Characterizing the Learning Curve of ION Robotic Bronchoscopy

摘要

BACKGROUND: To characterize the procedural learning curve for shape-sensing ION robotic-assisted bronchoscopy (ION) at a single institution, based on diagnostic performance and procedure time. METHODS: A retrospective analysis was conducted on 147 pulmonary nodules sampled with the ION SYSTEM across 129 procedures. The diagnostic yield was defined as the proportion of samples that led to a definitive diagnosis of malignancy or non-neoplastic disease. Learning curve analyses included cumulative sum (CUSUM) control chart methods for diagnostic yield and procedure time. To minimize procedural variability, only single-target cases were used in procedural time analysis. Generalized linear mixed-effects models (GLMMs) were used to identify predictors of diagnostic success. RESULTS: Among the 147 nodules, 129 (87.8%) yielded a definitive diagnosis, comprising 108 malignant and 21 non-malignant. The diagnostic CUSUM analysis revealed a steady performance improvement, approaching a plateau after 45 cases had been processed. For the subset of single-nodule procedures, the median procedure time decreased from 60 to 35 minutes, with visual change-point stabilization evident by case 40. GLMM identified a significant inverse association between procedure duration and diagnostic success (OR: 0.997; 95% CI: 0.994-0.999; P=0.005). Nodule size <15 mm was not significantly associated with a lower diagnostic yield (OR: 0.971; 95% CI: 0.873-1.080; P=0.611). No significant complications occurred; minor self-limited bleeding was observed in 7% of cases. CONCLUSION: Implementing the ION system demonstrates a measurable learning curve, with significant improvements in diagnostic yield and procedural efficiency observed after 40 cases have been performed. These findings underscore the importance of structured training and performance monitoring during the initial phase of adoption.