Learning curve analysis of colorectal endoscopic submucosal dissection (ESD) for laterally spreading tumors by endoscopists experienced in gastric ESD.

Abstract

Endoscopic submucosal dissection (ESD) for various colorectal neoplasms is more technically difficult than gastric ESD. We evaluated treatment outcomes and the learning curve for colorectal ESD of laterally spreading tumors (LSTs) based on the experience of a single endoscopist. We included 93 colorectal ESD procedures for colorectal LST that were performed between March 2009 and June 2012 by a single experienced endoscopist who previously performed hundreds of cases of gastric ESD. The cases were grouped chronologically into three periods by multi-dimensional analyses. For procedure time, the learning curve was analyzed using the moving average method, and for complication, the learning curve was analyzed using cumulative sum (cusum) method. The median procedure time for 93 colorectal ESD was 45min, and the rates of en bloc resection and R0 resection were 89.25 and 83.87%. When results were compared among three periods in order to determine the learning curve, the procedure time and en bloc resection rates were not significantly different. However, the procedure proficiency (about 0.16cm(2)/min) was significantly faster during the second period, after about 25 cases of colorectal ESD. In the third period (about 50 cases), the number and rate of en bloc resection (over 90%) reached the same as that of en bloc R0 resection. When comparing outcomes based on LST subtype, the procedure proficiency of LST-granular type (LST-G) was significantly faster than that of LST-non granular type (LST-NG) (LST-NG, 0.072cm(2)/min; LST-G, 0.157cm(2)/min; p=0.01). Endoscopists fully experienced in gastric ESD need a relatively short learning period for colorectal ESD in terms of procedure time and complication. However, approximately 50 cases might be needed to acquire an adequate skill of colorectal ESD for LST in an experienced gastric ESD endoscopist. Colorectal ESD for LST-NG seems to have higher technical difficulty and a longer learning curve than LST-G.