Use of a generative pre-trained transformer-based virtual patient for health assessment and communication training in...

This study evaluated the use of a generative pre-trained transformer (GPT)-based virtual patient in nursing education.

In nursing education, conventional training methods such as interactions with real-life or standardized patients exhibit limitations such as psychological distress, repetitive training and insufficient cost- and time-effectiveness. Because of their capacity to emulate human-like dialogue, GPTs have emerged as a valuable resource for educational nursing activities.

This study employed a mixed-methods design.

A GPT-based virtual patient with acute appendicitis was included. Twenty-eight new prospective nurses in South Korea, equipped with a head-mounted display, evaluated and communicated with the virtual patient. Usability, perceived virtual learning environment and self-efficacy in communication were measured. The GPT-generated dialogues and open-ended questions were subjected to qualitative analysis.

Among the subfactors of usability, the subdomains of perceived accessibility of functions and perceived virtual learning environments achieved high scores. Furthermore, a notable increase in self-efficacy for communication was observed (t = -2.82, p = .009). The participants' experiences with the GPT-based virtual patient were divided into "educational effects and learner experience" and "technical limitations and the need for improvement." Evaluation of the dialogue between the GPT-based virtual patient and participants revealed that the readability subdomain achieved the highest score, whereas the accuracy subdomain achieved the lowest score.

The findings of the present study provide insights into the advantages of employing GPT-based virtual patients, particularly regarding the perceived accessibility of functions, high scores for immersion and enhanced self-efficacy of communication.