Predictive Analytics for Outpatient Appointments

Healthcare is a very important industry where analytics has been applied successfully to generate insights about patients, identify bottleneck and to improve the business efficiency. In this paper, we aim to look at the patient appointment process as the hospital is experiencing high volume of "no shows". "No shows" have a high impact on longer appointment lead time for patients, poor patient satisfaction and loss of revenue for hospital. We use data analytics to identify pattern of "no shows", develop a statistical model to predict the probability of "no shows" and finally operationalizing the model to embed the analytics solution in the business process to reduce the number of "no shows" in the hospital. Exploratory data analysis (EDA) was used to find out the major causes of no shows based on patient demographic information, patient appointment detail and SMS reminder response. Data mining techniques such as logistic regression and recursive partitioning were used on training, test and validation data to predict patients who have high probability of "no show". We present the analytical outcomes and findings from our model. Our logistic regression model could predict around 70% of the "no show" cases correctly with a Kappa coefficient of 0.41 on validation data. Based on our finding, we have recommended different strategies to the operations staff for possible reduction of no show slots.