Learning curve of da vinci robotic lobectomy: a single-surgeon CUSUM analysis of operative time and perioperative outcomes

To characterize the operative-time learning curve for da Vinci robotic-assisted lobectomy in a single surgeon’s initial experience, and to examine, using trend analyses of patient comorbidity and surgical-complexity variables and adjusted regression models, whether any apparent change in operative time was influenced by a shift in case complexity. We retrospectively reviewed 47 consecutive patients who underwent da Vinci robotic-assisted lobectomy performed by a single board-certified thoracic surgeon at Kaohsiung Chang Gung Memorial Hospital between September 30, 2024, and July 6, 2026. Patients who underwent wedge resection, video-assisted thoracoscopic surgery (VATS), segmentectomy, or resection with the da Vinci SP system were excluded. Cases were ordered chronologically, and cumulative sum (CUSUM) of operative time was calculated as the cumulative deviation from the group mean. Because CUSUM based on the mean deviation is a descriptive method without formal decision limits, a 10,000-iteration permutation test was used to construct an empirical 95% decision-limit envelope and to test whether the observed CUSUM path exceeded the range expected under random case ordering; leave-one-out and moving-average sensitivity analyses were also performed. Early and late phases, divided at the CUSUM peak, were compared using the Mann-Whitney test for continuous variables and Fisher’s exact test for categorical variables; time to chest tube removal was additionally analyzed with the Kaplan-Meier method and log-rank test. The relationship between case sequence and Charlson Comorbidity Index and surgical-complexity variables (tumor size on pathology, malignant histology, pathologic T category, nodal status, stage, operative lobe, concomitant procedures, lymph node yield, pleural adhesions) was assessed using Spearman’s rank correlation and Fisher’s exact test; operative time was additionally modeled by linear regression with robust standard errors, adjusted successively for malignancy, tumor size, nodal status, lobe, concomitant procedures, and Charlson index. The CUSUM curve for operative time peaked at case 20 (value 265.0), with a nearly equal local peak at case 10 (260.5); leave-one-out and smoothing analyses showed the identified peak was unstable between cases 10 and 20, and the entire observed CUSUM path remained within the 95% permutation-derived decision-limit envelope (global p = 0.73), indicating that this descriptive breakpoint could not be formally distinguished from random case-to-case variation at this sample size. Operative time did not differ significantly between early ( n = 20) and late ( n = 27) phases (median 315.0 [IQR 283.8–357.2] vs. 284.0 [268.5–338.0] minutes, p = 0.186) and showed no monotonic trend with case sequence overall (Spearman’s ρ = 0.01, p = 0.93). Chest tube duration differed significantly between phases on both Mann-Whitney testing ( p = 0.041) and Kaplan-Meier/log-rank analysis (χ²=5.43, p = 0.020), although this and other secondary comparisons were not corrected for multiplicity. Charlson Comorbidity Index increased significantly with case sequence (ρ = 0.34, p = 0.02); the other surgical-complexity variables moved in the same direction without reaching significance (malignant histology 65% vs. 89%, p = 0.07; concomitant procedures 5% vs. 19%, p = 0.22), tumor size did not change (median 2.2 vs. 2.1 cm, p = 0.80), lobe distribution did not differ between phases, and no patient had pleural adhesions. Adjustment for malignancy and these variables did not materially change the association between case sequence and operative time (unadjusted − 0.03 min per case, 95% CI − 1.53 to + 1.46; fully adjusted + 0.12, − 1.48 to + 1.72) or the early–late difference (− 23.1 min unadjusted vs. −16.0 min adjusted; all p ≥ 0.27), although confidence intervals were wide. By the time of the CUSUM-identified breakpoint (lobectomy case 20), the surgeon had already accumulated 39 total robotic console cases when segmentectomy, wedge resection, and da Vinci SP cases are included. In this single-surgeon series, the operative-time CUSUM breakpoint at approximately case 10–20 should be interpreted as an exploratory, descriptive signal rather than a formally validated learning-to-proficiency transition, as it was not statistically distinguishable from random variation and the dichotomized early/late operative-time comparison was not significant. Comorbidity burden increased significantly over the study period and surgical-complexity indicators trended in the same direction, and adjustment for them did not alter the operative-time estimates; there was thus no evidence that the surgeon progressively selected easier cases, but the small sample precludes firm conclusions, and adhesion severity could not be studied because no patient had adhesions, so the findings may not apply to patients with adhesions. The surgeon’s cumulative robotic experience at the time of the breakpoint substantially exceeded the lobectomy case count alone once other robotic procedures are considered.