Virtual versus 3D-printed bronchoscopy simulation: changes in learner preferences, clinician appraisal, and modality-specific assessment outcomes

Abstract Background Virtual and three-dimensional (3D)-printed bronchoscopy simulators may support different aspects of early procedural learning. This study compared learner preferences, clinician appraisal, and assessment-platform performance following structured exposure to both modalities. Methods In this prospective comparative study, 68 bronchoscopy-naïve undergraduate medical students completed a five-week randomized, counterbalanced training program using both virtual and 3D-printed bronchoscopy simulators. Comparative preference surveys were administered before and after training. Ten clinician appraisers directly evaluated both simulators using a bronchoscopy-adapted Michigan Standard Simulation Experience Scale. Objective performance was assessed on both platforms using a modified Bronchoscopy Skills and Tasks Assessment Tool (BSTAT) and whole-tree navigation time. Because all participants trained with both modalities, performance changes were interpreted as assessment-platform-specific outcomes following combined multimodal training rather than isolated simulator-specific learning effects. Results After Holm adjustment across eight preference domains, clinical realism showed a significant shift toward the 3D-printed model (adjusted p < 0.0008), whereas changes in procedural realism and perceived training improvement did not remain significant. In prespecified exploratory comparisons, clinician appraisers assigned higher scores to the 3D-printed model for procedural and anatomical realism (20 [18.75–20] vs. 12 [12–15.25]; p = 0.0412) and educational usefulness (17 [16–20] vs. 13 [12.75–16]; p = 0.0339). Median total BSTAT scores increased from 19 (7.5–31.5) to 60 (41.5–65) on the virtual platform and from 9 (3–16) to 51 (39–61) on the 3D-printed platform. Larger total BSTAT and navigation score changes were observed on the 3D-printed assessment platform, whereas whole-tree navigation-time reduction was greater on the virtual platform (320.5 [192.5–413] vs. 172.5 [0–300] seconds; p < 0.0001; r = 0.65).