Machine Learning for the Prediction of Procedural Case Durations Developed Using a Large Multicenter Database:...

Area of Science:

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

Background:

• Accurate procedural case duration predictions are crucial for perioperative resource management and patient communication.
• Current estimation methods often fall short, impacting operational efficiency.
• Machine learning offers a promising avenue for enhancing procedure duration predictions.

Purpose of the Study:

• To evaluate the efficacy of a scalable machine learning algorithm for predicting surgical case durations.
• To determine if predictions can be achieved within acceptable tolerance limits across multiple institutions.
• To optimize operating room resource allocation through improved duration forecasting.

Main Methods:

• Development and comparison of deep learning, gradient boosting, and ensemble machine learning models.
• Utilized perioperative data from three distinct time points: scheduling, patient arrival, and procedure start.
• Performance assessed using mean absolute error (MAE) and proportion of predictions within 20% of actual duration, compared against a linear regression baseline.

Main Results:

• The gradient boosting machine model demonstrated superior performance, with an MAE of 34 minutes and 46% of predictions within 20% of actual duration.
• This represents a significant improvement over the baseline linear regression model (MAE 43 min, 39% within 20%).
• Key predictive features included surgeon's historical procedure duration, specific keywords in procedure text, and time of day.

Conclusions:

• Nonlinear machine learning models can generate highly accurate, automated, and explainable predictions for procedure duration.
• These models offer scalability across different healthcare settings.
• Implementation can lead to improved operational efficiency and resource management in surgical settings.