Recipe for Success? The Power of AI to Enhance Kidney Diet Support

Chronic kidney disease (CKD) is a progressive condition impacting over 10% of the global population, predominantly afflicting older individuals and those with diabetes mellitus or hypertension (1). The importance of kidney diet education in CKD cannot be overstated (2). Proper dietary management can significantly slow the progression of CKD, reduce the risk of complications such as hyperkalemia and hyperphosphatemia, and improve the quality of life for patients. A kidney diet typically involves the careful control of nutrient intake, including potassium, phosphorus, and protein, to alleviate the kidneys’ workload and prevent further damage. Navigating the complexities of a kidney diet for individuals with CKD requires careful planning and monitoring.

The rapid development of artificial intelligence (AI) technologies, particularly advanced generative models such as OpenAI's ChatGPT, has sparked extensive research and exploration across various domains, including health care (3). Within nephrology, we have thoroughly explored the utility of ChatGPT alongside other language models such as Google's Gemini (formerly known as Bard) and Microsoft's Copilot (formerly known as Bing Chat) (4). One of these investigations focused on the capability of various AI models to accurately identify the potassium and phosphorus levels in foods (5). A selection of 240 food items, derived from the Mayo Clinic Renal Diet Handbook (5), tailored for patients with CKD, was evaluated through each model. GPT-4, the latest iteration of ChatGPT, exhibited exceptional proficiency in identifying potassium levels, successfully classifying 81% of the food items (Figure, A). It demonstrated a remarkable accuracy rate of 99% in identifying high potassium foods, surpassing the performance of Gemini (79%) and Copilot (81%) (Figure, B). In the analysis of phosphorus content, Gemini emerged as the most accurate, achieving a perfect 100% accuracy rate, significantly outperforming Copilot (89%) and GPT-4 (77%) (Figure, A). The study also extended to evaluating the chatbots’ effectiveness in categorizing foods based on oxalate content (6). Out of 549 food items, Gemini led with an 84% accuracy rate in classifying food items based on the oxalate levels, followed by Copilot at 60% and GPT-4 at 52% (Figure, A). AI models demonstrated greater accuracy in identifying foods with low oxalate than those with moderate or high levels (Figure, B).

View Full Size

Performance of AI models in supporting kidney diet management

Citation: Kidney News 16, 6

• Download Figure
• Download figure as PowerPoint slide

Notably, a study showed that ChatGPT passed the Chinese Registered Dietitian exam, with 96% of ChatGPT's answers preferred by professional dietitians (7). It was also suggested that ChatGPT could offer personalized nutritional guidance for individuals with obesity, cardiovascular diseases, and type 2 diabetes, although its capability to formulate balanced meal plans should be further improved (8, 9). Additionally, research on ChatGPT's diet advice for those with food allergies found it could occasionally suggest unsafe diets that include allergens. ChatGPT might also make mistakes in food quantities and energy values and suggest repetitive diets lacking in variety (10).

Our findings, in conjunction with those from other studies, underscore the potential of AI as an impactful tool in enhancing dietary planning for patients with CKD, although its efficacy demands further improvement. This indicates a pivotal moment in health care utilizing technology to tackle complex health challenges. However, concerns over AI's ability to filter out erroneous information underline the importance of complementing, not replacing, professional judgment. It necessitates expert oversight to ensure the accuracy of AI-recommended diets. As medicine and AI evolve, their integration must reflect core medical values: care, empathy, and trust. Future studies should focus on the ethical integration of AI into health records, emphasizing safety and ethics. Health care professionals, including nephrologists, should endeavor to continuously adapt to the evolving landscape of AI. With AI's progress and increased accuracy, we are optimistic about its role in CKD dietary support. With ongoing research and addressing current limitations, we believe that AI will significantly aid in CKD diet planning.