Embracing the Future of Medical Education With Large Language Model-Based Virtual Patients: Scoping Review.

BACKGROUND: In recent years, large language models (LLMs) have experienced rapid development. LLM-based virtual patients have begun to gain attention, offering new opportunities for simulations in medical education. OBJECTIVE: This study aims to systematically analyze the current applications, research trends, and challenges of LLM-based virtual patients in medical education and to explore potential future directions for development. METHODS: This study adheres to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines. Five databases (Web of Science Core Collection, PubMed, IEEE Xplore, Embase, and Scopus) were searched from January 1, 2018, to June 24, 2025, to identify studies related to the application of LLM-based virtual patients in medical education. A comprehensive analysis of LLM-based virtual patients from research design to application and evaluation was conducted. RESULTS: A total of 28 studies were included in this scoping review. Analysis revealed that 92.9% (26/28) of the studies were published in the past 2 years, indicating that LLM-based virtual patient research is still in its early stages. The research primarily focuses on medical training and spans a wide range of medical disciplines. When using LLMs, advanced technologies such as social robots, virtual reality, and mixed reality are used to present LLM-based virtual patients. Combining these technologies with various supplementary tools enhances the realism of LLM-based virtual patients and improves user interaction. The evaluation of LLM-based virtual patients mainly emphasizes user experience. However, evaluation methods lack standardization, and only 13% (3/23) of studies used validated tools in assessing LLM-based virtual patients, while only 21.7% (5/23) of studies objectively measured learning outcomes facilitated by LLM-based virtual patients. All included studies expressed a positive attitude toward LLM-based virtual patients; however, they overlook privacy and security considerations in practical applications. CONCLUSIONS: LLM-based virtual patients hold significant innovation potential in medical education and are still in the early stages of development. They are primarily applied in medical training and show promise in communication skills training, although they cannot replace real-world interactions. Moreover, the heterogeneity of research designs, the absence of nonverbal cues in interactions, and concerns regarding privacy and security limit their broader implementation. Future research should focus on improving the reliability, realism, safety, and scientific efficacy of LLM-based virtual patients. TRIAL REGISTRATION: Open Science Framework Registries 10.17605/OSF.IO/DMC9Q; https://osf.io/DMC9Q/overview.