Detecting Diabetic Autonomic Neuropathy from Electronic Health Records Using Machine Learning

Diabetes is a disease that affects a large number of people worldwide, and diabetic neuropathy is one of its most common and serious complications. Diabetic autonomic neuropathy (DAN) is a type of diabetic neuropathy that is defined as a disorder of the autonomous nervous system and can affect various organs in the body, including the heart and kidney. DAN is widely under-diagnosed due to reasons such as the cost and unavailability of testing equipment, the difficulty of performing cardiovascular tests, and the oftentimes asymptomatic state of the disease in its early stages. However, a late diagnosis can lead to dangerous health complications in the long run. As such, this paper aims to use machine learning to detect DAN in the kidney and heart in diabetic patients by retrieving their information from electronic health records. For this purpose, a dataset of 1275 patient records was used with a variety of traditional machine learning and deep learning algorithms. The best performing model was TabNet with an F1 score of 85.82 for the heart and 73.37 for the kidney.