Artificial intelligence for precision obesity phenotyping and cardiometabolic risk stratification: A critical narrative review

Background: Body mass index is a widely used obesity metric but does not adequately capture adiposity distribution, metabolic dysfunction, or individual cardiometabolic risk, leading to misclassification of conditions such as normal-weight obesity, metabolically healthy obesity, and excess visceral fat. With the rising global burden of obesity and related cardiovascular and metabolic diseases, more precise and biologically informative risk-stratification approaches are urgently needed. Aim: This structured narrative review critically examines whether AI-enabled approaches provide incremental value beyond established anthropometric, body-composition, metabolic, imaging, and clinical phenotyping for cardiometabolic risk stratification in adults with obesity. Methods: We conducted a structured narrative review of peer-reviewed evidence identified through searches of PubMed/MEDLINE, Scopus, Web of Science, and supplementary sources through 22 July 2026. Evidence was synthesized thematically across imaging, electronic health records, laboratory data, omics, wearables, cardiometabolic risk prediction, and clinical implementation. Results: Artificial intelligence-enabled methods can identify complex associations and automate aspects of phenotype extraction, but evidence that they provide clinically important improvement beyond established anthropometric, body-composition, metabolic, and risk-assessment approaches remains limited. Most studies are retrospective, and reported gains in discrimination are often modest or insufficiently evaluated for calibration, external validity, decision impact, and patient outcomes. Conclusion: AI may complement established obesity phenotyping, but current evidence supports analytical feasibility more strongly than clinical superiority. Adoption should depend on demonstrated incremental value over conventional assessment, external validation, calibration, fairness, interpretability, workflow compatibility, and prospective evidence of improved decisions or outcomes.