Simulating Physician-Patient Interactions Using Generative Artificial Intelligence: An Educational Tool for Medical Students

Resumo (EN)

Introduction
History-taking and communication are core competencies in medical education, yet students often report insufficient confidence and limited opportunities for practice. Standardized patients are effective but resource-intensive. Virtual patients powered by large language models may provide a scalable alternative. This study aimed to evaluate the acceptability and perceived educational impact of a virtual patient simulation using ChatGPT version 3.5 (GPT-3.5) among undergraduate medical students.

Methods
This descriptive, mixed-methods study included 56 medical students recruited in the first semester of 2025. Participants completed an online questionnaire and interacted with an AI-based virtual patient programmed to simulate a dengue fever case. Performance was scored across six domains (maximum 100 points), and structured feedback was automatically generated. Students could repeat the simulation with minor variations in demographics and symptom presentation. Data collection included demographics, prior exposure to simulations and AI, self-assessment of knowledge and skills, and perceptions of the activity.

Results
The sample was balanced by sex (51.8% male), with a mean age of 24.3 years. Most students rated the virtual patient as realistic and useful, and nearly all considered the automated feedback beneficial. Willingness to repeat the activity was high. Among those who repeated, most reported significant performance improvement and greater confidence for real encounters. Chi-square tests confirmed significant deviations from uniformity across key outcomes. Thematic analysis identified four themes: educational utility, limitations of AI, suggestions for improvement, and overall positive appraisal. In terms of prior exposure, nearly half of the students had no prior experience with AI, and most reported only occasional institutional opportunities for simulation.

Conclusion
AI-enabled virtual patients using ChatGPT were well accepted by medical students, who perceived benefits for knowledge reinforcement, performance, and confidence in history-taking. These findings support the integration of AI-based simulations as scalable, formative complements to traditional approaches in medical education.