Improving Surgery Scheduling Times with AI

Scheduling elective surgeries is a complex operational challenge for hospitals. Each week, hospitals must assign surgeries to a limited number of operating-room (OR) time blocks, while also managing uncertainty in how long each procedure will take. Surgeons provide time estimates for their cases, but these estimates are often noisy and conservative, which can result in scheduling inefficiencies. When surgeries run longer than expected, hospitals incur overtime costs; when they finish early, valuable operating room capacity goes unused. Cases that cannot be scheduled must also be deferred to future weeks, increasing patient wait times.

Researchers Roham Bahri and Borzou Rostami from the Alberta School of Business at the University of Alberta, and Vahid Roshanaei from the Rotman School of Management at the University of Toronto, developed a decision-aware AI scheduling framework designed to reduce these operational costs and improve surgical throughput.

The model integrates machine learning with optimization techniques to improve how surgery durations are predicted and how schedules are generated. Instead of training the AI model solely to predict surgery times accurately, the approach learns in a way that directly improves scheduling decisions. This method accounts for the real costs hospitals face overtime, idle operating room time, and delayed surgeries when determining optimal schedules.

To evaluate the system, the researchers used three years of elective surgery data from a large Toronto academic hospital, including information about procedures, surgeons, patient characteristics, and actual surgery durations. The model was tested across hundreds of weekly scheduling scenarios.

The results showed significant operational improvements. The decision-aware scheduling policy reduced average weekly scheduling costs by approximately 38 percent, while also increasing the number of surgeries completed by about 8 percent by reducing deferrals. Importantly, the system only required small adjustments to surgeons’ original time estimates around 35 minutes on average, making it practical for real-world implementation.

Beyond improving efficiency, the framework provides hospital administrators with a flexible “trust parameter” that allows them to control how much the AI model adjusts existing surgeon estimates. This makes the approach more transparent and easier to adopt within existing clinical workflows.

As healthcare systems continue to face rising demand and constrained resources, decision-aware analytics models like this one demonstrate how artificial intelligence and operations research can work together to improve hospital efficiency while helping more patients receive timely care.

 

Authors: Roham Bahri (University of Alberta), Borzou Rostami (University of Alberta), Vahid Roshanaei (University of Toronto)