Systematic Review With Network Meta-Analysis

Endoscopic Techniques for Dysplasia Surveillance in Inflammatory Bowel Disease

Andrea Iannone; Marinella Ruospo; Suetonia C. Palmer; Mariabeatrice Principi; Michele Barone; Alfredo Di Leo; Giovanni F. M. Strippoli


Aliment Pharmacol Ther. 2019;50(8):858-871. 

In This Article

Materials and Methods

This systematic review was reported according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Network Meta-Analyses (PRISMA-NMA) guidelines.[16]

Data Sources and Searches

We searched for randomised trials in MEDLINE (1996 to May 2019), Embase (1996 to May 2019) and the Cochrane Central Register of Controlled Trials (CENTRAL; from 1996 to May 2019). We did not limit searching by publication date or language. An information specialist with expertise in systematic reviews of randomised trials designed the search strategy (Table S1). We also searched relevant trials from reference lists of all identified trials, guidelines and reviews on the topic.

Study Selection

Two reviewers (AI and MR) independently screened the retrieved search records by title and abstract. Potentially eligible citations were reviewed by the same two reviewers in full text. Disagreements were resolved through consensus and discussion with a third reviewer (SCP).

We included randomised and quasi-randomised controlled trials comparing any endoscopic technique (standard definition white-light endoscopy, high definition white-light endoscopy, chromoendoscopy, narrow band imaging, i-SCAN, autofluorescence, Fujinon intelligent colour enhancement [FICE] and full-spectrum endoscopy) for dysplasia surveillance in adults with ulcerative colitis and Crohn's disease. We included any study enrolling people with IBD or with concurrent primary sclerosing cholangitis, a condition at high risk of colorectal cancer that requires dysplasia surveillance following the diagnosis of biliary disease.[10] We excluded trials evaluating colorectal cancer surveillance in the general population or hereditary polyposis syndromes.

Data Extraction and Quality Assessment

Two reviewers (AI and MR) independently extracted data on characteristics of study, population, interventions and outcomes from included randomised trials using an electronic database. In case of randomised crossover trials, we extracted data regarding only the first phase of the study to avoid a carry-over effect. Any disagreements in data extraction were resolved through consensus and discussion with a third reviewer (SCP).

Outcomes. The key outcomes in this review were number of participants with one or more neoplastic lesions, number of participants with any type of lesion (including neoplastic and non-neoplastic lesions), number of neoplastic lesions detected by target biopsy, procedural time and any adverse events associated with the diagnostic procedure. According to the Vienna classification, we considered as neoplastic lesions those showing low-grade dysplasia, high-grade dysplasia or invasive neoplasia at histological examination.[17] Non-neoplastic lesions included endoscopic findings with no evidence of dysplasia or invasive neoplasia at histology. In the analysis of number of neoplastic lesions detected by target biopsy, we included only trials in which both target and random biopsy specimens were taken in all intervention arms and we calculated this outcome as the number of lesions detected by target biopsy divided by the total number of neoplastic lesions identified.

Additional outcomes included all-cause mortality, colorectal cancer-related mortality, interval colorectal cancer and health-related quality of life.

We also evaluated the number of morphological and histological subtypes of neoplastic lesions identified using different endoscopic techniques. According to the SCENIC consensus, we considered the morphological classification of neoplasia in polypoid (including sessile and pedunculated lesions) and nonpolypoid (including elevated, flat and depressed lesions).[13] We applied the histological classification of neoplastic lesions in low-grade dysplasia, high-grade dysplasia or invasive neoplasia, according to the Vienna classification.[17] We calculated the number of nonpolypoid neoplastic lesions, low-grade dysplastic lesions and high-grade dysplastic lesions as the number of each morphological or histological subtype of neoplastic lesions divided by the total number of neoplastic lesions identified.

If published outcome data were not reported or provided in sufficient detail in included trials, an author (AI) contacted the investigators up to three times by email to request any relevant additional information.

Risk of Bias Assessment. Two independent reviewers (AI and MR) assessed the risks of bias using the Cochrane tool.[18] Disagreements were resolved through consensus and discussion with a third reviewer (SCP). We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach for network meta-analysis to critically appraise the certainty of evidence for all evaluated outcomes.[19] For this purpose, we used the Confidence in Network Meta-Analysis (CINeMA) web application.[20]

Statistical Analysis

First, we performed random-effects pairwise meta-analysis. We assessed heterogeneity between trials in pairwise meta-analysis using the Chi-squared test and I2 statistics. Then, we used a frequentist framework, random-effects network meta-analysis to compare all endoscopic techniques for each pre-specified outcome.[21,22] We calculated treatment estimates for binary outcomes as odds ratios (ORs) and continuous outcomes as mean differences (MDs), together with their 95% CI.

We estimated the extent of heterogeneity in each network analysis using the restricted maximum likelihood method to generate a common heterogeneity variance (tau [τ]), which was compared with an empirical distribution of heterogeneity variances.[23] To explore for network inconsistency, we used a loop-specific approach, which compared the estimated effects derived from direct and indirect evidence in all triangular and quadratic loops in a network.[24] We used the design-by-treatment interaction approach to check the assumption of consistency in the entire analytical network.[25] We ranked endoscopic techniques to generate a hierarchy for each endpoint. The relative ranking probability of each endoscopic procedure being among the best technique was obtained using surface under the cumulative ranking (SUCRA) curves and displayed using rankograms. We planned to explore publication bias with funnel plots in which 10 or more studies were available.[26]

We performed pairwise and network meta-analyses in Stata (StataCorp LP), using the network command[27] and self-programmed Stata routines.[28]