Individualized No‐Show Predictions: Effect on Clinic Overbooking and Appointment Reminders

Abstract

Patient no‐shows and late cancellations lead to clinic inefficiency, high clinic costs and low patient satisfaction. The two main strategies clinics employed to alleviate the adverse effects of no‐shows are overbooking and patient appointment reminders. Developing effective overbooking schedules depends on accurately predicting each patient’s no‐show probability, while developing effective reminder systems requires a patient‐level estimate of communication sensitivity. Current methods of estimating no‐show probabilities do not produce such patient‐level predictions. To remedy this, we develop a Bayesian nested logit model which utilizes appointment confirmation data and estimates individual‐level coefficients for patient‐specific predictors. A log‐likelihood comparison of model fit on 12 months of appointment data shows that the Bayesian model outperforms the standard logit model by about 30% improvement in model fit. Additionally, our Bayesian model allows categorization of patients based on their appointment confirmation behavior. Finally, using patient‐specific no‐show probabilities as an input to a simulated appointment scheduler we find that the Bayesian model improves clinic profit over the standard logit model. The benefit comes mainly from waiting cost reduction when no‐show probability is low and from physician overtime and idle time cost reduction when no‐show probability is high. Our study has two managerial implications. First, the Bayesian method allows customizing appointment reminder effort based on patient’s confirmation behavior. Second, the Bayesian method also allows improved overbooking scheduling especially in clinics that experience large patient throughput.

Patient No-Show Prediction: A Systematic Literature Review

“…An approach that lies between building a single global model and many local models is to use a mixed effects logistic regression model (MELR). This approach was proposed by Lenzi et al [ 38 ] and Li et al [ 39 ]. In particular, the former, which determines the most parsimonious model based on the Akaike information criterion, groups by patient and provider, while the latter groups by patient and appointment confirmation.…”

“…Only 7 of the 50 analyzed articles include the intra-patient temporal dependence in the model. This dependence was incorporated using different approaches including Markov chains [ 5 ], weighting observations by their temporal closeness [ 18 , 19 ], using an exponential sum for regression [ 28 ], building various LR based on the number of previous visits [ 29 ], or using a MELR [ 38 , 39 ]. The last approach provides a promising approximation in the resolution of the problem since it allows to unify the behavior of the patient, the socio-demographic variables, and the environmental variables.…”

Overbooked and Overlooked: Machine Learning and Racial Bias in Medical Appointment Scheduling

“…The third area of research that we review includes studies that developed methodologies to schedule patients based on their individual no-show probabilities. Li et al (2019), Samorani and LaGanga (2015), Srinivas and Ravindran (2018), Zacharias and Pinedo (2014), and Samorani and Harris (2019) are some examples of a growing body of literature that promotes the predictive overbooking procedure depicted in Figure 1. The goal of those papers is to minimize the schedule cost (typically, the patients' waiting time and the provider's overtime and idle time) using individual no-show probabilities.…”

Coping with no-show behaviour in appointment services: a multistage perspective

“…Previous studies have proposed several coping strategies for no-show behaviour in appointment services, such as reminders, fines, overbooking, open-access scheduling, rescheduling, cancellation policies and changing panel size (Bech, 2005;Huang and Zuniga, 2014;Li et al, 2019;Liu et al, 2019a;Tsai and Teng, 2014). However, consumers' no-show behaviour still frequently arises.…”