Predictors of Healthcare Utilization in Patients with Diabetes: Comparing Traditional Statistical Methods to Supervised Machine Learning Approaches

Population health decision makers are interested in understanding patient characteristics associated with higher levels of healthcare utilization, particularly among patients with chronic health conditions. A variety of methodological approaches exist to identify such characteristics, including traditional biostatistical methods and machine learning methods. Understanding how these approaches compare, their limitations, and how results may vary across approaches is important for understanding which methods are fit for purpose. This project used methodological approaches from traditional statistics and supervised machine learning on a claims dataset to understand the different approaches and their results in identifying predictors of high healthcare utilization among a diabetic population