Narrow-band imaging does not improve detection of colorectal polyps when compared to conventional colonoscopy: a randomized controlled trial and meta-analysis of published studies

Assignment to interventions

During the first part of the procedure, the colonoscope was inserted through the rectum and advanced to the large intestine using conventional colonoscopy in both groups (we did not use chromoendoscopy during the process of conventional colonoscopy observation). Thereafter, patients were randomly assigned to colonoscope withdrawal using either conventional wide-angle or NBI wide-angle colonoscopy (Olympus Corp: Olympus 180, CF-Q180AL#2) in examinations conducted by a total of three experienced examiners (each with over 5,000 colonoscopies performed and more than 15 years of experience, including a minimum of two years of experience with NBI colonoscopy). The colonoscopies were performed using high definition monitors. Appropriate and complete bowel preparation before colonoscopy was ensured using four liters of polyethylene glycol lavage until clear rectal fluid was evacuated and a cleaning enema. We categorized the quality of bowel preparation into excellent, good, fair, poor or inadequate.

Randomization was performed in blocks of 4 and 6 using a random table. Once the caecum had been reached and appropriate bowel preparation confirmed, an opaque sealed envelope with sequential numbering was opened and participants were allocated to either NBI or conventional colonoscopy withdrawal of the instrument.

Polypectomies were performed in the same session during withdrawal when possible. Polyps were removed using snare polypectomy or forceps biopsy, depending on the size of the polyps.

Outcome measures

Sample size determination

The primary endpoint for this study was the mean number of detected polyps. We assumed from institutional data that the mean number of polyps per patient in the control group would be 0.32, with a standard deviation (SD) of 0.31. A clinically significant increase in polyp detection using the NBI system was determined to be 25%. Using a two-tail alpha error of 0.05, and beta error of 0.20 (power > 80%), 240 patients in each arm would be required to detect a difference.

Data management and analysis

The database, created in Excel, was double-checked and transferred to SPSS 15.0^©. Categorical variables were compared using the chi-square test. To evaluate the continuous variables, Student's t-test was used; P < 0.05 was considered statistically significant. The measurement of the intervention effect for dichotomous outcomes was the risk ratio (RR). The measurement of the intervention effect for continuous outcomes was assessed by the mean difference (MD).

Systematic review of the current evidence

During the conduction of this study, a number of other RCTs comparing the NBI technique with conventional colonoscopy were published. We performed an advanced search strategy of studies comparing the conventional colonoscopy to NBI to detect colorectal polyps/adenomas (appendix 1). Relevant RCTs were identified from the Cochrane Library (2009, Issue 4), PUBMED (1966-December 2009), LILACS (1982-December 2009) and Scirus (http://www.scirus.com; December 2009). We also scanned bibliographies of relevant studies for possible references to additional RCT. Two authors independently decided which trials fit the inclusion criteria. Eligible RCT were included regardless of the language of publication. Two reviewers independently extracted the relevant data using a pre-designed data extraction form and any disagreement was resolved by consensus with all authors. We extracted year of publication; patient population; number of patients (by intention to treat); sociodemographics; endoscopic, and histologic outcomes; and adverse effects. The main outcomes considered were the mean number of polyps, the mean number of adenomas and the rate of patients with at least one adenoma.

A risk of bias evaluation of each RCT was done following the Cochrane Collaboration's tool for the assessment of these features [35]. To estimate differences between treatments we calculated a weighted treatment effect across RCTs. We expressed the results as risk ratio (RR) with 95% confidence intervals (CI) for dichotomous outcomes and weighted mean difference (WMD) with 95% CI for continuous outcomes. We imputed conservative standard deviations where necessary using the p-value from an independent two-sample t-test [35]. For the pooled analysis, we calculated the I2 statistic, which describes the percentage of total variation across studies caused by heterogeneity [35].

Randomized Controlled Trial

Patients were enrolled during a three month period at the Reina Sofia Clinic and during a 2 month period at the Clinica Colombia, between September 2008 and May 2009. We included a total of 482 patients (62.5% female), with a mean age of 58.33 years (SD 12.91); 241 into the intervention (NBI) colonoscopy and 241 into the conventional colonoscopy group. Patients were enrolled during a three month period at the Reina Sofia Clinic and during a 2 month period at the Clinica Colombia.

The flow of participants through each stage of the randomized trial is described in Figure 1. Most polyps (67%) were found in the left colon; no significant difference was found between groups with regards to the location of the polyps. The total examination time did not differ significantly between the two groups (9.21 vs. 9.22; excluding polypectomy duration). Baseline characteristics of patients were similar among groups (Table 1).

Of the adenomas, 9% were high-grade, with no significant difference between groups. Two adenocarcinomas were found (one in each group). There were no differences between the different examiners in the rate of detection of adenomatous and hyperplastic polyps. No serious adverse events were reported among groups during the procedure.

Systematic Review of the current evidence

A total of 167 citations were identified from the diverse sources of information (Figure 2). Of the sixteen potentially RCTs screened [19–34], we excluded nine references because they were nonrandomized, they focused on predicting colon polyp histology, or they used other devices. Finally, seven RCTs meet the inclusion criteria [19–21, 23, 25–27] one of which was published as an abstract [27]. Characteristics of RCTs included in the meta-analysis are described in Table 4. One study was judged as having low risk of bias [21] and six RCTs were judged as having unclear risk of bias because the description of the method used to generate to conceal the allocation was unclear [19, 20, 23, 27]; only one evaluator performed all the colonoscopies [25]; inadequate distribution of the NBI procedure among all participating endoscopists [26]; and a possible learning effect from the NBI during the initial phase of the trial [19]. Only in one RCT each patient underwent back-to-back colonoscopy [21]. The primary outcomes of interest were frequently reported incompletely and we had to impute standard deviations in four studies [19, 21, 26, 27].

A meta-analysis of studies (including diagnostic, surveillance and/or screening colonoscopies) is showed in Table 5. No significant differences were found among groups in the mean number of polyps, the mean number of adenomas (Figure 3), and the rates of patients with al least one polyp or one adenoma (Figure 4). We performed a sensitive analysis comparing those studies that used Lucera or Exera II systems. Significant differences favoring the NBI system in the mean number of polyps (2 RCT, 457 participants; WMD -0.39 95% CI -0.62 to - 0.16; I2 0%) and the mean number of adenomas (2 RCT, 457 participants; WMD -0.22 95% CI -0.41 to - 0.04; I2 2%) were found when pooling data from RCTs that used the Lucera system [26, 27].

Study (Year)	Outcome	Conventional colonoscopy (N)	NBI system (N)	RR* WMD** (95%CI)	Heterogeneity (I2) §
Adler 2009 East 2009 Inoue 2008 Kaltenbach 2008 Sabbagh 2011	Mean number of polyps	1241	1238	WMD -0.07 (-0.21 to 0.07)	68%
Adler 2009 Adler 2008 East 2009 Inoue 2008 Kaltenbach 2008 Paggi 2009 Rex 2007 Sabbagh 2011	Mean number of adenomas	1768	1749	WMD -0.08 (-0.17 to 0.01)	62%
Adler 2009 Adler 2008 Inoue 2008 Kaltenbach 2008 Sabbagh 2011	Rate of patients with al least one polyp	1336	1330	RR 0.90 (0.73 to1.11)	70%
Adler 2009 Adler 2008 East 2009 Inoue 2008 Kaltenbach 2008 Paggi 2009 Rex 2007 Sabbagh 2011	Rate of patients with at least one adenoma	1763	1749	RR 0.96 (0,88 to 1.04)	0%
Adler 2009 Adler 2008 Paggi 2009 Sabbagh 2011	Rate of patients with carcinoma	1181	1169	RR 1.20 (0.58 to 2.47)	0%

*Relative Risk and 95% confidence intervals (CI) for dichotomous outcomes were calculated by the Mantel-Haenszel fixed-effects model when I2 < 50%. Relative Risk and 95% confidence intervals (CI) for dichotomous outcomes were calculated by the Mantel-Haenszel random-effects model when I2 > 50%.

**Mean difference and 95% confidence intervals (CI) for continuous outcomes were calculated using the inverse variance and the fixed-effects model when I2 < 50%. Mean difference and 95% confidence intervals (CI) for continuous outcomes were calculated using the inverse variance and the random-effects model when I2 > 50%.

§ Low, moderate, and high levels of heterogeneity approximately correspond to I2 values of 25%, 50% and 75%, respectively [35].

No serious adverse events were reported. Taken into account that we only found eight RCTs, funnel plots for assessing publication bias was not performed.

Main findings

According to our findings, the polyp detection rate per patient by visual inspection was significantly higher in the conventional colonoscopy group compared to the NBI group. In addition, significant differences favoring the conventional colonoscopy group were also found for some types of histological examination polyps (hyperplastic polyps and tubulovillous adenoma). However, the adenoma detection rate was similar in both groups.

Although meta-analysis of RCTs showed no significant difference for pre-specified primary outcomes, individual studies reported diverse findings [19–28]. Statistical heterogeneity may be explained by difference in the prevalence of polyps and adenomas in the population, the indication for colonoscopy (screening, surveillance and/or diagnostic), the age of the included population, and the examiner's experience among others. Findings of our review update those of a previous systematic review that included three RCTs concerning the detection of colorectal adenomas [36].

Only two RCTs found a significant difference in the mean rate of adenomas favouring the NBI group [21, 26]. One RCT, which included 243 patients, found a significant difference in the rate of adenoma detection favoring the NBI group (22% vs. 14%), including the subgroups of patients having polyps measuring less than 5 mm. The authors of the study recommended the routine use of the NBI system for surveillance of diminutive adenomas [26]. Another study [21] found that the NBI system significantly increased the total number of adenomas detected as well as the number of diminutive adenomas in the distal colon [21]; however the rate of missed lesions between the NBI and conventional group was similar. Another RCT found that the number of diminutive (< 5 mm) adenomas was significantly higher in the NBI group [26]. One RCT reported a significantly higher detection in the mean number of flat adenomas in the NBI group [27], one study reported the opposite [20] and two RCTs did not find any significant difference [23, 25].

In our study, colonoscopies were performed by experienced examiners in both techniques and included diagnostic and screening colonoscopies from two different institutions. Adenoma rates in larger colonoscopy trials vary widely. The overall detection rate of polyps and adenomas by histological examination were 35.1% and 22% respectively, which is similar to the rates reported in other studies [20, 37–40]. However some RCTs found higher rates of polyps and adenomas; the difference can be explained by heterogeneous included population. The lower detection rate in our study may be due to the withdrawal time but also to the lower prevalence in our population. A number of published studies have evaluated the prevalence of polyps and adenomas in Colombia. Overall, the prevalence of colonic adenomas is lower when compared with rates reported in other regions [41–44]. In addition, around one third of patients in our RCT had already had a colonoscopy. Although differences in polyp frequency between screening and diagnostic colonoscopy have been reported, some studies have found similar rates [40]. Additionally, no differences were found in primary outcomes between colonoscopies performed in the two different clinics, or when different periods of time were compared (first 241 colonoscopies vs. last 241 colonoscopies).

Limitations of the study

The findings of this RCT have some limitations, mostly due to the lack of tandem colonoscopy in both groups. Difference in the overall rate of polyps could have been due to selection bias (patients with more polyps could have been included in the conventional group by chance). However, we found no significant differences among participants in the baseline characteristics and the colonoscopy performance; we included a significant number of patients and we concealed the allocation of patients to minimize any possible bias. Our data shows that the white-light group had 20.8% higher detection rate of adenomas than did the NBI group. As both groups had similar withdrawal time, the white-light group could have had better mucosal visualization during the withdrawal phase compared to that of the NBI group (the darkening of the image associated with the use of NBI). This may have lead to the finding of significantly greater number of polyps found in the white-light group.

In addition, the fact that bowel preparation was not excellent in one third of patients may have contributed to the poorer performance of the NBI visualization. As screening colonoscopy is not usually recommended in our country in people above 50 years of age with average risk, we had an important proportion of diagnostic colonoscopies.

Concerning the review, we pooled data from studies that included heterogeneous populations and indications. The use of varied endoscopic systems as well as differences in colon preparation of participants between studies may have had some impact on findings. Uraoka et al noticed that significant differences in the detection of adenomas where related to the type of endoscopic video system (either the sequential LUCERA series or the simultaneous EXERA-II series). They found that most positive studies used the LUCERA system while all of the negative studies used the EXERA-II system [45]. Although pooled estimates from two RCTs [26, 27] support the use of LUCERA series, both studies where judged as having unclear risk of bias. More research is still needed to determine the efficacy of different NBI system settings for screening and surveillance colonoscopies, particularly to enhance the detection rate for flat adenomatous lesions [45].

Finally, there were differences in the report of flat or non-polypoid type neoplasm among studies which did not permit further pooled analysis.