Abstract: TH-PO008
Application of Large Language Models such as ChatGPT to Support Nutritional Recommendations for Dialysis Patients
Session Information
- AI, Digital Health, Data Science - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Wang, Lin-Chun, Renal Research Institute, New York, New York, United States
- Zhang, Hanjie, Renal Research Institute, New York, New York, United States
- Ginsberg, Nancy, Fresenius Kidney Care, Waltham, Massachusetts, United States
- Kotanko, Peter, Renal Research Institute, New York, New York, United States
Background
The diversity of food preferences has increased over the past years and provides challenges to dietitians working in dialysis clinics. We explored the use of Large Language Models (LLM), such as ChatGPT, to support nutritional advice given to patients.
Methods
We applied a simple LLM syntax that allows us to consider a variety of factors that determine personalized dietary recommendations/menu options. The syntax integrates several domains, such as patient demographics (e.g., age, sex, dialysis vintage) and food preferences (e.g., vegetarian, Indian, Mediterranean, etc.), laboratory data (e.g., levels of phosphate, potassium), and clinical characteristics (e.g., weight, BMI, blood pressure). We deployed these characteristics as queries to ChatGPT (ChatGPT (openai.com)). The resulting dietary recommendations were reviewed by an experienced renal dietitian.
Results
The ChatGPT command line allowed simple input of inquiries. We were able to narrow down and personalize requests in a short dialogue with the bot, for example: “John is a hemodialysis patient aged 65, with diabetes, CHF and fluid overload, his weight is 95 Kg, BMI is 32, his albumin is 3.5 g/dl, potassium is 4.8 mmol/L, phosphorous is 5.3 mg/dl. John likes Mediterranean food. Could you develop a few recipes for him?” We created multiple requests for several imagined patients. The review of the bot’s dietary recommendations by a renal dietitian was in general satisfactory. However, while the amount of phosphate was in target, phosphate additives were not considered by the LMM. We also explored the use of ChatGPT to create personalized dietary recommendations/menu options in different languages, such as Mandarin. The translations, as judged by native speakers, were reliable.
Conclusion
While we identified some gaps, LLMs such as ChatGPT hold promise to provide personalized nutritional guidance to diverse populations of hemodialysis patients. As a next step, exploring their application in clinical practice will be important.
Funding
- Commercial Support – Renal Research Institute