PT59. - ARTIFICIAL INTELLIGENCE (AI) IN NUTRITION: A CASE-BASED COMPARISON OF GENERATIVE AI MODELS

PT59.

ARTIFICIAL INTELLIGENCE (AI) IN NUTRITION: A CASE-BASED COMPARISON OF GENERATIVE AI MODELS

J. Edakkanambeth Varayil^1,*, O. Mohamed Elfadil¹, M. Mundi¹, G. Kolar², R. Hurt¹

¹Home Parenteral and Enteral Nutrition, ²Hospital Internal Medicine, Mayo Clinic, Rochester, United States

Rationale: Clinical nutrition (CN) is increasingly complex due to the rising prevalence of chronic illness, cancer, and malnutrition-related conditions such as short bowel syndrome (SBS) and refeeding syndrome. Despite its clinical significance, nutrition education among U.S. physicians remains limited. Simultaneously, large language model (LLM)–-based AI assistants have emerged as tools to support complex clinical decision-making (CDM) but remain largely untested in CN. 

Methods: This retrospective, IRB-approved study evaluated five LLM-based AI assistants—ChatGPT, OpenEvidence, Gemini, Copilot, and DeepSeek—using five real-world CN cases. Each AI assistant was queried with patient-specific nutrition questions. Responses were blinded and reviewed by five physician CN experts using an eight-item assessment tool evaluating clarity, relevance, evidence, and clinical utility. A linear mixed-effects model analyzed score differences. 

Results: All AI assistants produced clinically appropriate responses, with Gemini scoring highest in relevance (4.04) and clarity (4.16). Overall satisfaction scores ranged from 3.08 (Copilot) to 3.84 (Gemini). However, no statistically significant performance differences were observed across the five cases. Citation quality and novel insights were variable and generally limited

Conclusion: LLM-based AI tools can reliably replicate expert reasoning in CN. While not yet a source of novel clinical insights, they may serve as adjuncts in CDM, education, and improving access to nutrition care, especially in underserved settings that lack CN expertise

Disclosure of Interest: None declared