Comparing ChatGPT Responses with Clinical Practice Guidelines for Diagnosis, Prevention, and Treatment of Diabetes

Recent developments in artificial intelligence (AI) bring us to a new era. Especially pioneering research on deep learning and transformers leads to the success of current generative AI chatbots like ChatGPT. ChatGPT can be utilized in assisting various tasks such as the diagnosis and prevention of diabetes mellitus (diabetes). Diabetes is not a chronic disease but it is the inability of the body to produce enough insulin or to use the existing insulin properly. Untreated diabetes can cause a series of damages, such as kidney damage, heart attacks, and stroke. With early diagnosis, a healthy diet and exercise, most diabetes cases can be prevented. However, people in disadvantaged counties and with low incomes may not have access to proper medical services. In these cases, ChatGPT is a potential tool that can be utilized for diabetes education, such as early diagnosis of the symptoms, prevention of the disease with healthy diet and exercises, and also can provide possible treatment plans with the guidance of clinicians. However, the accuracy of ChatGPT answers to diabetes-related questions is unclear. In this work, for the first time, answers given by ChatGPT to diabetes-related questions are compared with the well-known clinical practice guidelines of KDIGO and the International Diabetes Federation. In particular, we assess how well ChatGPT answers are matching with the clinical guidelines that is the novelty of the proposed work. We also discuss ethical and security issues when using ChatGPT for disease diagnosis, prevention, and treatment.