Data-Driven with Process Mining to Analyze and Identify Key Features for No-Show Risks in Patients with Chronic Diseases to Achieve Treatment Improvement

Kantapat Chaichareenon, Natawut Nupairoj

Abstract: This study addresses the significant challenge of chronic diseases, such as diabetes, which require continuous medical care and follow-up. It aims to visualize the revisit interval that affects patient treatment with multi-level process mining and determine factors that cause patients to miss their appointments or “no-show” by using machine learning. Our research demonstrates that patients who have improvement have significantly shorter revisit intervals compared top atients who are deteriorating in their treatment, with a P-value of 3.83x10−9. A primary challenge is the substantial number of patients failing to attend their appointments. To uncover the reasons behind these no-shows, we employ PyCaret which facilitates the comparison of the performances of 14 models. This approach enables us to determine the importance of various factors contributing to patient no-shows, including our features from multi-level process mining. The result was that patients who usually miss their follow-up schedules have a higher chance of “no-show” at their next appointment. A significant root cause i s “historical attendance of individual patients” which is one of the features from multi-level process mining analysis. Our model has an AUC score of 0.80 in predicting the no-show risk of diabetes patients by using the Gradient Boosting Classifier. O ur study shows that exploring important features can be discovered by the exploration data analysis process. This insight offers a pathway to enhance patient engagement and reduce missed appointments, ultimately aiming to improve care for individuals with diabetes.