Development of Multiservice Machine Learning Models to Predict Postsurgical Length of Stay and Discharge Disposition...

Area of Science:

• Machine learning applications in healthcare analytics
• Predictive modeling for hospital operations
• Surgical outcomes and resource management

Background:

• Current surgical scheduling prioritizes operating room availability over downstream resource needs.
• Predicting postsurgical length of stay and discharge disposition is crucial for efficient hospital management.
• Early prediction facilitates better resource allocation and discharge planning.

Purpose of the Study:

• To develop machine learning models for predicting postsurgical length of stay (LOS) and discharge disposition (DD).
• To utilize only data available at the time of surgical case posting for these predictions.
• To apply these models across multiple surgical services for adult elective inpatient cases.

Main Methods:

• Retrospective study of 63,574 adult patients undergoing elective inpatient surgery.
• Development of gradient-boosting decision tree classification models.
• Prediction of LOS (short, medium, prolonged) and DD (home vs. nonhome) using case posting data.

Main Results:

• The LOS model achieved an area under the receiver operating characteristic curve (AUC) of 0.81.
• The DD model demonstrated an AUC of 0.88 for home versus nonhome prediction.
• Incorporating relative value unit and historical LOS improved short and prolonged LOS prediction accuracy.

Conclusions:

• Machine learning models can effectively predict postsurgical LOS and DD at the time of case posting.
• These models support improved case scheduling, resource allocation, and bed utilization.
• Predictive models aid in earlier discharge planning and prevention of case cancellations due to bed unavailability.