Keywords : Type 2 Diabetes Risk Prediction, Social Determinants of Health (SDOH) and Machine Learning

Abstract : Aims: To develop a clinically relevant risk score model for Type 2 Diabetes Mellitus (T2DM) by integrating conventional risk factors with social determinants of health (SDOH) using population-level data from the CDC 2015 Behavioral Risk Factor Surveillance System (BRFSS). Methods: Random Forest, Support vector machines (SVM), Logistic Regression, Gradient Boosting, and XGBoost were applied to the cleaned and curated CDC 2015 BRFSS dataset. The dataset was divided into a training dataset (n=2700) and a testing dataset (n=300). 10-fold cross-validation on all algorithms and clustering using K-means algorithm were conducted for data analysis. Logistic regression was used to generate the risk score for each factor. Results: The study identified positive and negative predictors of T2DM. The significant positive predictors included well-established clinical factors like BMI, Age, and sex. Emerging risk factors for T2DM, referred to as social determinants of health (SDOH), were also found to be significant in this study. For instance, higher income and improved access to healthcare were associated with lower predicted risk. Conclusion: These findings support targeted screening and preventive interventions for populations at elevated risk of Type 2 Diabetes. The logistic regression-based risk score model has great predictive power and is unique in that it utilizes underrepresented features with promising potential like the SDOH along with conventional clinical risk factors. The risk score relevance in public health initiatives to halt diabetes is based on its use of a diverse survey population that simulates real-world scenarios. The risk score model thus has great real-world applicability and is appropriate for public health initiatives.

Download