ChatGPT Shows Promise for Guiding Dietary Management of MASLD

New research suggests artificial intelligence, namely ChatGPT, may be a viable tool for assisting the development of dietary plans and nutritional recommendations for managing metabolic dysfunction-associated steatotic liver disease (MASLD).1

The simulation study used GPT-4o, the most recent version of ChatGPT's underlying model, to assess the content and appropriateness of generated 1-day menu plans according to disease-specific guidelines. Results suggest ChatGPT demonstrates foundational capabilities in formulating nutrition plans and offering recommendations for managing MASLD, but notable discrepancies were observed in macronutrient distributions and certain evidence-based recommendations, including the Mediterranean diet and regular physical activity.1

In March 2024, the US Food and Drug Administration granted accelerated approval to Madrigal Pharmaceuticals’ resmetirom (Rezdiffra) for the treatment of noncirrhotic metabolic dysfunction-associated steatohepatitis (MASH) with moderate to advanced fibrosis. The oral, thyroid hormone receptor-β selective agonist made history as the first drug to be approved for this indication and continues to be the only pharmacologic treatment option available for MASH.2

Despite now having an FDA approved drug to offer patients with MASH, multidisciplinary care continues to serve as a cornerstone of disease management, including lifestyle modifications and dietary interventions.3

“While literature indicates that AI implementation may enhance diagnostic accuracy for steatotic liver disease, there remains a paucity of research evaluating AI-driven dietary interventions in the management of MASLD,” Tugce Ozlu Karahan, PhD, an assistant professor at Istanbul Bilgi University, and colleagues wrote.1

To assess the potential of AI in developing personalized nutritional management plans for MASLD, investigators conducted a simulation study in a virtual cohort of 48 patients designed to ensure a balanced gender distribution and a standardized age of 50 years. Investigators strategically assigned varying weight values to each gender group to represent underweight, normal weight, overweight, and obese categories according to established body mass index (BMI) classifications.1

All simulated patients were designated as having a sedentary physical activity level to maintain consistency across the cohort. Additionally, all patients were assigned a CAP value of 325 dB/m to represent significant steatosis. LSM values range from 2.5 to 75 kPa to represent varying levels of fibrosis.1

In the ChatGPT prompts, investigators provided no specific weight loss targets, and only patients’ MASLD disease status was specified. All conversations were made independent by using a new session for each simulated patient.1

The content and appropriateness of the 1-day menu plan were evaluated according to disease-specific guidelines from the American Association for the Study of Liver Diseases, the European Association for the Study of the Liver, the European Association for the Study of Diabetes, and the European Association for the Study of Obesity. The energy and nutrient values obtained from the AI-generated menu plans were compared with calculated targets to determine accuracy.1

A total of 48 diet plans were generated by ChatGPT for 48 simulated patients with MASLD. Comparison of energy and macronutrient content provided by ChatGPT versus values calculated by registered dietitians revealed a mean accuracy of 91.3 ± 11.0% for energy; 136.3 ± 10.3% for protein; 133.4 ± 2.0% for fat; 51.2 ± 4.3% for carbohydrates; 136.7 ± 4.3% for saturated fatty acid (SFA) as a percentage of energy; and 88.1 ± 2.5% for fiber. Results indicated the AI-generated dietary plans provided lower than recommended energy, carbohydrate, and fiber recommendations, whereas protein, fat, and SFA as a percentage of energy were higher than recommended.1

Further analysis of accuracy by BMI category revealed significant differences between groups for all parameters except fiber (P = 0.963) and SFA as a percentage of energy (P = 0.257). The accuracy of energy (P = 0.001) and fat (P = 0.015) recommendations increased from underweight to overweight patients, while the accuracy of protein (P <.001) and carbohydrates (P = .006) recommendations decreased from underweight to obese patients.1

Investigators also pointed out that while the AI model advised weight loss for patients with obesity, it did provide such recommendations for normal-weight or overweight individuals. Additionally, ChatGPT did not specify the extent of weight loss necessary to ameliorate liver adiposity and lobular inflammation.1

Regarding dietary modifications, investigators noted ChatGPT's recommendations aligned with guidelines advising reduced intake of SFAs, high-fructose foods and beverages and ultra-processed foods. However, they pointed out the AI model did not provide guidance on adhering to a Mediterranean diet or engaging in physical activity.1

“Our findings collectively suggest that while AI tools hold promise for personalized nutrition management, further refinement is essential to ensure dietary recommendations align with established guidelines,” investigators concluded.1

References

1. Ozlu Karahan T, Kenger EB, Yilmaz Y. Artificial Intelligence-Based Diets: A Role in the Nutritional Treatment of Metabolic Dysfunction-Associated Steatotic Liver Disease? J Hum Nutr Diet. doi:10.1111/jhn.70033

2. Brooks A. Resmetirom (Rezdiffra) Receives Historic FDA Approval for Noncirrhotic NASH. HCPLive. March 14, 2024. Accessed March 3, 2025. https://www.hcplive.com/view/resmetirom-rezdiffra-receives-historic-fda-approval-for-noncirrhotic-nash

3. Brooks A. Multidisciplinary Care Model Improves Liver, Metabolic Outcomes in MASH. HCPLive. February 20, 2025. Accessed March 3, 2025. https://www.hcplive.com/view/multidisciplinary-care-model-improves-liver-metabolic-outcomes-mash