Machine Learning Overbooking Framework for Outpatient Appointments: Improving Resource Allocation and Correcting Socioeconomic Bias

Abstract

Outpatient care constitutes the primary healthcare service across different countries. Appoint- ment scheduling within this care setting faces the significant challenge of patient no-shows, which is detrimental to service quality, leading to treatment delays and economic losses for healthcare centers. With the rise of artificial intelligence, the combination of strategies such as overbooking and machine learning (ML) models has emerged as a promising approach. However, there are concerns regarding group bias (GB) and its potential to result in unfair services, perpetuating historical barriers and disparities that society is striving to eliminate. In the ML-enabled overbooking framework proposed in this study, we demonstrate the presence of socioeconomic GB due to the under-representation of a socioeconomically vulnerable population within the dataset, and how this worsens the service quality for the vulnerable group. We illustrate how including post-modeling strategies in the two proposed overbooking methodologies can com- pletely mitigate this effect, ensuring fairness in the framework that combines overbooking and ML