Evaluation of a single surgeon's learning curve of laparoscopic pancreaticoduodenectomy: risk-adjusted cumulative summation analysis.

Abstract

Laparoscopic pancreaticoduodenectomy (LPD) is technically demanding and there is much controversy about its safety. We evaluated the learning curve for pure LPD based on the clinical outcomes of consecutive patients treated by a single surgeon. We reviewed the medical records of 119 consecutive patients who underwent LPD by a single surgeon between June 2013 and August 2018. The learning curve was evaluated using the cumulative summation (CUSUM) and risk-adjusted CUSUM (RA-CUSUM) methods. Perioperative outcomes were compared among the learning curve phases. CUSUM analysis of the operation time showed that the operation time improved after the 47th case. RA-CUSUM analysis showed the learning curve for surgical failure, defined as severe complications (Clavien-Dindo grade ≥ 3) or open conversion, comprised three phases (phase 1: cases 1-60; phase 2: cases 61-83; phase 3: cases 84-119). There were no significant differences in operation time among the three phases. Intraoperative blood loss decreased significantly over the three phases (P = 0.032). There were no postoperative deaths. The rates of postoperative complications, pancreatic fistula (grade B/C), and post-pancreatic hemorrhage were significantly lower in phase 3 than in phase 2 (2.8% vs. 21.7%, P = 0.019; 2.8% vs. 17.4%, P = 0.049; 0% vs. 13.0%, P = 0.026), but not between phases 1 and 2. Postoperative hospital stay decreased progressively, and was significantly shorter in phase 3 than in phase 1 (9.1 vs. 16.7days, P = 0.001). The LPD failure rate decreased after the first 60 cases and stabilized after 84 cases. For safe dissemination of LPD, it is important to shorten the long learning curve and decrease the unfavorable outcomes in the early phase of the learning curve.