AI-ENABLED PERSONALISED NUTRITION FOR CARDIOMETABOLIC HEALTH: A SYSTEMATIC REVIEW OF CLINICAL EFFECTS AND GENERATED DIETARY ADVICE

F. KHOZHAMKUL, A. SHARMAN

Vestnik, 2026

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.53065/kaznmu.2026.78.3.004
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Introduction.AI-enabled nutrition includes predictive dietary personalisation, digital lifestyle programmes and generated dietary advice. Aim.To assess clinical outcomes of AI-enabled personalised nutrition in adults with cardiometabolic risk and the accuracy and safety of generated dietary advice. Materials and Methods.Five bibliographic databases and two trial registries were searched from inception to August 2026, with supplementary searches in Semantic Scholar, medRxiv and selected web sources. Eligible interventions were digital tools in which artificial intelligence, machine learning, a large language model or a conversational agent generated individualised dietary, menu or meal recommendations. Clinical outcomes and generated-advice accuracy were synthesised separately. Risk of bias was assessed using RoB 2, ROBINS-I and an adapted QUADAS-2 instrument; certainty was evaluated with GRADE. Results.The final working evidence inventory comprised 62 studies (64 reports): 26 in the clinical domain and 36 in the accuracy domain. A machine-learning-guided diet reduced HbA1c by 0.08 percentage points more than a Mediterranean diet at six months (95% CI, −0.14 to −0.02), whereas a separate personalised-diet trial did not improve six-month weight loss over a low-fat diet (difference +1.05 percentage points of baseline weight; 95% CI, −0.40 to 2.50). Personalised LLM prompting improved weight loss compared with manual ChatGPT prompting, while an AI-based sodium programme did not improve salt intake or blood pressure. A 368-participant trial met it prespecified noninferiority criterion for referral to an AI-led rather than human-coached diabetes-prevention programme. Generated-advice studies reported favourable expert ratings alongside nutrient and guideline discrepancies; complex profiles were more challenging in an exploratory chatbot comparison. Conclusion.Selected interventions improved specific outcomes, but a class-wide benefit remains unestablished. Programme effects do not isolate the AI component, and favourable content ratings do not establish nutritional adequacy or patient safety.