JMIR Publications

ABSTRACT

Standardized Patient (SP) simulations are integral but resource-intensive components of medical education, aimed at developing clinical interviewing and diagnostic skills. The nature of in-person SP encounters poses challenges in scalability and access. We introduce a web-based application leveraging a Large-Language Model (LLM) to simulate physician-patient interactions, thereby offering a scalable, cost-effective alternative for practicing clinical skills while receiving personalized formative feedback. The AI Patient Actor utilizes GPT-4o to create responsive, contextually appropriate simulated patient interactions based on expert-created clinical case scenarios. Python 3.10 and Streamlit provide the development framework, while LangChain optimizes prompt and case file processing. Whisper-3 speech recognition and synthesis enables multi-language vocal interaction. Formative feedback is generated immediately during the simulations, based on established medical education rubrics. The application delivers a conversational simulation environment where students can practice and refine interviewing and diagnostic skills. It minimizes resource constraints of traditional SP methods while enabling more equitable access across medical institutions. The design safeguards against the generation of erroneous medical advice by relying on expert-created case content. LLM-powered simulations present a new frontier for medical training in clinical communication and reasoning. The application stands out for its low operational costs, scalability, and multi-language capabilities, enhancing global accessibility. Continuing to evaluate the models’ fidelity, bias, and performance across different languages is pivotal to ensure its robustness and effectiveness in medical education.