Study population, data collection and study outcomes in one tertiary setting
The data from patients who underwent ESD for colorectal neoplasms (CRNs) in a tertiary referral center between January 2015 and April 2017 were retrieved and analyzed. Patients were excluded in terms of the following criteria: 1) Postoperative histopathology evaluation showed the submucosal penetration was deeper than 1000 µm from the muscularis mucosa. 2) Not suggestive of precancerous lesions or adenocarcinomas (neuroendocrine tumor, lipoma, gastrointestinal stroma tumors, et al) based on histopathology. 3) Locally residual or recurrence colorectal lesions on site after endoscopic treatment.
Data related to demographics of patients, lesions, procedures, and adverse events were collected. All ESD cases were categorized as en bloc or non-en bloc resection group. The ESD procedure was classified as technically difficult in the case of the non-en bloc resection.
Meta-Analysis
A systematic review was conducted using meta-analytical approaches outlined in the Cochrane Handbook for Systematic Reviewers and reported according to the Preferred Reporting Items for Systematic Reviews and MetaAnalyses (PRISMA) criteria.
Literature search strategy and selection criteria
A comprehensive literature search was performed through database such as Pubmed, EMBASE, Cochrane library and large bibliographic database in China such as Wanfang MedOnline, China National Knowledge Infrastructure (CNKI) and China Biology Medicine disc (CBMdisc) through 1 June 2019. The subject terms were used: “Endoscopic submucosal dissection,” “ESD,” “Difficult ESD”, “difficult endoscopic submucosal dissection” and “colorectal neoplasm.” An additional search was performed among references of included studies to find potentially eligible studies.
Inclusion criteria were as follows: 1) no limitations on study design, including prospective or retrospective observational cohorts, case–control studies, and randomized controlled trials (RCTs), and 2) studies reporting the therapeutic outcomes of ESD for colorectal neoplasm lesions, such as en bloc resection rate, R0 resection rate, and follow-up results, including follow-up period, recurrence rate, and rescued surgery, etc.
Exclusion criteria were as follows: 1) case reports, 2) literature on ESD for other diseases, such as esophageal, gastric lesions. 3) experiments on animals, 4) reviews, comments, or letters, 5) studies published in other languages besides English and Chinese, 6) studies with unavailable full text. W.J. and Y.G. evaluated the quality of the included studies using the Newcastle–Ottawa scale.
Data extraction
Data collection objectives included the following items. 1) Study design, duration, country, setting, Authors, year of publication; 2) Preoperative factors: patient demographics (such as age, gender distribution, previous abdominal surgeries, diagnosis of ulcerative colitis), location (colon vs rectum), morphology and tumor size, flexure, special anatomic regions such as Bauhin’s valve/dentate line and the experience of endoscopists; 3) Intraoperative factors: submucosal fibrosis, non-lifting sign. For comparison with data in our center, we only retrieved these two intraoperative factors. 4) The prognostic outcomes were also involved as data collection objectives. The factors reported in more than three literatures would be recruited to meta-analysis. As some factors (the cutoff of age and tumor size, morphology, location) may be classified differently among references, for the convenience of pooling these estimates, we included the values as follows. That is, we categorized tumor size into “≥40mm” and “<40mm”. The morphology was classified as “Laterally spreading tumors” (LST) and “Protruding”. The cutoff of age is 70-year-old. The location consists of colon and rectum. The follow-up data such as period, perforation rate, local recurrence rate and rescued surgery rate would be extracted from references reporting the failure of en-bloc resection or R0 resection as the prognostic outcome.
Definition of outcomes
The primary outcome was to evaluate the preoperative and intraoperative predictors of the non-en bloc resection or R0 resection for colorectal neoplastic lesions. Inconsistent with the data in our center, we have also enrolled references which reported “R0 resection” as the primary outcome for retrieving data as much as possible. Secondary prognostic outcome was follow-up outcomes in non-en bloc resection group and en-bloc resection group: 1) Local recurrence was defined as the cancer has recurred on the primary resection site during the follow-up period. 2) Rescued surgery referred that the lesion cannot be treated under endoscopy, so that they underwent open surgery or laparoscopy for resection.
Statistical analysis
For retrospective analysis, continuous parameters were analyzed using Student’s t-test or the Mann-Whitney U test, whereas categorical variables were compared using the χ2 test and Fisher’s exact test, as appropriate. Values of p < 0.05 were considered statistically significant. Risk factors for the non-en bloc resection were analyzed using univariate method. A logistic regression model was used for the multivariate analysis of significant factors detected by univariate analysis, defined as p < 0.05, with forward stepwise selection. SPSS version 25 for Mac (IBM Corp.; Armonk, NY, USA) was used for the statistical analysis.
For meta-analysis, all pooled odds ratio (OR) and 95% confidence intervals (CIs) of risk factors were calculated and compared using Review Manager (RevMan version 5.3). χ2 test and I2 statistics were used to evaluate heterogeneity. It was considered to have minor heterogeneity if I2 ≤ 25%; if 25% < I2 ≤ 75%, it was considered to have moderate heterogeneity; if I2 > 75% and p < 0.1, it was considered to be significant heterogeneity; When there was no statistically significant heterogeneity among the studies (I2 < 75%, P > 0.1), a fixed-effects model was used to calculate the pooled estimates, otherwise a random effects model was used. Multiple sensitivity analyses were conducted to explore the potential heterogeneity and the possible causes. For potential publication bias, a comprehensive literature search and removal of the duplicated data were conducted to minimize the reporting biases and funnel plots were also constructed to explore the publication bias.