Colonoscopy is still regarded as the standard diagnostic test for detecting colonic neoplasia, especially when the lesions are discrete and in the early stages. It has become the procedure of choice for colorectal cancer screening, as it also makes it possible to carry out polypectomy, which is considered to interrupt the adenoma–carcinoma sequence.1^–3 Previous studies have, however, shown that there is a polyp miss rate of 10–20% during back-to-back colonoscopies,4 5 and reports of missed cancers being identified after colonoscopy have recently been increasing.6 7 The reasons for the miss rate may be technical (incomplete colonoscopy or insufficient attentiveness during excessively fast instrument withdrawal), but missed lesions may also be caused by the imaging method. Smaller lesions, particularly flat ones, may be missed as a result of their subtle appearance and limited contrast in relation to the surrounding mucosa. If screening colonoscopy is to achieve its preventive purpose, every effort needs to be made to help the endoscopist detect all adenomas, including smaller and subtle lesions, with the greatest possible accuracy.

Techniques for reducing the adenoma miss rate have been tested to some extent in randomised studies, with methods including wide-angle colonoscopes,8 the use of transparent caps mounted on the colonoscope tip,9 and imaging techniques for better highlighting of lesions—simple staining has been shown to improve neoplasia detection in inflammatory bowel disease,10 but less consistent results have been reported for diagnostic colonoscopy for standard indications.11 12 Staining of the entire colon during standard colonoscopy is, however, usually considered to be too time-consuming and therefore unsuitable for routine application.

The new technique involving the incorporation of narrow-band imaging (NBI) facilities into conventional video colonoscopes uses special filters to narrow the bandwidth of light components, eliminating red and rendering vascular structures in black.13 NBI leads to the enhancement of structures and may therefore improve visualisation. The aim of the present study was to test the hypothesis that NBI can improve adenoma detection in comparison with standard video colonoscopy.

PATIENTS AND METHODS

During a 9-month study period, 401 eligible patients (52.6% male) with a mean age of 59.4 years (SD 13.4) were included in the study, depending on the availability of the study instruments (only one NBI and one non-NBI wide-angle colonoscope were available). Patients are scheduled at 45 minute intervals for colonoscopy in our unit, and were selected for informed consent only when otherwise eligible and both instruments were available within an acceptable time.

Patients were then randomly assigned to colonoscope withdrawal using either conventional (n = 201) or NBI wide-angle colonoscopy (n = 200; Olympus Corp., Hamburg, Germany) in examinations conducted by a total of seven experienced examiners (each with over 500 colonoscopies performed). Patients were selected if they presented for diagnostic colonoscopy for a variety of indications (eg positive fecal occult blood test, abdominal pain, diarrhoea, post-polypectomy surveillance). Patients were excluded if they had known colonic neoplasia or inflammatory or other significant colonic disease, such as patients specifically presenting for polypectomy. They were also excluded if there was open bleeding or they were receiving an emergency colonoscopy, if they had previously undergone colonic surgery, if they were in poor general condition (more than American Society of Anesthesiologists grade III), or if they were receiving anticoagulant medication. Randomisation lists were used for group allocation, and the study was approved by the university’s ethics committee (EA2/188/05).

Bowel preparation consisted of polyethylene glycol lavage with 4–6 litres until clear rectal fluid was evacuated. Special care was taken to wash and clean the entire colon during instrument introduction and withdrawal, in order to provide optimal conditions. Patients in whom this was deemed impossible as a result of poor bowel preparation were excluded from the analysis.

The following parameters were documented:

• age and sex of the patient

• indication for colonoscopy

• type and dosage of sedation

• examination time, separately for instrument introduction and withdrawal

• polyp characteristics: size (measured by open forceps or snare), shape (pedunculated, sessile, flat, the latter defined as maximal height of 1.3 mm13), and location

• polyp histology after removal using snare polypectomy or forceps removal (for polyps <3 cm), or biopsy if there were contraindications

• other lesions found.

Assessment of the colon to search for polyps and other lesions started when the caecum was reached, and the NBI function was activated throughout the withdrawal of the instrument in the NBI group. Switching back to conventional imaging was allowed in the NBI group, and the number of occasions for switching back and the reasons for it were documented in this group. Each examiner had carried out five specific training examinations in order to become acquainted with the NBI method, which was generally considered to be easy. Experience with NBI colonoscopy was available for a few months before the study start, so that a general impression of the technique could be gained.

The main outcome parameter was the adenoma detection rate in the two groups. This also determined the case number calculation, assuming an adenoma rate of 10% on the basis of a recent study among Berlin private practice gastroenterologists14 and an increase to 20% with NBI (80% power, significance level 0.05); 200 patients needed to be enrolled in each group.

Secondary outcome measures included:

• total number of polyps in both groups

• total number of flat/sessile adenomas in both groups

• total number of adenomas less than 1 cm in both groups

• total number of hyperplastic polyps and hyperplastic polyps less than 1 cm in both groups

• right-sided versus left-sided location in both groups.

Data analysis

Continuous variables were compared using the t-test if normally distributed and the Mann–Whitney test if not normally distributed. Categorical variables were compared using the chi-square test or Fisher’s exact test when appropriate. The chi-square test was used to analyse the main outcome measure—the difference in the proportion of patients with adenomas. The main outcome was examined for potential confounding variables using a logistic regression analysis. To compare the detection of all adenomas and hyperplastic polyps (per-polyp analysis), the Poisson regression model was used. Test of trend was applied to examine differences in detection over time.

RESULTS

A total of 401 patients (52.6% male), with a mean age of 59.4 years (SD 13.4) were enrolled. Of the colonoscopies, 45.9% were inpatient procedures and 54.1% were outpatient procedures. Sedation was used in 87.0% of cases (NBI 174, controls 175) and consisted of midazolam (mean dose 4.6 mg, range 2.0–10.0 mg) in 72% and propofol (mean dose 98.7 mg, range 40–460 mg) in 15% (59/401) of the total group. The caecum was reached in 98% of cases (392/401); failures were caused by stenosing cancers (n = 2), poor bowel preparation (n = 6), and failure to pass beyond the hepatic flexure (n = 1), which was later achieved during a second examination with propofol and did not form part of the study. The total examination time (mean 22.5 minutes, SD 11.1) did not differ significantly between the two groups. Table 1 shows in detail that there were no significant differences in any of the parameters assessed.

Five patients were excluded secondarily during the study audit as a result of violation of the protocol (four with known cancers or polyps who were sent for polypectomy, and one because of advanced ovarian malignancy with possible colonic involvement). Five carcinomas were found (three in the NBI group and two in the control group), but these were not included in the further analysis. None of the patients was excluded as a result of failure to reach the caecum.

The overall polyp detection rate (polyps of all histology) per patient in the entire study group was 33.6% (n = 133), with adenomas and hyperplastic polyps found in 19.9% (n = 79) and 19.7% (n = 78) of all patients, respectively. Of the adenomas, 99% were low grade and there was one adenoma with high-grade intraepithelial neoplasia in the NBI group.

Detailed results of the study are shown in table 2. Analysed per patient—ie, the rates of patients with polyps—the differences between NBI and controls were 41.4% versus 25.8% (p<0.001) for all polyps, 22.7% versus 16.7% (p = 0.129) for adenomas, and 28.3% versus 11.6% (p<0.0001) for hyperplastic polyps. The aim in colonoscopy is to detect and remove all (adenomatous) polyps, and therefore an analysis was also carried out on a per-polyp basis (table 3).

In a number-needed-to-diagnose analysis, the number of NBI colonoscopies needed to find one additional adenoma patient would be 17 (with adenoma detection on a per-patient basis of 17% versus 23%).

There were no differences between the different examiners in the rate of detection of adenomatous and hyperplastic polyps. Nor were there any differences in the polyp detection rate between the two groups (both adenomas and hyperplastic polyps) when analysed for subgroups in relation to polyp size, form, and localisation (see table 3). Adjustment for age and gender had no effect on the results either.

In the NBI group, a mean of 1.7 switchbacks to conventional imaging was recorded (minimum 0, maximum 15, median 1, range 1–2), and the main reason for switching was to confirm the presence or absence of lesions with conventional imaging. Generally, switchbacks were more frequent in the initial phase (for the first and last hundred cases, the mean values were 2.3 versus 1.2, p<0.001). This can be accounted for as representing a learning effect.

The influence of experience with NBI examination on the detection of lesions was secondarily evaluated by examining the change in the rate of detection of patients with adenomas with increasing numbers of NBI endoscopies carried out, during the first, second, third, and final hundred endoscopic examinations. To take into account the smaller numbers of endoscopies conducted by some examiners, the proportions of patients with lesions seen in the first 25 and the second 25 of their endoscopies were also calculated. Possible changes were evaluated with a test of trend.

We found that the adenoma detection rates differed significantly in the first 100 patients (NBI 26.5%, control group 7.8%; p = 0.02), but not in the last 100 cases (26.5% versus 25.5%; p = 0.91). This is illustrated in fig 1 for all patients as well as for examiner subgroups. Subanalysis did not show that small adenomas (<10 mm) or sessile/flat adenomas were more frequently detected with increasing NBI experience in the non-NBI group, probably because of limited numbers in the subgroups. The number of hyperplastic polyps showed a similar trend, but not as pronounced (37.6% versus 9.8% in the first 100 cases, 34.0% versus 14.3% in the last 100 cases; p = 0.563). Withdrawal times were not statistically significant between the groups of 100 cases each (mean of 8.8, 10.5, 10.3 and 9.2 minutes for cases without polypectomy; p = 0.204)

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 (A) Overview of adenoma rates per group of 100 study patients each. (B) The adenoma rates for all examiners with more than 30 examinations during the first and second half of their study cases. NBI, Narrow-band imaging.

Examples of polyps imaged conventionally and with NBI are shown in fig 2 and fig 3.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 A flat adenoma visualised with narrow-band imaging.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 (A) A flat adenoma visualised with narrow-band imaging. (B) The same lesion as seen with subsequent conventional imaging.

DISCUSSION

The study showed that the new technique of NBI detected more colonic adenomas in comparison with conventional high-resolution video endoscopy without NBI. Although 36% more adenomas were found by NBI, the difference was not statistically significant for the total study group. Analysed on a number-needed-to-diagnose basis, 17 colonoscopies would be required to detect one additional adenoma patient by means of NBI. In the search for an explanation of this fact, consecutive groups of 100 patients each were analysed secondarily. In the NBI arm, adenoma rates were mostly approximately 25%. In the control group without NBI, however, the adenoma detection rate rose continuously, from 8% to 15%, 17%, and 26.5% in each consecutive hundred patients, and this trend was also observed for all of the examiners participating. It can only be speculated that this increase might be the result of some form of learning effect resulting from the NBI contrast-enhancement technique. We hypothesised that improved recognition of lesions on NBI may also subsequently have improved the examiners’ ability to perceive such lesions on conventional high-resolution colonoscopy, although the study was not set out and powered to test a possible learning effect. Interestingly enough, an ongoing UK study observed a similar trend. In the first interim analysis on the basis of 52 cases, NBI detected 61% more adenomas (42 versus 26), whereas in a later analysis with 91 patients, this increase was only 45% (96% versus 66%).14 15 Among other explanations, a variable motivation of examiners (eg better motivation with the NBI application) cannot fully be excluded. Several examiners were, however, involved and such an effect represents an inherent potential bias in any study dealing with a comparison of new and conventional tests. We have also previously published negative results with imaging tests.16 17

Another issue to be discussed from our study is a low adenoma detection rate at least in the control group, and especially in the initial study cases. Adenoma rates in the two other randomised studies on NBI colonoscopy, namely from Indianapolis18 and London,15 were much higher, with some 60–65% of patients in the control group being diagnosed with adenomas. Other large-scale studies from the United States also reported high adenoma rates, in the range of 37.5%.19 In general, however, adenoma rates in larger colonoscopy trials vary widely. In other studies from the United States, adenoma rates were much lower, with 23.5% (n = 2053),20 or even as low as 14.5% in another trial from New York (n = 4043).21 A large Polish screening study including 43 042 cases reported a neoplasia detection rate of 9.4%.22 As for studies from Germany, a recent study from 39 private practices in Berlin found a polyp rate of 15% (>6 mm, no histological specification).23 Two other German trials reporting on screening colonoscopies had neoplasia rates of either 12.4% (n = 1117)24 or 20.7% (n = 109 989).25 The adenoma rate reported in our study (17% in the control group) is thus well within the limits of the other trials. The setting (screening versus diagnostic colonoscopy) may also have an influence on the results; however, we recently found no differences in polyp frequency between screening and diagnostic colonoscopy in the private practice multicentre study cited above.23

Other methodological factors should also be discussed. Wide-angle colonoscopes may contribute to better detection, but these were used in both groups and have previously been shown not to increase adenoma detection rates.8 A clean colon is essential for proper visualisation of the entire colonic mucosa in order not to miss subtle and flat lesions. Data from the recent UK sigmoidoscopy trial have indeed shown a good correlation between colon cleanliness and the adenoma detection rate.26 There were no differences between the groups in this respect, because all residual fecal material was removed by flushing in all study cases if possible, and the remaining cases were excluded.

Study methodology, patient population, and adenoma prevalence are all factors that may explain the differences between different studies. Only a prospective randomised study is capable of detecting differences in adenoma rates. It is debatable whether a tandem methodology (double colonoscopy with and without the new technique being assessed during the second withdrawal phase) or simple randomisation between the two methods is preferable. A recent randomised tandem colonoscopy trial found more high-grade neoplasias in the staining group, but this was as a result of detection during the first withdrawal, which (in both groups) consisted of conventional colonoscopy.11 In the present study the simpler design of comparing the two methods was chosen, similarly to others,15 18 and larger numbers of patients were included in the trial to compensate for the possible general miss rates caused by performing one instead of two colonoscopies, as is the case with the tandem approach.

The level of experience of the examiners is another crucial issue, and this has only recently been systematically investigated. A large community-based study recently showed that adenoma detection rates decreased below a cutoff point of six minutes of withdrawal time.20 In another study, an offshoot from the UK sigmoidoscopy trial, videotaped examinations were independently and blindly assessed; the completeness of visualisation and the examiner “quality” correlated with the adenoma detection rate.27 The present study included only experienced examiners with at least 500 cases of colonoscopy experience, and this was also reflected in the high caecal intubation rate of 98%. Withdrawal times were not significantly different between the groups and were well within the limits set out for quality assurance (10 minutes).

Both the staining studies mentioned above as well as our NBI trial showed that a (non-significantly) increased adenoma detection rate was accompanied by a significantly increased yield of hyperplastic polyps.28 29 This may impact on the endoscopist’s workload as a result of an increased polypectomy or at least biopsy rate. Being able to achieve a differential diagnosis of polyps on the basis of imaging criteria would then become crucial in order to save time and effort. This has been attempted with staining and/or magnification,30 31 and was also recently assessed with NBI.32 Good sensitivity was, however, usually associated with poorer specificity, with some 10–20% of adenomas possibly being misinterpreted as hyperplastic polyps. These would be inappropriately left in place, perhaps even without biopsy, if imaging alone was used as a basis for decision-making. Better non-invasive methods of achieving effective differential diagnosis therefore need to be sought. It remains to be seen whether the new imaging techniques, in which the aim is to achieve “endoscopic histology”,33 possibly in combination with techniques such as NBI incorporated into the same scope,34 will really be helpful in terms of accuracy and costs during ordinary clinical routine work.

In summary, this study has shown that in a single-centre diagnostic setting with several experienced examiners involved, NBI colonoscopy initially increased the adenoma detection rate, but that this effect was diluted as a result of better adenoma recognition developing in the conventional group as the study proceeded. It can only be speculated whether NBI induces a training effect leading to increasing adenoma detection in the control group, but this effect was obviously also observed by others.14 15 Further studies will need to demonstrate first, whether these results are reproducible by other groups, possibly in different settings, and second what would be the best method of dealing with the increased yield of hyperplastic polyps.