Predicting diabetes self-management education engagement: machine learning algorithms and models.

Abstract

Introduction: Diabetes self-management education (DSME) is endorsed by the American Diabetes Association (ADA) as an essential component of diabetes management. However, the utilization of DSME remains limited in the USA. This study aimed to investigate current DSME participation among the older population and to identify comprehensive factors of DSME engagement through employing various machine learning (ML) models based on a US nationally representative survey linked to claims data.

Research design and methods: Data from the Medicare Current Beneficiary Survey were employed, and this study included data on US Medicare beneficiaries with diabetes from 2017 to 2019. Comprehensive variables following the National Institute on Aging Health Disparities Research Framework were employed to ensure a comprehensive evaluation of factors associated with DSME using five common ML approaches.

Results: In our study, 37.94% of participants received DSME after the application of inclusion and exclusion criteria. A total of 95 variables were used and all ML models achieved accuracy scores exceeding 70%. Random forest had better predictive performance, with an accuracy of 85%. Seventy-four of 95 variables were identified as key variables. Racial/ethnic disparities in predictors for DSME were identified in this study.

Conclusions: This study identified comprehensive and critical factors associated with DSME engagement from biological, behavioral, sociocultural, and environmental domains using different ML models, as well as related racial/ethnic disparities. Aligning these findings with the DSME National Standards from the ADA would enhance the guidelines' effectiveness, promoting tailored and equal diabetes management approaches that cater to diverse races/ethnicities.