Drafting a residency call schedule is time consuming due to variables such as resident physician contract stipulations, practice conventions within a residency program, resident wellness and personal scheduling requests. Resident call duties have been a major topic of discussion in the past few decades and the evidence support the notion that unreasonable call requirements can have significant impacts on both resident performance and resident burnout. Therefore, generating call schedules is a vital task undertaken with careful attention. Scheduling has traditionally implemented linear programming which is less flexible when compared to Artificial Intelligence (AI). We have applied constraint programming (Google’s OR Tools), a subdomain of AI, to build a pragmatic call scheduling algorithm for the University of Saskatchewan’s Core Internal Medicine Program and compare its efficacy to traditional methods.
The newly implemented AI algorithm would select a single call schedule from >1 million potential call schedules in 10 minutes or less based on fixed and flexible constraints. When compared head to head, the new AI algorithm performed significantly better with respect to scheduling errors by virtually eliminating errors. It was also demonstrated to be statistically equal in all other metrics assessing scheduling quality as determined by an anonymous resident survey, namely minimizing resident workload, reducing fly-in shifts, and maximizing desirable weekend call combinations. Most importantly, with automation, the scheduler was immune to last minute changes.
This quality improvement project is a demonstration of the potential and role of AI beyond the clinical domain. With AI becoming more accessible, we implore other residency programs to entertain the creation of innovation-committees to situate resident physicians at the forefront of AI to guide the advancement of medicine in all domains, including education and administration.