LLM-Based Clinical History Taking System: A Persona-Driven Approach.

Area of Science:

• Medical Education Technology
• Artificial Intelligence in Healthcare
• Clinical Skills Training

Background:

• Standardized patients (SPs) are crucial for medical training but are resource-intensive to develop and manage.
• Traditional SP programs face challenges related to cost, scalability, and consistency.
• There is a need for innovative, efficient methods to create high-quality clinical training materials.

Purpose of the Study:

• To develop and validate a systematic approach for generating standardized patient cases using GPT-4o.
• To assess the clinical authenticity and accuracy of AI-generated SP cases.
• To evaluate the effectiveness and student perception of virtual patient interactions.

Main Methods:

• A systematic case generation protocol was developed utilizing GPT-4o.
• Twelve SP cases were created across three distinct chief complaints.
• Expert faculty validated the cases for clinical authenticity and accuracy.
• Medical students (n=15) interacted with virtual patients and provided feedback.

Main Results:

• Expert validation indicated good clinical authenticity (information flow: 3.7±0.5, realism: 3.4±0.8) and accuracy (3.4±0.5).
• Medical students rated the virtual patient interactions positively, with high scores for appropriateness (7.9/10) and accuracy (7.5/10).
• The study demonstrated the feasibility of using LLMs for creating effective SP training cases.

Conclusions:

• Large Language Models like GPT-4o can systematically generate standardized patient cases.
• AI-generated SP cases demonstrate promising clinical authenticity and accuracy for educational purposes.
• This approach offers a viable, resource-efficient complement to traditional standardized patient programs in medical education.