Development, Deployment, and Implementation of a Machine Learning Surgical Case Length Prediction Model and...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Surgical Operations Management

Background:

• Operating rooms are critical, costly hospital resources, generating significant revenue.
• Accurate surgical case length prediction is essential for efficient scheduling and resource utilization.

Purpose of the Study:

• To develop a machine learning model for predicting surgical case length using limited data available at case creation.
• To improve operating room scheduling and resource management through enhanced prediction accuracy.

Main Methods:

• Developed a gradient-boosting machine learning model incorporating a similarity cascade for case complexity and surgeon influence.
• Customized the model's loss function to balance over- and under-prediction.
• Established a production pipeline for seamless institutional deployment.

Main Results:

• The model outperformed human schedulers in predicting surgical case length, reducing underpredictions by 11.2% and increasing accuracy within 20% of actual length by 5.9%.
• Prospective analysis showed gradual scheduler adoption and improved prediction accuracy across 33,815 surgical cases.
• The model assisted schedulers, improving predictions by 3.4% within 20% of actual length and reducing underpredictions by 4.3%.

Conclusions:

• A novel framework was created for accurate surgical case length prediction at the time of posting.
• The developed machine learning model is actively used daily, demonstrating its practical utility.
• The framework has the potential for deploying future machine learning models in healthcare settings.