Abstract

Artificial intelligence (AI) and big data are reshaping how we understand, predict, and manage 
obesity-associated type 2 diabetes (T2D). The diabesity phenotype arises from heterogeneous interactions 
among genes, behaviors, environments, and health-care systems. At scale, routinely collected data electronic 
health records (EHRs), pharmacy claims, continuous glucose monitoring (CGM), wearables, meal logs, imaging, 
multi-omics, and social determinants of health (SDOH) capture this complexity but are noisy, incomplete, and 
biased. Modern machine-learning (ML) methods can transform these substrates into actionable insights: 
predicting incident T2D and complications; detecting subclinical trajectories; stratifying patients into 
mechanistic endotypes; recommending individualized nutrition, activity, and pharmacotherapy; and monitoring 
for relapse or adverse events. Time-series deep learning, survival modeling, graph neural networks, and causal 
inference frameworks enable robust forecasting and counterfactual reasoning, while reinforcement learning (RL) 
personalizes dynamic regimens. However, translation hinges on trustworthy data engineering, external 
validation, calibration, explainability, privacy, and equity. Federated learning and differential privacy protect 
data; fairness auditing and participatory design mitigate bias; and MLOps governs monitoring, drift detection, 
and post-deployment updates. Integrating AI into clinical workflows requires human-in-the-loop decision 
support, interoperable standards, and pragmatic evaluation focused on outcomes that matter to patients and 
systems. This review synthesizes the data foundations, predictive analytics, digital phenotyping, and 
decision-support paradigms relevant to diabesity; outlines implementation, safety, and governance 
requirements; and maps a path toward multimodal, foundation-model–enabled “digital twins” that couple 
physiology with behavior to modify disease trajectories. Done well, AI augments not replaces clinicians and 
patients, enabling earlier intervention, precise therapy matching, and durable cardiometabolic risk reduction.