Leveraging on Predictive Analytics to Manage Clinic No-Show and Improve Accessibility of Care

Clinic No-Shows result in inefficient use of precious hospital resources as it deprives needy patients of timely appointments. Accurately predicting a patient's risk of clinic No Show enables clinic managers to intervene with appointment reminders, optimize allocation of staff, balance clinic workload and use intelligent overbooking techniques to reduce appointment waiting time. While previous work have identified significant factors associated with clinic No-Show, little work have showcased the real life deployment and application of predictive modelling of No-Shows. In this paper, we present a feature engineering approach with text mining, the modelling of our No-Show predictive model, the insights gained, and discuss modeling considerations on the trade-offs between model accuracy and complexity of deployment. We also present our deployment architecture and demonstrate how the risk scores produced by the model were being used as a business intelligence report in real clinic settings for intervention. Finally, through case studies, we showcase that our model could be used to manage No-Shows and improve accessibility of care.