Duke Health algorithm improves OR scheduling accuracy by 13% across 33,000 surgical cases

Human schedulers at Duke University Hospital struggled to accurately predict surgical time needed in the operating room, leading to scheduling errors that disrupted clinical workflow and generated costly overtime. Operating rooms are among the most expensive resources in a hospital, making inefficiencies particularly costly.

A team of Duke Health data scientists, clinicians, and researchers trained three AI models on thousands of historical surgical cases to predict procedure duration. The algorithm was validated in a study published in the Annals of Surgery and subsequently deployed directly into clinical operations at Duke University Hospital.

The machine-learning models were 13% more accurate in predicting surgical time compared to human schedulers. The algorithm has been used on over 33,000 cases. Reductions in scheduling errors and resulting overtime could save approximately $79,000 in overtime labor expenses over a four-month period.

• Even a modest improvement in scheduling accuracy (13%) translates to meaningful cost savings and workflow improvements at scale.
• Training AI models on institution-specific historical data enables rapid real-world deployment with immediate clinical utility.
• Reducing OR overtime is a high-leverage target — small gains compound across thousands of cases per year.