JMIR Publications

ABSTRACT

Background:

Thoracic surgery residency training requires residents to integrate imaging, staging, guideline-based evidence, and multidisciplinary treatment recommendations when managing patients with lung cancer or pulmonary nodules. Artificial intelligence–assisted clinical decision support systems may provide structured support for clinical reasoning, but empirical evidence regarding their educational use in residency training remains limited.

Objective:

This study aimed to evaluate the immediate assistive effect of an AI-assisted clinical decision support system on clinical decision-making performance, task efficiency, and user satisfaction among thoracic surgery residents.

Methods:

This prospective self-controlled study included 49 residents rotating in the Department of Thoracic Surgery. Following a three-stage standardized AI training curriculum, participants completed a two-stage assessment: Phase I, independent diagnosis and treatment planning based on patient data; and Phase II, AI-assisted planning. Clinical performance was evaluated by blinded faculty raters using a Clinical Practice Assessment Scale with predefined scoring criteria, and user satisfaction was measured using a Likert-scale questionnaire. Data were analyzed using Wilcoxon signed-rank tests, Spearman correlation analysis, and linear regression models.

Results:

Clinical practice scores were significantly higher in the AI-assisted phase than in the independent phase (mean 89.39, SD 6.73 vs mean 73.56, SD 13.82; P<.001), while task completion time was shorter (mean 5.57, SD 1.08 minutes vs mean 11.20, SD 1.90 minutes; Z=−6.126; P<.001). Linear regression showed that residents from nonmajor specialties achieved a greater magnitude of improvement than those from thoracic/cardiac-related specialties (B=11.277; P<.001). Conversely, senior residents showed smaller gains compared with junior residents (B=−5.446; P<.001). Among residents, satisfaction was positively correlated with the magnitude of performance improvement (r=.446; P=.001).

Conclusions:

AI assistance was associated with significantly higher clinical decision-making scores and shorter task completion time during case-based assessment, with greater immediate performance gains among nonmajor and junior residents. Because this study measured performance under real-time AI assistance rather than durable, AI-independent learning gains, the findings should be interpreted as evidence of the platform’s immediate assistive value. Future studies incorporating delayed AI-free assessments and controlled designs are needed to determine whether these short-term gains translate into durable educational benefits. Clinical Trial: Not registered. This single-center educational evaluation study involving resident physicians was not prospectively registered as a clinical trial. Ethics approval was obtained from the Ethics Committee of Southwest Hospital, Army Medical University (Approval No.KY2025035).