Related Experiment Video
Updated: May 11, 2025
09:55
Bridging the Technology Divide in the COVID-19 Era: Using Virtual Outreach to Expose Middle and High School Students to Imaging Technology
Published on: September 28, 2022
1.6K
Development of a GPT-4-Powered Virtual Simulated Patient and Communication Training Platform for Medical Students to
Dan Weisman1, Alanna Sugarman2, Yue Ming Huang1,3
1UCLA Simulation Center, University of California, Los Angeles, Los Angeles, CA, United States.
JMIR Formative Research
|April 17, 2025
Summary
Generative pretrained transformer (GPT-4) technology enables the creation of virtual simulated patients (VSPs) for medical training, offering more dynamic interactions than traditional methods. This AI-driven approach streamlines development and enhances communication skills practice for students.
Area of Science:
- Artificial Intelligence in Medical Education
- Natural Language Processing for Healthcare Simulation
- Virtual Patient Simulation Development
Background:
- Standardized patients (SPs) are valuable for training medical students in difficult conversations but face resource limitations.
- Artificial intelligence (AI) and large language models (LLMs) like generative pretrained transformers (GPT) offer new avenues for virtual simulated patient (VSP) development.
- GPT-4 allows for dynamic, text-prompt-based scenario design, moving beyond traditional branching path simulations.
Purpose of the Study:
- To describe the development process and lessons learned in creating a GPT-4-driven VSP for medical student training.
- To enable medical students to practice discussing abnormal mammography results with a virtual patient.
- To assess GPT-4's capability in generating realistic VSP responses and providing performance feedback.
Main Methods:
- A multidisciplinary team developed the VSP using an agile design process informed by user interviews.
- GPT-4 was prompted with scenario details, emotional states, and learner expectations for dialogue.
- The VSP was iteratively refined through testing, issue documentation, and prompt revision; GPT-4 provided exploratory feedback on learner communication.
Main Results:
- In-depth interviews informed the communication protocols for discussing abnormal mammography results.
- The GPT-4 driven VSP demonstrated sensible responses and appropriate emotional inflections during simulated conversations.
- GPT-4 feedback identified learner strengths and weaknesses with relevant quotes, though occasional inaccuracies in protocol adherence assessment occurred.
Conclusions:
- GPT-4 significantly streamlined VSP development and enabled more dynamic, humanlike interactions compared to traditional simulations.
- Further pilot testing with medical students is planned to evaluate the VSP's feasibility and acceptability.
- This study highlights the potential of LLMs in advancing communication skills training in medical education.
Keywords:
AIGPT-4LLMabnormal mammography resultsagentartificial intelligencebiopsycommunication skills traininglarge language modelstandardized patientvirtual simulated patient
