JMIR Publications

ABSTRACT

Background:

Empathic communication is essential in nursing, yet opportunities for practicing emotionally challenging interactions and receiving individualized feedback remain limited. Multimodal artificial intelligence (AI) may complement simulation-based education by incorporating verbal, nonverbal, and physiological information into feedback.

Objective:

This study evaluated the educational outcomes of a multimodal AI-supported simulation compared with conventional role-play among undergraduate nursing students and explored participants’ learning experiences. Physiological responses and selected technical outputs of the feedback system were examined as supplementary findings.

Methods:

A mixed-methods quasi-experimental study with a pretest-posttest nonequivalent control group design included 134 nursing students (experimental group, n=68; control group, n=66). Both groups completed three clinical scenarios. The experimental group received AI-supported virtual-patient interactions, individualized multimodal feedback, and structured reflection, whereas the control group received conventional role-play and faculty-led debriefing. Questionnaire outcomes were analyzed using baseline-adjusted analysis of covariance. Heart rate variability (HRV) was summarized descriptively across four simulation phases. A preliminary technical assessment examined 42 recorded scenario encounters, and semistructured interviews with 12 students were analyzed thematically.

Results:

All participants completed the intervention. Compared with the control group, the experimental group had higher baseline-adjusted postintervention empathy (F(1,131)=31.84, P<.001, partial η²=.196) and communication self-efficacy (F(1,131)=24.91, P<.001, partial η²=.160). Group effects were also significant for emotional, behavioral, and cognitive engagement (all P<.001; partial η²=.133-.163). The technical assessment reported facial-expression classification accuracy of 81.0% and verbal-expression classification accuracy of 84.5%. HRV showed greater descriptive variation across simulation phases in the experimental group. Qualitative themes concerned empathy awareness, emotional immersion, communication confidence, feedback-guided reflection, and adaptation to AI-mediated interaction.

Conclusions:

Multimodal AI-supported simulation was associated with higher self-reported empathy, communication self-efficacy, and learning engagement than conventional role-play. The qualitative findings provided context for interpreting these outcomes, while physiological and technical findings offered complementary information about the feedback system. AI-supported simulation may serve as an additional approach to empathic communication training rather than a replacement for educator judgment.