Learning curve stratified outcomes after robotic pancreatoduodenectomy

Background: Robotic pancreatoduodenectomy is increasingly being implemented worldwide, with good results reported from individual expert centers. However, it is unclear to what extent outcomes will continue to improve during the learning curve, as large international studies are lacking. Methods: An international retrospective multicenter case series, including consecutive patients after robotic pancreatoduodenectomy from 18 centers in 8 countries in Europe, Asia, and South America until December 31, 2019, was conducted. A cumulative sum analysis was performed to determine the inflection points for the feasibility (operative time and blood loss) and proficiency (postoperative pancreatic fistula grade B/C and major morbidity) learning curves. Outcomes were compared in 3 groups on the basis of the learning curve inflection points. Results: Overall, 2,186 patients after robotic pancreatoduodenectomy were included. The feasibility learning curve was reached after 30–45 robotic pancreatoduodenectomy procedures and the proficiency learning curve after 90 robotic pancreatoduodenectomy procedures. These inflection points created 3 phases, which were associated with major morbidity (24.7%, 23.4%, and 12.3%, P < .001) but not 30-day mortality (2.1%, 2.0%, and 1.5%, P = .670). Other outcomes mostly continued to improve, including median operative time 432, 390, and 300 minutes (P < .0001), conversion 6.0%, 4.7%, and 2.7% (P = .002), bile leakage 7.2%, 4.1%, and 2.4% (P < .001), postpancreatectomy hemorrhage 6.5%, 6.1%, and 1.8% (n = 21) but not R0 resection (pancreatic ductal adenocarcinoma only) 78.5%, 73.9%, and 82.8% (P = .35), and 90-day mortality rate 3.1%, 3.5%, and 2.1% (P = .191). Centers performing >20 robotic pancreatoduodenectomies annually had lower rates of conversion, reoperation, and shorter median operative time as compared with centers performing 10–20 robotic pancreatoduodenectomies annually. Conclusion: This international multicenter study demonstrates that most outcomes of robotic pancreatoduodenectomy continued to improve during 3 learning curve phases without a negative effect on 90-day mortality. Randomized studies are needed in high-volume centers that have surpassed the first learning curves, to compare these outcomes with the open approach.