A recent study published in the journal Nutrients evaluated the potential of the Chat Generative Pretrained Transformer (ChatGPT) to provide nutritional counseling.
Non-communicable diseases (NCDs) are the leading cause of death, accounting for 74% of global deaths. The 2019 Global Burden of Disease study estimated 43.8 million cases of type 2 diabetes (T2D), 1.2 billion cases of non-alcoholic fatty liver disease (NAFLD), and 18.5 million cases of hypertension. The prevalence of obesity nearly tripled between 1975 and 2016.
Various studies have consistently highlighted the influence of lifestyle and dietary factors on the onset and progression of non-communicable diseases. In recent times, there has been an increase in internet searches for health-related information. ChatGPT is a widely used chatbot that generates responses to text queries. It can understand context and provide coherent answers.
ChatGPT has become an accessible and efficient resource for medical inquiries. Chatbots can provide interactive, personalized patient education and support in real time, contributing to improved patient outcomes. However, limited data exist on the benefit of ChatGPT for improving nutrition in NCD patients.
Study: Is ChatGPT an Effective Tool for Nutritional Counseling?
The Study and Findings
In the study, researchers compared nutritional counseling from ChatGPT with recommendations from international guidelines related to non-communicable diseases. The analyses were conducted using the standard ChatGPT model (version 3.5). The team included diseases requiring specific nutritional treatments, such as arterial hypertension, T2D, dyslipidemia, obesity, NAFLD, sarcopenia, and chronic kidney disease (CKD).
A series of prompts for these conditions, formulated by physicians and nutritionists, were used to obtain nutrition advice from the chatbot. Separate chat sessions were conducted for each prompt conversation. The responses from ChatGPT were compared to recommendations from international clinical guidelines. Two nutritionists independently evaluated and categorized the responses from ChatGPT. Responses were deemed „appropriate“ if they aligned with the guidelines, and „inappropriate“ if they were contradictory.
In addition, responses were classified as „unsupported“ if they were not confirmed in the guidelines, „not fully compliant“ if they did not fully meet the guidelines, and „general advice“ if they were unspecific but overall promoted a healthy diet. The team also assessed whether ChatGPT could replace the consultation of a nutritionist in treating complex cases, presenting a scenario with (a patient with) multiple concurrent conditions (CKD, obesity, and T2D).
Overall, the advice from ChatGPT was correct. The appropriateness rates ranged from 55.5% for sarcopenia to 73.3% for NAFLD. There was one contradictory response each for NAFLD and obesity. In the case of obesity, the chatbot recommended regular meals and snacks to stabilize blood sugar levels, while the guidelines emphasize avoiding snacks between meals.
For NAFLD, benefits were reported from dietary supplements like omega-3 fatty acids, antioxidants, and vitamin E under medical supervision. However, existing guidelines do not recommend antioxidants and omega-3 fatty acids for NAFLD treatment. A T2D-related recommendation was not supported by the guidelines.
Specifically, ChatGPT suggested dividing meals into smaller, balanced portions as an alternative to three large meals. While this was not in direct contradiction to the guidelines, it is not addressed in existing T2D guidelines. In addition, numerous responses were categorized as not fully compliant with the guidelines.
ChatGPT, for example, emphasized using low glycemic index foods and monitoring portion sizes for hypertriglyceridemia, while guidelines recommend combating overweight and reducing carbohydrates. Additionally, ChatGPT offered general health advice for various conditions, such as consuming adequate fluids, avoiding processed foods, and consuming lean protein. The chatbot also repeatedly emphasized overall well-being and appetite control.
The general advice included foods to be included in the diet that are not frequently reported in the guidelines. Regarding the scenario of a patient with obesity, CKD, and T2D, some suggestions from ChatGPT were deemed inappropriate or contradictory. For example, there was an emphasis on prioritizing lean proteins for muscle health and then a suggestion to limit overall protein intake. The responses were generally broad and repeatedly emphasized consulting a nutritionist.
The findings highlight several concurrences and differences in ChatGPT’s responses to nutritional guidelines. The responses were clear and included practical examples of foods to be included or excluded from the diet. Some recommendations from ChatGPT were partially compliant. The chatbot could not provide adequate guidance in the case of multiple concurrent conditions. While ChatGPT was fairly accurate in providing nutritional recommendations for non-communicable diseases, limitations were evident in complex scenarios. While ChatGPT has the potential to be helpful, it cannot replace expert advice.