Artificial Intelligence Length-of-Stay Forecasting and Pediatric Surgical Capacity

Area of Science:

• Artificial Intelligence in Healthcare
• Machine Learning for Clinical Operations
• Surgical Patient Flow Optimization

Background:

• Hospitals face challenges with unpredictable patient volumes, impacting surgical capacity.
• Variable inpatient loads strain resources and complicate bed management.
• AI offers a potential solution for stabilizing hospital capacity.

Purpose of the Study:

• To predict postoperative length of stay (LOS) for elective surgery patients using machine learning.
• To implement an LOS prediction model to optimize surgical scheduling and hospital bed capacity.
• To evaluate the model's impact on clinical operations and resource utilization.

Main Methods:

• A retrospective cohort study at a tertiary children's hospital.
• Used Extreme Gradient Boosting (XGBoost) to predict postoperative LOS.
• Implemented and evaluated the model in a pre/post-implementation design over 22 months.

Main Results:

• The LOS prediction model achieved 85.6% accuracy with a 0.6-day mean absolute error.
• Post-implementation, the median number of weekday elective surgeries increased by 5.
• Significant reduction in midweek bedded day variation (43-44% IQR decrease) and underused capacity (33% to 10%).

Conclusions:

• Machine learning-based LOS prediction effectively optimizes elective surgical scheduling.
• The model reduces day-to-day variation in surgical procedures and hospital bed occupancy.
• Implementation led to increased surgical throughput and decreased hospital bed underuse.