Can better surgical outcomes be obtained in the learning process of robotic rectal cancer surgery? A propensity score-matched comparison between learning phases

Background Although studies of robotic rectal cancer surgery have demonstrated the effects of learning on operation time, comparisons have failed to demonstrate differences in clinicopathological outcomes between unadjusted learning phases. This study aimed to investigate the learning curve of robotic rectal cancer surgery for clinicopathological outcomes and compare surgical outcomes between adjusted learning phases. Study design We enrolled 506 consecutive patients with rectal adenocarcinoma who underwent robotic resection by a single surgeon between 2007 and 2018. Risk-adjusted cumulative sum (RA-CUSUM) for surgical failure was used to analyze the learning curve. Surgical failure was defined as the occurrence of any of the following: conversion to open surgery, severe complications (Clavien-Dindo grade >= 3a), insufficient number of harvested lymph nodes (LNs), or R1 resection. Comparisons between learning phases analyzed by RA-CUSUM were performed before and after propensity score matching. Results In RA-CUSUM analysis, the learning curve was divided into two learning phases: phase 1 (1st-177th cases, n = 177) and phase 2 (178th-506th cases, n = 329). Before matching, patients in phase 2 had deeper tumor invasion and higher rates of positive LNs on pretreatment images and preoperative chemoradiotherapy. After matching, phase 1 (n = 150) and phase 2 (n = 150) patients exhibited similar clinical characteristics. Phase 2 patients had lower rates of surgical failure overall and these components: conversion to open surgery, severe complications, and insufficient harvested LNs. Conclusions For robotic rectal cancer surgery, surgical outcomes improved after the 177th case. Further studies by other robotic surgeons are required to validate our results.