Paper ID
1934
Paper Type
full
Description
Appointment no-shows are common in outpatient clinics and increase clinic costs and patients’ dissatisfaction. We develop a framework to predict the no-show probabilities of a given set of patients, and to subsequently employ these predictions to find the optimal appointment schedule. Some existing work assumes that all patients have the same no-show probability (1-class approach); other work assumes that patients have either a low or a high no-show probability (2-class approach). In contrast, we utilize probabilistic classifiers to obtain the individual patients’ no-show probabilities (N-class approach). Our approach results in better-quality schedules, as measured by a weighted average of patient waiting time and provider overtime. We also find that a small increase in the prediction performance (measured by the Brier score) translates into a large decrease in the schedule cost. Our results are obtained through a large-scale computational study and validated on a real-world data set from an outpatient clinic.
Recommended Citation
Samorani, Michele and Harris, Shannon, "The Impact of Probabilistic Classifiers on Appointment Scheduling with No-Shows" (2019). ICIS 2019 Proceedings. 7.
https://aisel.aisnet.org/icis2019/is_health/is_health/7
The Impact of Probabilistic Classifiers on Appointment Scheduling with No-Shows
Appointment no-shows are common in outpatient clinics and increase clinic costs and patients’ dissatisfaction. We develop a framework to predict the no-show probabilities of a given set of patients, and to subsequently employ these predictions to find the optimal appointment schedule. Some existing work assumes that all patients have the same no-show probability (1-class approach); other work assumes that patients have either a low or a high no-show probability (2-class approach). In contrast, we utilize probabilistic classifiers to obtain the individual patients’ no-show probabilities (N-class approach). Our approach results in better-quality schedules, as measured by a weighted average of patient waiting time and provider overtime. We also find that a small increase in the prediction performance (measured by the Brier score) translates into a large decrease in the schedule cost. Our results are obtained through a large-scale computational study and validated on a real-world data set from an outpatient clinic.