BackgroundArtificial intelligence (AI) and large language models (LLMs) like ChatGPT have transformed information retrieval, including in healthcare. ChatGPT, trained on diverse datasets, can provide medical advice but faces ethical and accuracy concerns. This study evaluates the accuracy of ChatGPT-3.5's answers to frequently asked questions about oral cancer, a condition where early diagnosis is crucial for improving patient outcomes.MethodsA total of 20 questions were asked to ChatGPT-3.5, selected from Google Trends and questions asked by patients in the clinic. The responses provided by ChatGPT were evaluated for accuracy by medical oncologists and oral and maxillofacial radiologists. Inter-rater agreement was assessed using Fleiss’s and Cohen kappa tests. The scores given by the specialties were compared with the Mann-Whitney U test. The references provided by ChatGPT-3.5 were evaluated for authenticity.ResultsOf the 80 responses from 20 questions, 41 (51.25%) were rated as very good, 37 (46.25%) as good, 2 (2.50%) as acceptable. There was no significant difference between oral and maxillofacial radiologists and medical oncologists in all 20 questions. Of the 81 references to ChatGPT-3.5 answers, only 13 were scientific articles, 10 were fake, and the remaining references were data from websites.ConclusionChatGPT provided reliable information about oral cancer and did not provide incorrect information and suggestions. However, all information provided by ChatGPT is not based on real references.
Read full abstract