A Randomized Controlled Trial of the Effects of a Large Language Model on Medical Students' Clinical Questioning Skills

At present, the main form of clinical questioning skills teaching is to let undergraduates who participate in the apprenticeship first learn the characteristics and diagnostic points of cases, and then practice questioning on real patients in the wards. However, due to the large number of trainee students, it is difficult to meet the teaching demand in terms of the number of cases available for questioning and the richness of disease types under the current teaching mode. Therefore, it is necessary to utilize new intelligent technologies and create a new model of questioning skills teaching to improve teaching efficiency and enhance students' clinical thinking.

Large-scale language modeling (LLM) is a deep learning technology that can learn knowledge from a large amount of text, and AI chatbots such as ChatGPT are a typical example of its application. AI chatbots are characterized by anthropomorphic comprehension and diversified natural language generation abilities in different contexts, and have been initially applied in the medical field, such as passing the U.S. Medical Licensing Examination, assisting in ophthalmic history documentation and answering ophthalmic questions. However, it has been found that although LLM has fair modeling performance in general medical knowledge, it still needs to be improved in the area of specialty diseases. Based on this, the researcher's team has used the ophthalmology textbook, clinical guideline consensus, the Internet conversation data and knowledge base of Zhongshan Ophthalmology Center in the early stage, combined with artificial feedback reinforcement learning and other techniques to fine-tune and train the LLM, and developed "Digital Twin Patient", a localized large language model that has the ability to answer ophthalmology-related medical questions, and also constructed a combination of automated model evaluation and manual evaluation by medical experts. The evaluation system combining automated model evaluation and manual evaluation by medical experts was constructed at the same time.

This project intends to integrate "Digital Twin Patient" into undergraduate ophthalmology apprenticeship, simulate the consultation process of real patients through the online interaction between students and "Digital Twin Patient", explore the effect of "Digital Twin Patient" consultation teaching, provide emerging technology tools for guiding medical students to actively learn a variety of ophthalmology cases, cultivate clinical thinking, and provide the possibility of creating a new mode of intelligent teaching.