AI-driven virtual standardized patients combined with scenario-based simulation in anesthesiology training: a randomized pilot study

Abstract

Background Anesthesiology trainees require repeated opportunities to practice perioperative crisis recognition, decision-making, communication and teamwork, which are central to crisis resource management (CRM) and non-technical skills. High-fidelity simulation supports these competencies but is resource intensive to deliver repeatedly. AI-driven virtual standardized patients (AI-VSPs) may complement simulation and structured debriefing by enabling interactive practice and feedback. Methods We conducted a prospective, randomized, three-arm pilot study among 60 students enrolled in a professional master's degree program in anesthesiology at Zhongnan Hospital of Wuhan University between January and June 2026. Participants were assigned to conventional teaching (group A, n = 20), scenario-based simulation with structured debriefing (group B, n = 20), or the same simulation pathway with AI-VSP training (group C, n = 20); each group received 16 h of teaching. In group C, four 1-hour AI-VSP sessions replaced 4 h of conventional case discussion. The primary outcome was short-term simulated clinical performance measured using a locally developed 100-point competency rubric. Secondary outcomes were theoretical examination, skills examination and teaching satisfaction, with satisfaction interpreted as acceptability rather than effectiveness. Results All participants completed the study and were analyzed. No statistically significant between-group differences were detected in the reported baseline characteristics. In participant-level post-intervention analyses, total simulated clinical performance was higher in group C than in groups A and B (88.2 ± 4.8 vs. 78.0 ± 7.7 and 80.4 ± 6.5; F(2, 57) = 13.72, P < 0.001, partial η2 = 0.325). Bonferroni-adjusted comparisons favored group C for clinical decision-making, communication, total performance, theoretical examination, skills examination and teaching satisfaction. Conclusions A time-matched pathway in which AI-VSP practice replaced part of conventional case discussion while simulation and structured debriefing were retained was associated with better short-term performance on aligned anesthesiology crisis tasks. These findings support the feasibility of this approach and suggest potential benefits in selected CRM-relevant domains, but do not establish comprehensive CRM competence, definitive effectiveness or durable transfer to clinical practice. Larger multicenter studies with prespecified analyses, detailed reporting of AI-system characteristics, delayed assessment and workplace-based outcomes are needed.