Integrating large language model-based agents into a virtual patient chatbot for clinical anamnesis training

Area of Science:

• Medical Education
• Artificial Intelligence in Healthcare
• Clinical Communication

Background:

• Effective healthcare communication is vital for patient trust and clinical decisions.
• Traditional training methods like role-playing are costly and lack realism.
• Current simulation tools face limitations in scalability and personalized practice.

Purpose of the Study:

• To develop and evaluate a virtual patient system utilizing Large Language Models (LLMs) for healthcare communication training.
• To provide a scalable, autonomous, and interactive platform for practicing clinical communication skills.
• To address the limitations of traditional and existing simulation-based training methods.

Main Methods:

• Development of a virtual patient system powered by Large Language Models (LLMs).
• Implementation of interactive chatbot scenarios simulating patient roles.
• Inclusion of features for student case selection and teacher-created custom cases.
• Evaluation of the system's response consistency, plausibility, and usability.

Main Results:

• The LLM-powered agent provided consistent and plausible responses aligned with clinical case descriptions.
• The system achieved a high Chatbot Usability Questionnaire (CUQ) score of 86.25/100.
• The approach facilitates flexible, repetitive, and asynchronous practice with real-time feedback.

Conclusions:

• LLM-powered virtual patient systems offer a promising, scalable solution for healthcare communication training.
• This technology enhances learning by providing personalized, accessible, and interactive practice opportunities.
• The developed tool supports the improvement of clinical reasoning and decision-making skills through simulated patient interactions.