•  
  •  
 

Rehabilitation Practice and Science

Abstract

Background: Effective history taking in stroke rehabilitation requires comprehensive assessment skills, yet opportunities for medical undergraduates to practice such skills remain limited. Virtual patients (VPs), especially those powered by large language models (LLMs), offer a promising solution. This study aimed to develop and evaluate a GPT-based virtual training system for functional assessments in stroke rehabilitation history taking.

Methods: A mixed-methods design was employed in which VPs were modeled on a clinical stroke case and 10 core domains of functional assessments in stroke rehabilitation were integrated. A virtual examiner (VE) module provided automated scoring and feedback. Phase 1 evaluated the reliability and semantic validity of the VP/VE system. Phase 2 involved nine medical undergraduates who engaged with the VP, completed a self-efficacy questionnaire, received VE feedback, and participated in semi-structured interviews.

Results: VPs demonstrated high test-retest reliability (90.6%) and consistent VE scoring. Students’ VE scores ranged from 29 to 39 out of 40, with stronger performance in observable domains and lower scores in psychosocial aspects. Self-efficacy showed a moderate correlation with objective scores (ρ = 0.614). Qualitative feedback confirmed users’ learning effects by the four identified themes, namely, system realism and interaction, learning support and feedback, educational value in functional assessments of history taking, and understanding holistic patient care needs.

Conclusion: The GPT-based VP/VE system is a reliable, valid, and learner-accepted tool that enhances comprehensive functional assessments for stroke rehabilitation historytaking training. The proposed system also offers scalable, self-paced practice aligned with competency-based medical education for medical students.

Additional Files
Appendix_0721.docx (23 kB)
Download

Included in

Physiotherapy Commons

Share

COinS