Artificial Intelligence Versus Human-Controlled Doctor in Virtual Reality Simulation for Sepsis Team Training: Randomized Controlled Study

Background: Interprofessional communication is needed to enhance the early recognition and management of patients with sepsis. Preparing medical and nursing students using virtual reality simulation has been shown to be an effective learning approach for sepsis team training. However, its scalability is constrained by unequal cohort sizes between medical and nursing students. An artificial intelligence (AI) medical team member can be implemented in a virtual reality simulation to engage nursing students in sepsis team training.

Objective: This study aimed to evaluate the effectiveness of an AI-powered doctor versus a human-controlled doctor in training nursing students for sepsis care and interprofessional communication.

Methods: A randomized controlled trial study was conducted with 64 nursing students who were randomly assigned to undertake sepsis team training with an AI-powered doctor (AI-powered group) or with medical students using virtual reality simulation (human-controlled group). Participants from both groups were tested on their sepsis and communication performance through simulation-based assessments (posttest). Participants' sepsis knowledge and self-efficacy in interprofessional communication were also evaluated before and after the study interventions.

Results: A total of 32 nursing students from each group completed the simulation-based assessment, sepsis and communication knowledge test, and self-efficacy questionnaire. Compared with the baseline scores, both the AI-powered and human-controlled groups demonstrated significant improvements in communication knowledge (P=.001) and self-efficacy in interprofessional communication (P<.001) in posttest scores. For sepsis care knowledge, a significant improvement in sepsis care knowledge from the baseline was observed in the AI-powered group (P<.001) but not in the human-controlled group (P=.16). Although no significant differences were found in sepsis care performance between the groups (AI-powered group: mean 13.63, SD 4.23, vs human-controlled group: mean 12.75, SD 3.85, P=.39), the AI-powered group (mean 9.06, SD 1.78) had statistically significantly higher sepsis posttest knowledge scores (P=.009) than the human-controlled group (mean 7.75, SD 2.08). No significant differences were found in interprofessional communication performance between the 2 groups (AI-powered group: mean 29.34, SD 8.37, vs human-controlled group: mean 27.06, SD 5.69, P=.21). However, the human-controlled group (mean 69.6, SD 14.4) reported a significantly higher level of self-efficacy in interprofessional communication (P=.008) than the AI-powered group (mean 60.1, SD 13.3).

Conclusions: Our study suggested that AI-powered doctors are not inferior to human-controlled virtual reality simulations with respect to sepsis care and interprofessional communication performance, which supports the viability of implementing AI-powered doctors to achieve scalability in sepsis team training. Our findings also suggested that future innovations should focus on the sociability of AI-powered doctors to enhance users' interprofessional communication training. Perhaps in the nearer term, future studies should examine how to best blend AI-powered training with human-controlled virtual reality simulation to optimize clinical performance in sepsis care and interprofessional communication.

Trial Registration: ClinicalTrials.gov NCT05953441; https://clinicaltrials.gov/study/NCT05953441.