Assessing the Accuracy of ChatGPT in Answering Questions About Prolonged Disorders of Consciousness.
Brain sciences April 13, 2025 Sergio Bagnato, Cristina Boccagni, Jacopo Bonavita 3 citations
Two versions of the ChatGPT large language model (4o and o1) answered 57 open-ended questions about prolonged disorders of consciousness, written as if from a patient's relative. Both models gave predominantly correct answers (80.7-96.8% accuracy). ChatGPT 4o showed greater empathy, while ChatGPT o1 more often recommended consulting a healthcare professional (especially in Italian). English responses were more accurate than Italian only for ChatGPT 4o on clinical data. The findings suggest chatbots could help support caregivers of people with disorders of consciousness, but occasional inaccuracies mean information should be verified with a doctor.