Classification and Prediction of Hypoglycemia in Patients with Type 2 Diabetes Mellitus Using Data from the EHR and Patient Context

Luis Claudio Gubert, Felipe André Zeiser, Cristiano André da Costa, Rafael Kunst

Abstract: The increase in obesity, a sedentary lifestyle, and population aging are considered the main factors for the increase in Type 2 Diabetes Mellitus (T2DM) worldwide. Global estimates indicate that around 400 million people live with T2DM, reaching 600 million in 2035. This scenario generates a high social and financial cost for the patient and the healthcare system. In this context, this work evaluates machine learning models to classify and predict hypoglycemic crises in patients with T2DM. A dataset with data from a clinical center in southern Brazil is constructed. Patient data involves Electronic Health Records (EHR) and data collected in the patient context through Internet of Things (IoT). This dataset is used to run classification and prediction models. Results show that the proposed approach is promising, achieving an AUC of 0.8200 and a sensitivity of 90.00% for classifying hypoglycemia. In addition, the Clarke Error Grid plot demonstrates an assertiveness of prediction for hi