Journal of Regional Medical Campuses

Machine Learning to Improve Resident Scheduling: Harnessing Artificial Intelligence to Enhance Resident Wellness

Aazad Abbas

Division of Orthopaedic Surgery, University of Toronto, Toronto, Canada

Jay Toor

Division of Orthopaedic Surgery, Max Rady College of Medicine, University of Manitoba

Denis Ariel Margalik

Temerty Faculty of Medicine, University of Toronto, Toronto, Canada

Jin Tong Du

Division of Orthopaedic Surgery, University of Toronto, Toronto, Canada

Anne Versteeg

Division of Orthopaedic Surgery, University of Toronto, Toronto, Canada

Nicholas J Yee

Division of Orthopaedic Surgery, University of Toronto, Toronto, Canada

Joel A Finkelstein

Division of Orthopaedic Surgery, University of Toronto, Toronto, Canada

Jihad Abouali

Division of Orthopaedic Surgery, Michael Garron Hospital, Toronto, Canada

Markku T Nousiainen

Division of Orthopaedic Surgery, Sunnybrook Health Science Centre, Toronto, Canada

Hans J Kreder

Division of Orthopaedic Surgery, Sunnybrook Health Science Centre, Toronto, Canada

Jeremy Hall

Division of Orthopedic Surgery, St Michaels Hospital, Toronto, Canada

Cari Whyne

Sunnybrook Research Institute, Orthopaedics Biomechanics Laboratory

Jeremie Larouche

Division of Orthopaedic Surgery, Sunnybrook Health Science Centre, Toronto, Canada

DOI: https://doi.org/10.24926/jrmc.v8i1.6332

Keywords: resident, medication education, orthopaedics, machine learning, scheduling

---

Abstract

Introduction: Excessive resident duty hours (RDH) is a recognized issue with implications for physician well-being and patient safety. A significant component of the RDH concern is on-call duty. While other industries have adopted machine learning models (MLMs) to optimize scheduling and employee well-being, medicine has lagged. This study aimed to investigate the use of MLMs to predict demand on orthopaedic residents to optimize scheduling.

Methods: Daily surgical handover emails over an eight-year (2012-2019) period at a level I trauma centre were used to model demand on residents. Various MLMs were trained to predict the workload, with their results compared to the current approach. Quality of models was determined by using the area under the receiver operator curve (AUC) and accuracy. The top ten most important variables were extracted from the most successful model.

Results: The reduction in orthopaedic resident shifts possible per annum was 24.7%. The most successful model during testing was the neural network (AUC: 0.81, accuracy: 73.7%). All models were better than the current approach (AUC: 0.50, accuracy: 50.1%). Key variables used by the neural network model were (descending order): spine call duty (y/n), year, weekday/weekend, month, and day of the week.

Conclusion: This was the first study using MLMs to predict demand for orthopaedic residents at a major academic institution. All MLMs were more successful than the current scheduling approach. Future work should look to incorporate predictive models with optimization strategies, matching scheduling with demand to improve resident well-being and patient care.