Defining the Learning Curve for Robotic Pancreaticoduodenectomy at a Single Center with an Established Laparoscopic Pancreaticoduodenectomy Program

Abstract

Introduction: Robotic pancreaticoduodenectomy (RPD) has a learning curve (LC) of approximately 30-250 cases to reach proficiency. The LC for laparoscopic pancreaticoduodenectomy (LPD) at Duke University was previously defined as 50 cases. This study compares the LC for LPD and subsequently RPD at a single institution. Methods: LPD and RPD were retrospectively analyzed. Continuous pathologic and perioperative metrics were compared and LCs were defined with respect to operative time using CUSUM analysis. Results: Thirty-nine LPD and 69 RPD were analyzed. Introduction of LPD was completed first as a hybrid technique (excluded), followed by total LPD. LPD had a bimodal or inverted LC possibly accounting for proficiency attained from hybrid LPD and introduction of additional surgeons later in the experience. The LC for RPD had three phases: accelerated early experience (cases 1-10), skill consolidation (cases 11-40), and improvement (cases 41-69), marked by reduction in operative time. Compared to LPD, RPD had shorter operative time (379 vs 406 minutes, p<0.03), improved LN harvest (15 vs 19, p<0.02), and similar estimated blood loss and R0 resection. RPD also had improved LOS (7 vs 9 days, p<0.002), and lower rates of surgical site infection (10% vs 38%, p<0.001), abscess (7% vs 23%, p<0.04), and readmission (13% vs 33%, p<0.03). Conclusion: Experience in LPD may shorten the learning curve for RPD. The gap in surgical quality and perioperative outcomes between RPD will likely widen as exposure to robotics in General Surgery, Hepatopancreaticobiliary, and Surgical Oncology training programs increase.