We made a preliminary search of published reports using the following terms: ([narrow band imaging OR NBI] AND [(adenomatous polyps OR adenoma OR polyps) AND (colon OR rectum OR colorectal OR colorectum OR colon rectum)]) OR (adenomatous polyps [MeSH major topic]) from the Cochrane Center Register of Controlled Trials, PubMed/MEDLINE, EMBASE, and Chinese Biomedicine Literature Database. In addition, we performed a manual review of references, including all research articles and reviews of interest on this topic for the identification of additional relevant studies. All research articles were reviewed independently and crosschecked by three investigators (X-F. J., T-H.C., and J-W.S.). Disagreements among them were resolved by discussion and consensus.
Individual reports complying with the following inclusion criteria were included in the meta-analysis. There was no restriction on any study and limitation for beginning data. The search was updated until the end of June 2011. The inclusion criteria were: (i) surveillance colonoscopy of known colorectal adenomas or cancer, or screening colonoscopy in patients with a positive test of fecal blood; (ii) diagnostic colonoscopy in patients with symptoms of rectal bleeding, abdominal pain, and changes in bowel habits, diarrhea, and iron-deficiency anemia; (iii) reports with the raw data on the comparison of NBI with white light endoscopy concerning one or more of the following parameters: patients with at least one adenoma in both groups, patients with at least one flat adenoma in both groups, the number of adenomas per patient, the number of flat adenomas per patient, and mean withdrawal time; (iv) written in English; (v) published as a full article or abstract; and (vi) study with high homogeneous populations, but without another disease. In addition, all studies should exclude patients with known familial adenomatous polyps, inflammatory bowel disease, known colorectal cancer, or inadequate bowel preparation. When multiple reports came from the same population or subpopulation, the most recent or most informative report was included. When two articles were reported from the same study, the more informative publication was selected.
Quality Assessment of Studies
The quality of items was assessed by randomization, allocation concealment, blinding of outcome assessment, withdrawals, and dropout analysis.
Data were analyzed using the RevMan analysis statistical program in the Review Manager. Polled results (with 95% confidence intervals) were analyzed using the fixed effect model (the Mantel–Haenszel method), but the data with significant heterogeneity were analyzed using the random effects model (DerSimonian and Laird method Relative risk (RR) was used as a summary statistic for the statistical analysis of dichotomous variables. Continuous variables were processed using the values of mean and standard deviation. Heterogeneity was tested using the Cochran Q-test, and a P-value of < 0.1 was considered statistically significant.
The data with significant statistical heterogeneity were analyzed by the random effect model, and subsequently, by a sensitivity analysis to evaluate the consistency of the results. We first examined whether the exclusion of this study substantially could alter the magnitude of heterogeneity of the summary estimate by repeating the analyses with exclusion of individual studies (one study per analysis). We further performed stratification analyses, according to factors that could potentially influence the result.
Publication bias was assessed by constructing funnel plots, which were obtained by plotting the log RR versus precision (1/SE) of individual studies. Their symmetry was estimated by Egger's regression asymmetry test (evidence of asymmetry based on P < 0.1).