Efeitos de intervenções nutricionais guiadas por IA no peso corporal: uma revisão sistemática e meta-análise

Abstract

Artificial intelligence (AI) evolves rapidly, yet previous systematic reviews evaluated obsolete computational tools, necessitating an updated analysis of modern generative AI and chatbots in nutrition. The objective of this study was to evaluate the applicability and effectiveness of AI models in dietary planning, monitoring, adherence, and body weight reduction. Following Cochrane and PRISMA guidelines (registered on OSF: osf.io/d952s), electronic databases (PubMed, Cochrane Library, IEEE Xplore, Web of Science, Google Scholar, OATD) were searched up to April 2026 for randomized and non-randomized interventional trials evaluating AI tools versus controls in humans aged 19 to 60 years. Two independent reviewers performed study selection and risk of bias assessment using RoB 2 and ROBINS-I. Quantitative random-effects metaanalysis was executed using Jamovi software. Thirteen studies (k=13, n=2,535) met all inclusion criteria. Overall pooling demonstrated a statistically significant weight reduction favoring AI interventions (µ = -0.35; 95% CI: -0.54 to -0.16; p < 0.001), translating to a mean additional reduction of 2.39 kg, though heterogeneity was high (I² = 77.87%). Non-randomized studies (k=6) consistently favored interventions (µ = - 0.37; p < 0.001), retaining significance after sensitivity outlier removal (µ = -0.25; p < 0.001; I² = 0%). In randomized controlled trials (k=7), the initial pooled effect (µ = -0.35; p = 0.04) shifted upon removing an influential outlier, assuming a borderline statistical character (µ = -0.21; 95% CI: -0.44 to 0.02; p = 0.071; I² = 60.89%). Risk of bias ranged from low to moderate across most trials. Limitations include a modest number of included studies, technological intervention variability, and short follow-up durations (median: 12 weeks). AI models demonstrate promising clinical potential for weight management, but require professional oversight within a Human-in-the-Loop framework to ensure nutritional safety. There wasn't any funding for this project.