BAYESIAN STATISTICAL MODELS FOR PREDICTING TYPE 2 DIABETES PREVALENCE IN URBAN POPULATIONS

Abstract

This systematic review investigates the application of Bayesian statistical models in predicting the prevalence of type 2 diabetes mellitus (T2DM) within urban populations, with a focus on methodological innovations, model performance, data integration, and public health relevance. The study followed the PRISMA guidelines and synthesized findings from 84 peer-reviewed articles published between 2000 and 2025. These studies encompass diverse urban contexts across North America, South Asia, Latin America, East Asia, and sub-Saharan Africa, reflecting a broad and globally relevant evidence base. The review identifies Bayesian hierarchical models as the dominant approach for capturing multilevel dependencies between individuals, neighborhoods, and city-wide determinants. Spatio-temporal Bayesian models were also extensively used to estimate dynamic changes in urban T2DM prevalence, employing structured priors such as Conditional Autoregressive (CAR) models and Gaussian Markov Random Fields (GMRFs). Approximately half of the reviewed studies integrated heterogeneous data sources—including electronic health records (EHRs), satellite imagery, surveys, and census data—through Bayesian data fusion frameworks. These techniques enabled cross-level modeling and imputation of missing data, enhancing robustness and predictive validity. The review also highlights the use of hybrid models such as Bayesian neural networks and ensemble frameworks, which offered improved predictive performance while preserving probabilistic interpretability. Despite these strengths, the review identifies key challenges, including computational burden, sensitivity to prior specification, ethical concerns in spatial labeling, and potential bias in underrepresented urban populations. Comparative evaluations show that while machine learning methods often achieve higher raw accuracy, Bayesian models provide superior interpretability, uncertainty quantification, and policy relevance. The findings affirm that Bayesian modeling offers a statistically rigorous and context-sensitive approach to urban diabetes epidemiology. The study concludes with recommendations emphasizing methodological transparency, ethical safeguards, participatory modeling, and investment in computational capacity to maximize the benefits of Bayesian inference in urban public health decision-making.