Surgery duration: Optimized prediction and causality analysis.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Surgical Operations Research

Background:

• Accurate estimation of surgery duration (DOS) is crucial for efficient operating room utilization and reduced patient wait times.
• Previous studies have focused on prediction models but often lacked comprehensive feature sets and causal analysis.

Purpose of the Study:

• To develop a superior supervised nonlinear regression model for predicting DOS.
• To identify influential and causal features affecting DOS.
• To analyze the causal relationship between identified features and DOS.

Main Methods:

• Implemented various machine learning algorithms trained on a comprehensive dataset of 23,293 surgery records over 10 years.
• Introduced novel features combined with existing ones for a robust feature set.
• Utilized feature importance and causal inference methods for analysis.

Main Results:

• The developed DOS prediction model achieved a mean absolute error (MAE) of 14.9 minutes, outperforming existing models.
• Gradient Boosted Trees (GBT) algorithm yielded the best performing model.
• Identified top 10 influential and 10 causal features, with 40% showing a significant causal relationship with DOS.

Conclusions:

• The novel feature set and advanced model significantly improve DOS prediction accuracy.
• The identified causal relationships offer opportunities for hospitals to actively influence and manage surgery duration.
• The model's feature importance analysis provides explainability for predictions, aiding clinical understanding.