An Ensemble Learning Approach to Improving Prediction of Case Duration for Spine Surgery: Algorithm Development and...

Area of Science:

• Neurosurgery and Health Informatics
• Data Science in Healthcare
• Operating Room Management

Background:

• Accurate estimation of surgical case duration is vital for operating room efficiency.
• Traditional predictive methods in spine surgery, like statistical models, are often less sophisticated.
• Machine learning has been applied to predict outcomes but not specifically case duration.

Purpose of the Study:

• To evaluate an ensemble learning approach for enhancing the accuracy of scheduled spine surgery durations.
• To compare machine learning models against the institution's current predictive methods.

Main Methods:

• Retrospective analysis of 3189 spine surgery cases over 4 years.
• Comparison of multivariable linear regression, random forest, bagging, and XGBoost models.
• Evaluation using R-squared, RMSE, explained variance, and MAE with k-fold cross-validation.
• Feature importance determined using SHAP (Shapley Additive Explanations) analysis.

Main Results:

• The institution's current method showed poor prediction accuracy (R-squared=0.213).
• XGBoost regression demonstrated superior performance with an R-squared of 0.770, RMSE of 92.95 minutes, and MAE of 44.31 minutes.
• Key predictive features included body mass index, spinal fusions, surgical procedure, and number of spine levels.

Conclusions:

• Ensemble learning, specifically XGBoost regression, significantly enhances the accuracy of spine surgery time estimations.
• This approach offers a more reliable tool for operating room scheduling and efficiency.