BACKGROUND: Large language models (LLMs) are becoming increasingly important as they are being used more frequently for providing medical information. Our aim is to evaluate the effectiveness of electronic artificial intelligence (AI) large language models (LLMs), such as ChatGPT-4, BingAI, and Gemini in responding to patient inquiries about retinopathy of prematurity (ROP).
METHODS: The answers of LLMs for fifty real-life patient inquiries were assessed using a 5-point Likert scale by three ophthalmologists. The models\' responses were also evaluated for reliability with the DISCERN instrument and the EQIP framework, and for readability using the Flesch Reading Ease (FRE), Flesch-Kincaid Grade Level (FKGL), and Coleman-Liau Index.
RESULTS: ChatGPT-4 outperformed BingAI and Gemini, scoring the highest with 5 points in 90% (45 out of 50) and achieving ratings of \"agreed\" or \"strongly agreed\" in 98% (49 out of 50) of responses. It led in accuracy and reliability with DISCERN and EQIP scores of 63 and 72.2, respectively. BingAI followed with scores of 53 and 61.1, while Gemini was noted for the best readability (FRE score of 39.1) but lower reliability scores. Statistically significant performance differences were observed particularly in the screening, diagnosis, and treatment categories.
CONCLUSIONS: ChatGPT-4 excelled in providing detailed and reliable responses to ROP-related queries, although its texts were more complex. All models delivered generally accurate information as per DISCERN and EQIP assessments.

Artificial intelligence (AI) chatbots utilizing large language models (LLMs) have recently garnered significant interest due to their ability to generate humanlike responses to user inquiries in an interactive dialog format. While these models are being increasingly utilized to obtain medical information by patients, scientific and medical providers, and trainees to address biomedical questions, their performance may vary from field to field. The opportunities and risks these chatbots pose to the widespread understanding of skeletal health and science are unknown. Here we assess the performance of 3 high-profile LLM chatbots, Chat Generative Pre-Trained Transformer (ChatGPT) 4.0, BingAI, and Bard, to address 30 questions in 3 categories: basic and translational skeletal biology, clinical practitioner management of skeletal disorders, and patient queries to assess the accuracy and quality of the responses. Thirty questions in each of these categories were posed, and responses were independently graded for their degree of accuracy by four reviewers. While each of the chatbots was often able to provide relevant information about skeletal disorders, the quality and relevance of these responses varied widely, and ChatGPT 4.0 had the highest overall median score in each of the categories. Each of these chatbots displayed distinct limitations that included inconsistent, incomplete, or irrelevant responses, inappropriate utilization of lay sources in a professional context, a failure to take patient demographics or clinical context into account when providing recommendations, and an inability to consistently identify areas of uncertainty in the relevant literature. Careful consideration of both the opportunities and risks of current AI chatbots is needed to formulate guidelines for best practices for their use as source of information about skeletal health and biology.
Artificial intelligence chatbots are increasingly used as a source of information in health care and research settings due to their accessibility and ability to summarize complex topics using conversational language. However, it is still unclear whether they can provide accurate information for questions related to the medicine and biology of the skeleton. Here, we tested the performance of three prominent chatbots—ChatGPT, Bard, and BingAI—by tasking them with a series of prompts based on well-established skeletal biology concepts, realistic physician–patient scenarios, and potential patient questions. Despite their similarities in function, differences in the accuracy of responses were observed across the three different chatbot services. While in some contexts, chatbots performed well, and in other cases, strong limitations were observed, including inconsistent consideration of clinical context and patient demographics, occasionally providing incorrect or out-of-date information, and citation of inappropriate sources. With careful consideration of their current weaknesses, artificial intelligence chatbots offer the potential to transform education on skeletal health and science.