Artificial Intelligence in the Optimisation of Glucagon-Like Peptide-1 Receptor agonist Therapy for Obesity and Type 2 Diabetes: A Critical Narrative Review of Treatment Monitoring and Clinical Decision Support

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Glucagon-like peptide-1 receptor agonists have changed what can be achieved pharmacologically in obesity and type 2 diabetes, yet the average effects reported in registration trials conceal wide interindividual variation in weight and glycaemic response, high rates of early discontinuation, incomplete dose escalation in routine care, and pronounced inequity in who receives treatment. These features have prompted the claim that artificial intelligence, and machine learning in particular, can improve candidate selection, treatment monitoring and prescribing decisions. This critical narrative review evaluates whether the accessible evidence supports that claim. Literature was identified through structured searching of a biomedical bibliographic index and a scholarly metadata registry, supplemented by backward citation searching and verification against publisher records. Five interrelated domains were examined: prediction of response and identification of candidates; monitoring through continuous physiological, behavioural and imaging data streams; decision support ranging from registry-derived rule induction to large language models; automated safety signal detection; and the methodological, regulatory and equity conditions under which such systems would have to operate. The evidence base is expanding quickly but remains immature. Most predictive models are developed retrospectively in single data environments, evaluated by discrimination alone, and reported without calibration, decision-curve analysis or external validation. Several report performance figures that appear strong but depend on outcome definitions dominated by baseline values, so the apparent accuracy reflects arithmetic rather than clinical insight. Monitoring applications and digital engagement platforms show consistent associations with better outcomes, but the observational designs cannot separate the effect of the technology from the characteristics of people who engage with it. No randomised evaluation has yet demonstrated that an artificial intelligence system improves patient-relevant outcomes specifically in glucagon-like peptide-1 receptor agonist therapy. Prospective, calibration-aware, externally validated and equity-audited evaluation is the principal requirement before deployment can be justified.