A Risk Score Model to Support Early Identification and Prevention of Type 2 Diabetes Incorporating Social Determinants of Health

This study presents the development of a clinically relevant risk score model for Type 2 Diabetes Mellitus (T2DM) by integrating conventional clinical risk factors with social determinants of health (SDOH) using population-level data from the CDC 2015 Behavioral Risk Factor Surveillance System (BRFSS). Multiple machine learning algorithms—including Random Forest, Support Vector Machines (SVM), Logistic Regression, Gradient Boosting, and XGBoost—were applied to a cleaned and curated dataset. The data were split into training (n = 2,700) and testing (n = 300) cohorts, with 10-fold cross-validation performed across all models. K-means clustering was used for exploratory pattern identification, while logistic regression was employed to derive an interpretable risk score for individual predictors. The analysis identified both positive and negative predictors of T2DM. Established clinical factors such as body mass index (BMI), age, and sex emerged as significant positive predictors. Importantly, several social determinants of health were also found to be independently associated with diabetes risk, with higher income levels and improved access to healthcare correlating with reduced predicted risk. The resulting logistic regression–based risk score demonstrated strong predictive performance and offers a novel approach by incorporating underrepresented SDOH variables alongside traditional clinical factors. By leveraging a large, diverse, and nationally representative survey population, this model closely reflects real-world conditions. The findings support the use of integrated risk assessment tools for targeted screening, prevention strategies, and public health initiatives aimed at reducing the burden of Type 2 Diabetes Mellitus.