Evaluating diabetes risk: Bayesian hierarchical models and machine learning integration

Type 2 diabetes mellitus (T2DM) is a global health concern influenced by factors such as obesity, sedentary behavior, and poor diet. This study aims to analyze both regional and individual predictors of T2DM in Bangladesh by integrating Bayesian hierarchical models with machine learning methods. This study uses data from the 2017-18 Bangladesh Demographic and Health Survey (BDHS). Employing a Bayesian multinomial mixed-effects model, we account for regional variability and individual factors like age, gender, BMI, residence, wealth, education, employment, and hypertension. Our results indicate significant regional differences and associations between demographic and health-related factors with diabetes risk. Younger individuals and those with a higher BMI are more likely to develop diabetes, and hypertension significantly increases the risk. We applied machine learning (ML) models, including logistic regression, decision trees, k-nearest neighbor, linear discriminant analysis, and random forest, to classify diabetic status using these predictors, assessing their accuracy through 10-fold cross-validation. Logistic regression and linear discriminant analysis demonstrated robust performance across various response distributions. Simulation studies further examined the impact of different response distributions on model performance, revealing significant differences in classification accuracies. This approach of estimating parameters with a Bayesian model, applying ML for prediction, and conducting simulation studies to explore various scenarios highlights the importance of integrating these methodologies for effective diabetes prediction, providing insights for public health strategies to mitigate T2DM's impact.