Application and evaluation of a virtual patient system based on a multi-branch dynamic decision model in orthopedic graduate medical education.

BACKGROUND: Orthopedic Graduate Medical Education (GME) necessitates the cultivation of comprehensive clinical competencies across various subspecialties—including trauma, joint disorders, spinal conditions, and sports medicine—within a constrained training period. Conventional rotation-based training models often result in uneven exposure to infrequent yet high-risk clinical scenarios, thereby limiting the systematic development of integrative clinical reasoning and awareness of long-term patient prognoses. This study aims to design and evaluate an orthopedic generalist virtual patient artificial intelligence system (OrthoSim AI Agent) that employs a multi-branching, multi-outcome, temporally dynamic decision-making framework spanning the entire continuum of patient care, from initial diagnosis to long-term follow-up. The objective is to enhance orthopedic graduate students’ clinical proficiency within standardized educational programs. METHODS: This single-center, prospective study employing a historical control design enrolled 38 orthopedic graduate students. The control group (n = 18) completed conventional rotation training, whereas the intervention group (n = 20) received supplementary interactive core disease reinforcement training through the OrthoSim AI Agent in addition to the standard rotation curriculum. The primary outcomes evaluated were the overall and domain-specific scores derived from the Mini-Clinical Evaluation Exercise (Mini-CEX). Secondary outcomes included performance scores obtained from the Objective Structured Clinical Examination (OSCE) station focused on complication management, as well as responses to a custom-designed 5-point Likert scale satisfaction questionnaire. RESULTS: A longitudinal Mini-CEX analysis (T1–T5) showed both groups started similarly, but the intervention group’s performance significantly surpassed the control group from T4 onward (T4: P = 0.012; T5: P < 0.001). At T5, the intervention group scored higher in clinical judgment, history-taking, and organizational efficiency (all P < 0.05), with greater overall score gains. In the OSCE, the intervention group achieved significantly higher total scores (86.95 ± 9.97 vs. 80.15 ± 4.04, P = 0.019). Student satisfaction surveys (5-point Likert scale) revealed high ratings for cross-subspecialty integration (4.25 ± 0.72), safe simulation of high-risk complications (4.45 ± 0.83), and enhanced decision-making confidence (4.15 ± 0.81) (all P < 0.001). 80% were “Extremely Likely” to recommend the system. CONCLUSION: The OrthSim AI Agent-Assisted Training provides an integrated learning model based on multi-branch dynamic decision-making for orthopedic Graduate Medical Education. By simulating the complete clinical causal chain in a safe environment, it effectively promotes cross-subspecialty thinking, high-risk scenario management, and awareness of long-term prognosis.