Ensemble Learning for Addressing Class Imbalance in Cardiology Appointment Scheduling and Overbooking

Abstract Purpose Outpatient appointment scheduling is a critical aspect of healthcare services; however, the unpredictable nature of patient behavior poses challenges. This study focuses on predicting patient behavior in cardiology appointment scheduling in an outpatient cardiology practice within the Mount Sinai Health System (MSHS) to optimize overbooking strategies. Methods By reviewing the literature and conducting exploratory data analysis, significant features influencing patient behavior were identified. An ensemble learning model for an imbalanced class was developed to accurately predict the likelihood of no-show appointments and enable strategic overbooking decisions. Results The findings demonstrate that an increased lead time is associated with a greater probability of appointment no-shows. Patient-initiated rescheduling and cancellations were identified as the primary reasons for appointment changes. Moreover, while average wait time increases with age, younger individuals tend to have longer appointments. A stacking ensemble model for imbalanced classes and three machine-learning approaches were evaluated. The stacking ensemble model outperformed traditional techniques with an impressive F1 score of 92.3% and an AUC of 91%. Conclusion The proposed model enables allocating appointments based on patient preferences and characteristics, optimizing resource use, and reducing the number of cancellations and no-shows.