Abstract
•Accurately predicting patient service time will lead to more effective scheduling.
•We employ an analytical approach, with a data-driven methodology.
•Machine learning and simulation modeling evaluate common policies in practice.
•This study supports the understanding of factors that impact service time.
•We provide clinical policy implications and recommendations for practitioners.
Most outpatient clinics apply deterministic block scheduling policies to patient visits even though patients utilize varying amounts of time, leaving patients, operations managers, and clinicians frustrated because patients and physicians are kept waiting. This paper offers a decision-making model for schedulers so that the service time needed for a specific patient can be predicted to allow outpatient clinics to schedule more effectively. We employed an analytical approach, with a data driven methodology consisting of two phases. In phase one, machine learning algorithms are used to predict service time for outpatient clinics servicing patients with various characteristics. This study supports the understanding of factors that impact service time. A large dataset from an outpatient clinic is obtained and used in the analyses. Four dominant data mining models are developed to predict service time, and their performances are compared: neural networks (NNs), generalized linear model (GLM), linear regression (LR), and support vector regression (SVM). The NN models performed the best. The reason for visiting the doctor and patient type are identified as the primary characteristics to aid in predicting patient service time. We compare the proposed NN models with commonly used scheduling policies in practice in the second phase via simulation modeling and analysis. This paper contributes to the literature in four ways. First, we obtained a large dataset and extracted quality data to test the prediction accuracy of multiple models to determine which one improves scheduling the best. Second, patient characteristics are identified through machine learning modeling and sensitivity analysis to understand which ones are most important for service time prediction accuracy. Third, we analyzed the performance of standard scheduling policies used in clinics. Lastly, we provide clinical policy implications and recommendations that will provide insights and support appointment scheduling decisions.