Exploratory benchmarking of AI-generated diet plans for inherited protein metabolism disorders: a simulation-based evaluation of nutritional accuracy and clinical safety
Aug 2026· Frontiers in Nutrition· Vol 13· 0 citations· 30 references
TL;DR
It was demonstrated that conventional measures of nutritional adequacy alone were insufficient to determine the clinical appropriateness of AI-generated dietary plans for inherited protein metabolism disorders, and disease-specific benchmarking against established metabolic nutrition guidelines was provided.
Abstract
Artificial intelligence (AI)-based large language models (LLMs) are increasingly used to support nutrition-related decision-making; however, their ability to generate clinically appropriate dietary plans for inherited protein metabolism disorders remains largely unexplored. This study aimed to perform an exploratory simulation-based benchmarking analysis of AI-generated dietary plans for phenylketonuria (PKU), maple syrup urine disease (MSUD), and propionic acidemia (PPA) using disease-specific metabolic nutrition guidelines.
Standardized pediatric case scenarios were developed for PKU, MSUD, and PPA. Using identical English-language prompts, ChatGPT-5.3 Pro and Gemini 3 Pro Advanced each generated 3-day dietary plans. Nutrient composition was analyzed using the BeBiS Nutrition Information System and evaluated against Dietary Reference Intakes (DRIs). Disease-specific nutritional targets, including amino acid intake, protein distribution, and energy provision, were benchmarked against recommendations from Genetic Metabolic Dietitians International (GMDI). Nutritional characteristics of the dietary plans generated by the two AI models were compared using exploratory statistical analyses.
Both LLMs generated structured dietary plans with generally acceptable overall nutritional characteristics; however, clinically relevant deviations from disease-specific nutritional targets were identified across all three disorders. In the PKU case, both models achieved the recommended phenylalanine range, but neither simultaneously met protein and tyrosine recommendations. In the MSUD case, differences were primarily related to energy provision and branched-chain amino acid targets, while in the PPA case neither model achieved the recommended balance between intact protein and total protein. These findings demonstrated that conventional measures of nutritional adequacy alone were insufficient to determine the clinical appropriateness of AI-generated dietary plans for inherited protein metabolism disorders.
General-purpose LLMs can generate structured dietary plans for inherited protein metabolism disorders; however, disease-specific metabolic targets are not consistently achieved. Evaluation of AI-generated dietary plans should therefore extend beyond conventional nutritional assessment and incorporate disease-specific benchmarking against established metabolic nutrition guidelines. This study provides a disease-specific benchmarking framework for evaluating AI-generated dietary plans in inherited protein metabolism disorders.
BACKGROUND
Personalized meal planning by registered dietitian nutritionists (RDNs) is time-intensive. Large language models (LLMs) may automate drafting meal plans, but their nutritional accuracy in clinical practice is uncertain.
METHODS
In this proof-of-concept study, five outpatient RDNs and four LLMs (Gemini, CoPilot, ChatGPT 4.0, and customized ChatGPT 4.0) each generated 3-day meal plans for five validated clinical scenarios. Effectiveness was defined as accuracy in meeting pre-specified energy, protein, carbohydrate, fat, and sodium targets. Time to create plans and RDN comfort (self-rated confidence in nutritional accuracy and clinical appropriateness on 1-5 Likert scale) were recorded. Three independent RDNs, blinded to source, analyzed nutrient content using Nutritionist Pro. Group differences were assessed with t-test and ANOVA.
RESULTS
All LLMs and RDNs produced feasible meal plans. LLMs generated meal plans in under 1 min, whereas RDNs required a mean of 44 min per scenario. RDNs reported comfort levels ranging from 3.8 to 4.8. Across most scenarios, LLM plans delivered a smaller proportion of requested energy than RDN plans, which more consistently approached energy targets. Both groups performed similarly for the Mediterranean diet scenario. Overall, protein accuracy did not differ. However, in chronic kidney disease, LLMs undershot the guideline-based protein target, while RDNs tended to modestly exceed it. Accuracy for low-carbohydrate, fat, and sodium diets was comparable.
CONCLUSION
LLMs can rapidly generate clinically plausible meal plans but are less reliable than RDNs in achieving prescribed energy and selected macronutrient goals. Prompt precision is essential for nutrient-specific targets. A hybrid model in which RDNs refine LLM-generated drafts may leverage efficiency without sacrificing clinical accuracy.
M. Mundi, Osman Mohamed Elfadil, Danielle P. Johnson et al.· Nutrition in clinical practi...· 0 citations
The findings support the technical feasibility of the MAI-DIET framework and its preliminary acceptability among healthcare professionals, and further evaluation is required before conclusions can be drawn regarding its practical effectiveness.
Konstantinos Divanis, Alexandra Foscolou, Georgios Prosalentis et al.· Nutrients· 0 citations
Nutrition is one of the few modifiable risk factors for health status. The diet-disease relationship, however, is highly complex, with many endogenous and exogenous factors contributing to risk and with variation in responses to diet between individuals. The 2020-30 National Institutes of Health (NIH) Strategic Plan for Nutrition Research states that precision nutrition is a unifying and holistic approach to developing comprehensive and dynamic nutritional recommendations relevant to both individual and population health. Recognizing the growing importance of advanced computational approaches in nutrition research, in 2022 the NIH Office of Nutrition Research and participating NIH Institutes and Centers published a Funding Opportunity Announcement for training programs in AI for Precision Nutrition (AIPrN) aiming to develop a new generation of the scientific workforce, literate in both nutritional and computer sciences, that leverages the growing data resources available for nutrition research to tackle complex and emerging health-disease associations. In this paper, we describe the approaches developed by the first four institutional awardees of the T32 AIPrN training grant, and we highlight applied strategies for interdisciplinary training, data integration, and methodological innovation as the research community prepares to analyze and interpret data generated by major initiatives such as the NIH-funded Nutrition for Precision Health Initiative.
Saurabh Mehta, S. Huey, Paraskevi Massara et al.· Journal of NutriLife· 0 citations
Objective: This study aimed to evaluate the clinical accuracy and degree of individualization of diets generated by ChatGPT-3.5 for malnourished simulated cases with decompensated cirrhosis. A secondary aim was to quantitatively compare the nutrient composition of diets generated by ChatGPT with individualized medical nutrition therapy plans prepared by a dietitian.Methods: Ten simulated case profiles (five women and five men) representing malnourished individuals with decompensated cirrhosis of different ages and body mass indices were developed. The same clinical and demographic information was provided to both ChatGPT and the same registered dietitian for each simulated case to ensure comparability. ChatGPT and the dietitian independently generated individualized dietary plans and nutrition therapy recommendations. The nutrient composition of all diets was analyzed using the BeBiS nutrition software. Differences between the two approaches were evaluated using the Wilcoxon signed-rank test, with statistical significance set at p < 0.05.Results: Compared with dietitian-generated diets, ChatGPT-generated diets contained significantly higher amounts of protein and a higher percentage of energy derived from protein, whereas carbohydrate content, the percentage of energy from carbohydrates, dietary fiber, plant protein, and potassium were significantly lower, whereas sodium content was significantly higher in the ChatGPT-generated diets (p0.05). Furthermore, seven of the ten simulated cases received identical diets despite having different nutritional characteristics, indicating insufficient dietary individualization.Conclusion: Although ChatGPT can provide general nutrition guidance, it demonstrated important limitations in generating individualized medical nutrition therapy for malnourished simulated cases with decompensated cirrhosis. AI-generated dietary plans should therefore be reviewed and individualized by a registered dietitian before clinical implementation.
Hilal Toklu Baloğlu· MIDDLE BLACK SEA JOURNAL OF...· 0 citations
BACKGROUND
ecombinant human growth hormone (rhGH) has been used for four decades under a largely population-based dosing paradigm, yet growth response varies substantially among children with apparently similar auxological phenotypes. Advances in genomics, GH-insulin-like growth factor-1 (IGF-1) axis biomarkers, mathematical and machine-learning (ML) prediction models, and digital health technologies now allow individualized characterization of growth disorders, forming the basis of an emerging precision-endocrinology paradigm.
OBJECTIVES
(1) to synthesize evidence on monogenic and polygenic genetic determinants of pediatric growth faltering and/or short stature and their diagnostic yield; (2) to evaluate GH-IGF-1 axis biomarkers and pharmacogenetic markers, including the IGF-1/IGFBP-3 molar ratio as a likely better indicator of bioactive IGF-1 than IGF-1 alone, together with mathematical/ML prediction models, for individualizing rhGH therapy; and (3) to appraise artificial intelligence (AI) and digital-health tools, bone-age algorithms, facial-recognition phenotyping, adherence-prediction models, and smartphone growth-monitoring, as instruments for operationalizing precision endocrinology in children and adolescents with growth disorders.
METHODS
This is a systematic review employing narrative synthesis (a systematic scoping review). Reporting explicitly followed the PRISMA Extension for Scoping Reviews (PRISMA-ScR) checklist rather than the PRISMA 2020 statement for systematic reviews and meta-analyses, because the heterogeneous outcome metrics across genetic, diagnostic-accuracy, prediction-model, and AI/digital-health studies preclude meta-analytic pooling of a single quantitative effect size; PRISMA-ScR is the methodologically appropriate reporting framework for a review mapping evidence across such conceptually distinct domains. PubMed/MEDLINE was searched for English-language, pediatric (0-18 years) studies published between 2000 and 2026. Two-stage screening (title/abstract, then full text) was performed. Quality was appraised using design-appropriate tools: an adapted Newcastle-Ottawa Scale for genetic-association studies, QUADAS-2 for diagnostic-accuracy biomarker studies, PROBAST/TRIPOD-informed criteria for prediction-model and ML studies, and reference-standard/external-validation criteria for AI-imaging studies. Sixty-two studies were retained for qualitative synthesis.
RESULTS
Monogenic defects (SHOX, ACAN, NPR2) and exome-sequencing panels explain a meaningful minority (approximately one-quarter) of previously "idiopathic" short stature, while genome-wide association studies and polygenic scores capture a substantial share of the remaining heritable variance, with polygenic risk scores achieving areas under the receiver-operating-characteristic curve up to 0.84 for predicting adult short stature. The IGF-1/IGFBP-3 M ratio outperforms IGF-1 alone for diagnosing GH deficiency (sensitivity 87.5%, specificity 83.0%), reflecting the greater bioavailability of free, unbound IGF-1 relative to that carried in the ternary IGF-1/IGFBP-3/acid-labile-subunit complex. GH-receptor exon-3 (d3) pharmacogenetic variants and machine-learning models (random forest, transcriptomic classifiers) improve prediction of individual rhGH response beyond classical mathematical models. AI-based bone-age algorithms achieve near-radiologist accuracy with reduced inter-observer variability, computer-aided facial-phenotyping tools show comparable diagnostic accuracy for syndromic short-stature disorders such as Noonan and Turner syndrome, and connected-device/ML adherence-monitoring and smartphone growth-tracking tools objectively detect suboptimal adherence and growth faltering earlier than conventional clinic-based surveillance; network meta-analyses of once-weekly long-acting rhGH formulations further suggest that reduced injection burden can translate into modestly improved height outcomes relative to daily rhGH. These findings are synthesized into a Precision-Medicine Cascade, a practice-oriented framework showing how genotype, biomarker, and digital data streams can be layered onto routine auxological assessment to guide same-visit clinical decisions on diagnostic work-up, dosing, and monitoring frequency.
CONCLUSION
Converging genetic, biomarker, computational, and digital-health evidence supports a feasible, evidence-grounded trajectory toward individualized therapy, including rhGH and emerging growth-plate-targeted agents, in pediatric growth disorders. The Precision-Medicine Cascade proposed here offers pediatric endocrinologists an immediately applicable framework for integrating these tools into everyday practice. However, current tools remain adjunctive rather than replacement for clinical judgment, and prospective, ethnically diverse validation of integrated precision-endocrinology pathways is required before routine adoption.
Ashraf T. Soliman, F. Alyafei, Nada Alaaraj et al.· Growth Hormone & IGF Researc...· 0 citations
