Overbooked and Overlooked: Machine Learning and Racial Bias in Medical Appointment Scheduling
Michele Samorani, Leavey School of Business, Santa Clara University
Shannon L. Harris, Virginia Commonwealth University
Linda Goler Blount, Black Women’s Health Imperative
Haibing Lu, Santa Clara University
Michael A. Santoro, Santa Clara University
In using machine learning for appointment scheduling, the challenge arises when certain racial groups have a higher chance of missing appointments. This can lead to these patients being scheduled during busy times, impacting their service experience negatively. To address this, researchers proposed new ways of scheduling that balance the cost and avoid racial disparities. Their approach, tested on real-world data, shows that common scheduling systems can make black patients wait about 30% longer.
Recognized as the winner of the 2023 Manufacturing & Service Operations Management (M&SOM) Best Paper Award, their work introduces a race-aware method that achieves fairness and emphasizes the importance of considering race in scheduling objectives.
Let’s explore further with a brief Q&A session with a member of the research team, Associate Professor of Supply Chain Management and Analytics, Shannon Harris, Ph.D.:
Can you describe the challenges healthcare facilities might face in managing racial disparities in patient scheduling and why a race-aware method would be preferable?
Managing any disparities in a healthcare clinic, once they become known, can be challenging. Addressing the issues and correcting them involves acknowledging how the inequities occurred in the first place, and releveling expectations with all patients and staff involved with the clinic.
In our paper, we found that an approach that explicitly addresses the bias that is uncovered is best. Other approaches either resulted in a decreased amount of fairness or increased cost, a metric that was important in our model. The hope is that an explicit approach is a way to begin to address the problem, and with continued evaluation and consideration, clinic procedure and culture can be updated as needed.
How do you foresee other healthcare settings adopting the race-aware method, and are there potential challenges organizations might encounter when implementing?
Our method is “race”-aware, because based upon initial analyses, we found differences in waiting times between black and white patients. We mitigated that imbalance by minimizing the waiting time of the racial group expected to wait longer. A clinic that would like to understand if their procedures are leading to unintended bias can analyze their policies, not only based upon race, but upon any demographic or socio-economic factors of their patients. For example, if it is found that there are differences in a clinic metric based upon sex, then the clinic can mitigate the inequity by adopting a “sex”-aware” approach, where the metric is balanced among the sexes in the clinic.
Change can be difficult! A clinic may face many challenges when trying to implement a method to address bias. Some patients may question the need for change; others may take offense that the bias occurred, even if it was unintentional. It is important for patients to trust their healthcare providers, so a clinic must work to rebuild broken trust. Some bias may have been perpetuated by clinic staff, and the clinic will face challenges around realigning how staff interact with patients, and ensuring the culture of the clinic is one where patients feel cared for and safe.
Read the full research paper: https://pubsonline.informs.org/doi/10.1287/msom.2021.0999