Type 2 diabetes mellitus (T2DM) represents a major global health challenge, necessitating robust strategies for early detection and intervention. Machine learning has emerged as a powerful approach for enhancing clinical prediction of T2DM, employing a spectrum of algorithms from traditional logistic regression and support vector machines to advanced ensemble methods and deep learning architectures. This systematic review explores the integration of diverse multimodal dataincluding clinical measures (e.g., blood glucose, BMI, blood pressure), electronic health records (EHRs), genomic and proteomic profiles, and lifestyle indicatorsto improve predictive accuracy. Despite promising results, critical challenges remain, such as data quality issues (missing values, class imbalance, and privacy concerns), model interpretability, and limited generalizability across populations. Future research should prioritize the development of interpretable, fair, and clinically adaptable machine learning systems. Leveraging time-series data and novel AI techniques such as federated learning could further refine risk stratification and support the translation of predictive models into real-world clinical settings, ultimately contributing to personalized prevention and improved patient outcomes.
Bai et al. (Thu,) studied this question.