Duration

Area of Science:

• Spine Surgery
• Machine Learning
• Operating Room Management

Background:

• Accurate prediction of spine surgery duration is crucial for efficient operating room management.
• Discrepancies between estimated and actual surgical times disrupt scheduling and resource allocation.
• Machine learning (ML) offers a potential solution for improving duration prediction.

Purpose of the Study:

• To develop and compare ML algorithms for predicting spine surgery duration.
• To identify the most effective ML approach for this prediction task.
• To assess the impact of ML-based prediction on surgical scheduling accuracy.

Main Methods:

• Retrospective analysis of 3376 spine surgery patient records.
• Development of four models (Random Forest, XGBoost, MLP, WLS) using pre-operative data.
• Evaluation of models on an internal test set and comparison with a full-information model.
• Utilized SHapley Additive exPlanations (SHAP) for predictor importance analysis.

Main Results:

• XGBoost achieved the best predictive performance with an MSE of 3014.6 min² (RMSE 54.9 min) and R² of 0.622.
• ML models significantly reduced mean absolute error by ~20 min compared to surgeon estimates alone.
• Surgeon identity was the most influential predictor (24.7-41.9%), followed by procedure type and estimated duration.

Conclusions:

• Machine learning models substantially enhance spine surgery duration prediction accuracy.
• XGBoost demonstrates superior performance, improving operating room scheduling efficiency.
• Surgeon identity is a critical factor in predicting surgical duration, necessitating its inclusion in predictive models.