Meta-analysis: narrow band imaging for lesion characterization in the colon, oesophagus, duodenal ampulla and lung

Authors


Dr J. E. East, Wolfson Unit for Endoscopy, St. Mark’s Hospital, Watford Road, Harrow, Middlesex HA1 3UJ, UK.
E-mail: jameseast6@yahoo.com

Summary

Background  Narrow band imaging is a new endoscopic technology that highlights mucosal surface structures and microcapillaries, which may be indicative of neoplastic change.

Aim  To assess the diagnostic precision of narrow band imaging for the diagnosis of epithelial neoplasia compared to conventional histology both overall and in specific organs.

Methods  We performed a meta-analysis of studies which compared narow band imaging-based diagnosis of neoplasia with histopathology as the gold standard. Search terms: ‘endoscopy’ and ‘narrow band imaging’.

Results  Five hundred and eighty-two patients and 1108 lesions in 11 studies were included. Overall, sensitivity was 0.94 (95% confidence interval 0.92–0.95), specificity 0.83 (0.80–0.86); weighted area under the curve was 0.96 (standard error 0.02), diagnostic odds ratio (DOR) 72.74 (34.11–155.15). DORs were 66.65 (25.84–171.90), 61.19 (7.09–527.97), 69.74 (8.04–605.24) for colon, oesophagus and lung respectively. Studies with more than 50 patients had higher diagnostic precision, relative DOR 4.96 (1.28–19.27), = 0.022. There was no difference in accuracy between microvessel and mucosal (pit) pattern based measures, relative DOR 1.29 (0.05–35.16), = 0.87. There was significant heterogeneity overall between studies, Q = 31.2, = 0.003.

Conclusion  Narrow band imaging is accurate with high diagnostic precision for in vivo diagnosis of neoplasia across a range of organs, using simple microvessel-based measures.

Introduction

Cancer prevention is thought to be both better and cheaper than cure and policy makers and healthcare providers have started to investigate and invest heavily in screening and surveillance programmes, often at a national level, for detection of premalignant lesions for colon, stomach, oesophageal and cervical cancer.1 Colorectal cancer is a particularly attractive target as there is a clearly defined preneoplastic lesion, the adenomatous polyp, which has a long latent period before cancer develops, of the order of 5–10 years.2 Removal of these lesions through colonoscopic polypectomy may reduce future colorectal cancer risk by 70–90% for a period that may exceed 10 years;3 however this has led to endoscopists removing any lesion detected for fear of missing a potentially preneoplastic lesion. For smaller lesions, approximately half of these polyps are non-adenomatous, increasing cost and histopathology workload, as well as the small risk of additional unnecessary polypectomy.4 A similar problem exists in the surveillance of patients who have Barrett’s oesophagus (BO), where once detected, surveillance is now recommended.5, 6 Quadrantic biopsies are taken every 2 cm in the Barrett’s segment (Seattle protocol) with targeted biopsies of suspicious areas to try and detect dysplasia, leading to large numbers of nondysplastic biopsies.7 Endoscopists have sought an effective method to characterize suspicious areas accurately in vivo to allow better targeting of biopsies or polypectomy to lesions likely to be neoplastic, the so called ‘optical biopsy’.

In both the colon and BO surveillance, magnification chromoendoscopy has been used to characterize lesions with observation of the ‘pit pattern’, the shapes of the colonic crypts or glandular orifices in the columnar lined (Barrett’s) epithelium.4, 8–11 This has a high sensitivity and specificity for dysplasia in expert hands, but has not been widely adopted by Western endoscopists because of perceived need for additional equipment, time and training. Narrow band imaging (NBI) may offer similar benefits in terms of characterization to magnification chromoendoscopy, but without the need for dye-spray in a simple, push-button format.12 NBI, using optical filters, highlights surface structure and superficial mucosal capillaries.13 This can allow an enhanced appreciation of the mucosal pattern or ‘pit pattern’ and of superficial microvessel networks.14 This has led to NBI usefulness being assessed in many endoscopically accessible organs where superficial epithelial neoplasia may occur including the oro-pharynx, oesophagus, stomach, duodenal ampulla, lung, colon and bladder, with promising results.14–20 Many of these assessments have used novel microvessel assessments, relying on the increased or abnormal microvessel density seen in neoplastic lesions, common across the range of epithelial neoplasia in different organs, a significant break from the well known, but poorly utilized pit pattern classifications.21 This shared abnormality of microvessels in neoplasia seen with NBI makes combination of diagnostic studies from different organs potentially appropriate; however, overall study sizes have been relatively small in size with no study examining more than 140 patients. Although most studies report a relatively high level of accuracy, there is variability with sensitivity as low as 77% and specificity lower at 43%.22, 23 To consider the use of NBI in clinical practice, not only is a point estimate showing high accuracy when the results of multiple groups are combined needed, but also a narrow confidence interval around this estimate (diagnostic precision). If the lower limit of the confidence interval exceeds the accuracy felt necessary for clinical application, this would reassure those who wish to take NBI forward into clinical use.

The aim of this meta-analysis was to examine the diagnostic precision of NBI to characterize neoplastic lesions both across the reported spectrum of epithelial neoplasia overall and with emphasis on the colon, oesophagus and lung.

Methods

Search strategy and selection criteria

We searched Medline (using Pubmed as the search engine), as well as Web of Science, Embase and Ovid Medline, to identify studies where NBI, with or without magnification, was used to characterize endoscopically lesions into neoplastic and non-neoplastic categories in any organ until 8th August 2007. We used the search terms ‘narrow band imaging’ and ‘endoscopy’. Retrieved articles were assessed for relevance to the topic by two independent assessors (JE and PT). The reference lists of these articles and the personal libraries of the authors were searched for additional relevant articles.

Eligibility criteria and data extraction

Eligible studies were those where data on sensitivity and specificity for lesion characterization for dysplasia with NBI could be extracted. Studies without histological confirmation of dysplasia were excluded, as were case reports, editorials and commentaries and data reported as abstracts only. Publications with possible overlap of patients or lesions were discussed by JE, ET and PT and only the best quality study was used.

Data were obtained for: author, date of publication, study design (prospective or retrospective), patient characteristics (number, age), number of lesions assessed, anatomical site, use of magnification, additional use of chromoendoscopy, number of endoscopists/observers, scoring system for dysplasia (pit pattern/mucosal pattern and vascularity/microvessels) and degree of dysplasia. Where studies reported both mucosal pattern and microvascular measures, the measure with the highest accuracy was used for the overall analysis. If studies reported data for individual observers, these were treated as individual observations and not combined. Endoscopic system (Lucera or Excera II) was determined either by direct reporting in the paper or by the description of the endoscope or system used. Systems reported as using a tri-chromatic (red–green–blue) illumination system or endoscopes compatible with such a system (Olympus 240 and 260 series) were classed as Lucera. Systems reporting the use of colour chip charged coupled devices and white light illumination (Olympus 140 and 160 series endoscopes and systems) were classified as Excera II. Data were extracted independently by ET and JE. Discrepancy was resolved through consensus with PT.

Included studies were assessed for quality. Using The Standards for Reporting of Diagnostic Accuracy (STARD) guidelines, a quality score for every study was completed on the basis of title, abstract, introduction, methods, results and discussion. Quality scoring was also performed using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) methodology, where a score of 1 was given when a criterion was met, a score of −1 when a criterion was not met, and a score of 0 if it was unclear if a criterion had been achieved.24, 25

Endpoints and definitions

The primary end points were the sensitivity and specificity for NBI to characterize lesions correctly as neoplastic and non-neoplastic. In the case of BO, the distinction was made between specialized intestinal metaplasia, indeterminate for dysplasia, and low-grade dysplasia, considered non-neoplastic for the purposes of the study, and high-grade dysplasia and carcinoma, considered neoplastic. This distinction was made because there is a much higher level of agreement between histopathologists regarding what constitutes high-grade or cancer dysplasia compared to other lower grades of dysplasia.26 Also, the diagnosis of high-grade dyplasia is clinically important as this would usually prompt consideration of esophagectomy or ablation, whereas low-grade dysplasia would only lead to more intensive surveillance. In the lung, Shibuya et al. only report data for angiogenic squamous dysplasia, and therefore other types of dysplasia are not considered in assessment of their data.17

Statistical analysis

The frequency of true positives was defined as the number of lesions that were assessed with NBI measures to be dysplastic, which were subsequently shown to be dysplastic (as defined above) in histopathological analysis. True-negatives were defined as lesions assessed as being nondysplastic using NBI, which was confirmed in subsequent histological analysis. The frequency of false positives was defined as lesions assessed as dysplastic by NBI, but where no dysplasia was found at histopathology. False negatives were defined as lesions assessed as nondysplastic by NBI measures, but were dysplastic when assessed by histopathology. The sensitivity and specificity of NBI in each study were extracted or calculated by using 2 × 2 contingency tables of lesion diagnosis.

We calculated overall pooling of sensitivity [i.e. true positive/(true positive + false negative)] and specificity [i.e. true negative/(true negative + false positive)] with 95% confidence intervals (95% CIs) using a random-effect model. Summary receiver operating characteristic (ROC) analysis was used to assess the interaction between sensitivity and specificity. Diagnostic odds ratio (DOR), Q-statistic, and area under the ROC curve (AUC) were used to analyse the diagnostic precision of NBI. DOR was calculated from data for sensitivity and specificity, and was defined as: (frequency of true positives/frequency of false positives)/[(1 – frequency of true positives)/(1 – frequency of false positives)]. The higher the DOR, the greater the diagnostic precision of NBI.

The Q-statistic is a form of chi-squared test that measures heterogeneity between studies and is used to calculate the applicability of the summary ROC regression over the dataset. The AUC measures test precision from the summary ROC: a value of 0.5 infers a test that is equally likely to diagnose a positive result as either positive or negative. The perfect test, which gives a 100% correct diagnosis irrespective of patient characteristics, has an AUC of 1.0. Most clinical tests have a value of between 0.5 and 1.0, with a better diagnostic precision correlating with an AUC closer to 1.0.

We also analysed the effect of other covariates in univariate metaregression analysis (inverse variance weighted) on DOR including study quality score, whether more than 50 patients were assessed, colon vs. oesophagus (other organs were excluded as too few studies were available), use of magnification, comparison vs. chromoendoscopy and use of mucosal pattern (pit pattern) vs. microvascular measures. The relative DOR was calculated according to standard methods to analyse the change in diagnostic precision in the study per unit increase in the covariate.

Sensitivity analysis was performed to assess NBI in subgroups and to identify sources of heterogeneity between studies. The subgroups analysed were: studies with a quality score of 16 or more according to STARD criteria (maximum score 25);24 studies with a score of 13 or more according to QUADAS (maximum score 14);25 studies that assessed more than 50 patients; studies with two or more endoscopists. Studies were analysed by organ. Although no consensus yet exists for the reporting of meta-analyses of studies of diagnostic accuracy, the study reporting, as far as possible, conforms to the MOOSE guidelines for the reporting of meta-analyses of observational studies.27

Statistical analyses were performed by using STATA version 8.0 (StataCorp LP, College Station, TX, USA) and Meta-Test software version 0.9 (Tufts - New England Medical Center, Boston, MA, USA). The study was carried out in accordance with previously reported guidelines for meta-analyses of diagnostic tests.28–30

Results

The keyword search identified 51 abstracts. The abstracts were assessed to identify studies, which compared NBI based lesion characterization to histopathological diagnosis. Thirty-two articles were excluded: 20 reviews, seven commentaries or editorials and five case reports. Two further studies were identified from the authors own libraries.22, 23 Thus, 21 articles were fully assessed.14, 16–18, 20, 22, 23, 31–44 The reference lists for these studies were assessed but no further articles were identified. Seven studies reported data on either neoplastic or non-neoplastic lesions, but not both.20, 33, 34, 37, 39, 43, 44 In two studies, the study data were not extractable on a per lesion basis.31, 45 One study reported detection, but not characterization.32 Therefore, 11 studies were used for data extraction and analysis: there were five for the colon, three for the oesophagus, two for the lung and one for the duodenal ampulla14, 17, 18, 22, 23, 35, 36, 38, 40–42 (Figure 1). Tables 1 and 2 show the characteristics of these studies.

Figure 1.

 Flow chart of study selection for meta-analysis.

Table 1.   Patient demographic characteristics
Author (year)Number of subjectsLesions examinedAge (range)Inclusion criteriaExclusion criteriaStudy designSTARD/QUADAS score
  1. * Epub date.

  2. Inclusion criteria: 1 = undergoing endoscopy/surgery, 2 = screening, 3 = changes in bowel habit, 4 = unexplained weight loss, 5 = polypectomy follow-up, 6 = history of Barrett’s oesophagus, 7 = 18 years or older, 8 = smokes > 25/day, 9 = enlarged ampulla on endoscopy, 10 = bloody sputum.

  3. Exclusion criteria: 1 = unable to provide informed consent, 2 = erosive disease, 3 = evidence of cancer/mass, 4 = unable to stop aspirin/NSAID before study, 5 = chronic liver disease, 6 = uncontrolled coagulopathy, 7 = previous surgery to scoped area, 8 = unable to tolerate procedure safely.

  4. QUADAS, Quality Assessment of Diagnostic Accuracy Studies; STARD, The Standards for Reporting of Diagnostic Accuracy; R, retrospective; P, prospective.

Machida (2004)34431, 2 P15/11
Chiu (2006)*13318055 (29–87)1P18/12
Su (2006)7811051 (21–87)2, 3, 4, 5P18/12
Hirata (2007)991481R12/12
East (2007)20331P17/13
Kara (2006 – P155)631986P15/11
Sharma (2006)5120464 (43–85)1, 6, 71–7P18/12
Kara (2006 – P176)204766 (55–77)1P16/13
Shibuya (2003)486766 (42–77)1, 8, 10P11/9
Vincent (2007)226459 (28–77)1, 7, 8, 101, 8P16/12
Uchiyama (2006)141462 (51–82)9P12/10
Total5821108     
Table 2.   Technical characteristics of reported studies
Author (year)Anatomic siteReference (Gold) standardType of investigationNBI systemNo. of endoscopistsLesion characterizationPositive histology
  1. NBI, narrow band imaging.

  2. * Epub date.

  3. † Nonmagnified, all other studies used optical magnification.

  4. ‡ High definition endoscopy; however, both studies used still images recorded to nonhigh definition media for evaluation.

Machida (2004)ColonHistologyNBI, ChromoendoscopyLucera2MicrovesselsCancer/dysplasia
Chiu (2006)*ColonHistologyNBI, ChromoendoscopyLucera4MicrovesselsCancer/dysplasia
Su (2006)ColonHistologyNBI, Chromoendoscopy†‡Lucera2MicrovesselsCancer/dysplasia
Hirata (2007)ColonHistologyNBILuceraMicrovesselsCancer/dysplasia
East (2007)ColonHistologyNBI, Chromoendoscopy‡Lucera2Microvessels, pit patternCancer/dysplasia
Kara (2006a – P155)OesophagusHistologyNBILucera2Microvessels, mucosal patternCancer/high-grade dysplasia
Sharma (2006)OesophagusHistologyNBILuceraMicrovessels, mucosal patternCancer/high-grade dyplasia
Kara (2006b – P176)OesophagusHistologyNBILucera2Microvessels, mucosal patternCancer/high-grade dysplasia
Shibuya (2003)LungHistologyNBILuceraMicrovesselsAngiogenic squamous dysplasia
Vincent (2007)LungHistologyNBI†Excera II2MicrovesselsCancer/dysplasia
Uchiyama (2006)AmpullaHistologyNBILucera1Microvessels, mucosal patternCancer/dysplasia

The total number of patients assessed in the studies was 582 (range per study 14–133), who had 1108 (14–204) lesions assessed with NBI and compared with histopathology. All but one study was prospective. All eleven studies used NBI microvessel-based measures to assess for neoplasia, with a further five using mucosal pattern based measures. Four studies, all colonic, also reported data on chromoendoscopic pit pattern assessment. Figures 2–4 show the summary of overall sensitivity, specificity and DORs of NBI for diagnosing neoplasia. Overall sensitivity was 0.94 (95% CI 0.92–0.95), specificity was 0.83 (95% CI 0.80–0.86) and AUC for NBI was 0.96 [S.E. 0.02; DOR 72.74 (95% CI 34.11–155.15)], with significant heterogeneity between studies (Q value 31.20, = 0.003), Table 3. Figure 5 shows a summary ROC curve for NBI overall. Comparison between overall estimates between colon and oesophagus indicates very similar diagnostic performance for both sensitivity and specificity, with meta-regression confirming no overall statistical difference in diagnostic accuracy (Table 4). Diagnostic performance in high quality studies assessed by STARD and QUADAS criteria was very similar to that of low quality studies (Table 3); however, diagnostic precision was higher in studies that included more than 50 patients, AUC 0.98 [S.E. 0.02 (DOR 127.82 95% CI 48.81–334.71)], confirmed in the metaregression = 0.022. Sensitivity and specificity were similar depending on whether mucosal pattern ‘pit pattern’ based measures were used vs. microvascular measures and meta regression confirmed no significant difference in diagnostic accuracy overall, Table 4. There was substantial heterogeneity between studies for the overall pooled analysis and by organ with the exception of the lung and looking at oesophagus with studies, which included two or more endoscopists.

Figure 2.

 Forest plot of overall sensitivity of narrow band imaging for characterization of neoplasia.

Figure 3.

 Forest plot of overall specificity of narrow band imaging for characterization of neoplasia.

Figure 4.

 Forest plot of diagnostic odds ratio of narrow band imaging for characterization of neoplasia.

Table 3.   Pooled sensitivity, specificity and diagnostic odds ratio of NBI in the diagnosis of cancer/dysplasia
Pooled analysisNo. of Bx/totalNo. of studiesSensitivity (95% CI) HG (P-value) Specificity (95% CI) HG (P-value) DOR (95% CI) HG (P-value)AUC (S.E.)
  1. Bx, Biopsy; CI, confidence interval; DOR, diagnostic odds ratio; AUC, area under the receiver operating characteristic curve; S.E., standard error; No., number; HG, study heterogeneity (Q-value); QUADAS, Quality Assessment of Diagnostic Accuracy Studies; STARD, The Standards for Reporting of Diagnostic Accuracy.

All sites
 Overall analysis1108110.94 (0.92–0.95)37.62 (<0.001)0.83 (0.80–0.86)113.72 (<0.001)72.74 (34.11–155.15)31.20 (0.003)0.96 (0.02)
 Studies with ≥2 endoscopists67570.92 (0.89–0.94)17.16 (0.046)0.74 (0.69–0.79)42.39 (<0.001)44.15 (23.47–83.05)14.50 (0.106)0.94 (0.02)
 Studies with ≥50 patients84050.95 (0.93–0.96)25.14 (<0.001)0.87 (0.84–0.90)52.62 (<0.001)127.82 (48.81–334.71)16.22 (0.013)0.98 (0.02)
 STARD ≥1663860.92 (0.90–0.94)23.71 (0.003)0.84 (0.80–0.87)95.80 (<0.001)53.77 (21.74–132.97)23.06 (0.003)0.96 (0.03)
 QUADAS ≥1278670.94 (0.92–0.95)36.24 (<0.001)0.84 (0.80–0.87)97.27 (<0.001)69.24 (26.68–179.68)29.29 (0.001)0.96 (0.03)
 Pit/mucosal pattern44940.92 (0.86–0.96)13.70 (0.008)0.88 (0.84–0.92)51.83 (<0.001)64.66 (8.45–494.67)17.65 (0.001)0.98 (0.03)
 Hue/Vascularity (microvessels)1108110.94 (0.92–0.95)37.62 (<0.001)0.83 (0.80–0.86)113.72 (<0.001)72.74 (34.11–155.15)31.20 (0.003)0.96 (0.02)
Colon only
 Overall analysis51450.93 (0.91–0.95)28.47 (<0.001)0.84 (0.78–0.89)12.98 (0.073)66.65 (25.84–171.90)19.51 (0.007)0.93 (0.04)
 Studies with ≥2 endoscopists36640.92 (0.89–0.94)14.41 (0.025)0.83 (0.77–0.88)11.47 (0.075)50.79 (21.76–118.55)13.30 (0.038)0.94 (0.03)
 STARD ≥ 1632330.92 (0.89–0.94)14.40 (0.013)0.82 (0.76–0.88)8.06 (0.153)46.86 (19.05–115.24)12.69 (0.027)0.92 (0.05)
 QUADAS ≥ 1247140.93 (0.91–0.95)28.26 (<0.001)0.83 (0.77–0.88)9.76 (0.135)62.69 (22.88–171.76)19.02 (0.004)0.90 (0.07)
Oesophagus only
 Overall analysis44930.93 (0.85–0.97)7.07 (0.029)0.83 (0.77–0.88)23.57 (<0.001)61.19 (7.09–527.97)8.97 (0.011)0.99 (0.00)
 Studies with ≥2 endoscopists24520.88 (0.77–0.95)0.07 (0.795)0.73 (0.63–0.81)0.01 (0.910)19.50 (7.70–49.37)0.08 (0.781)
 STARD ≥ 1625120.95 (0.87–0.99)5.15 (0.023)0.93 (0.86–0.97)9.68 (0.002)188.29 (1.94–18271.20)7.11 (0.008)
 QUADAS ≥ 1225120.95 (0.87–0.99)5.15 (0.023)0.93 (0.86–0.97)9.68 (0.002)188.29 (1.94–18271.20)7.11 (0.008)
Colon & oesophagus
 Overall analysis96380.94 (0.92–0.96)37.73 (<0.001)0.86 (0.83–0.89)62.35 (<0.001)85.19 (35.14–206.52)28.44 (0.002)0.96 (0.03)
Lung only
 Overall analysis13120.96 (0.82–1.00)1.28 (0.26)0.68 (0.58–0.77)31.25 (<0.001)69.74 (8.04–605.24)1.43 (0.23)
Figure 5.

 Summary receiver operating characteristic (SROC) curve showing the diagnostic precision of narrow band imaging for characterization of neoplasia.

Table 4.   Weighted meta-regression comparison for accuracy of lesion characterization
ComparisonNumber of studiesCoefficientRelative DOR (95% CI)P
  1. DOR, diagnostic odds ratio; NBI, narrow band imaging.

  2. * Colonic studies only.

  3. Significant values are in bold.

NBI vs. chromoendoscopy*41.1913.29 (0.86–12.63)0.08
NBI vs. white light*32.0517.78 (2.01–30.05)0.009
Colon vs. Oesophagus8−0.8090.45 (0.04–4.62)0.459
Microvessels vs. Mucosal pattern50.2581.29 (0.05–35.16)0.865
Magnified vs. nonmagnified90.7992.22 (0.60–8.25)0.224
Studies with ≥50 vs. <50 patients111.6014.96 (1.28–19.27)0.022

Table 4 shows the findings of the meta regression analysis. This confirms no significant difference in overall diagnostic accuracy between oesophagus and colon, mucosal pattern and microvascular measures and between use of magnification and no magnification. Studies with 50 or more patients were more accurate, relative DOR 4.96 (1.28–19.27), = 0.022. There was no significant difference in diagnostic accuracy between NBI and chromoendoscopy in studies in the colon, relative DOR 3.29 (95% CI 0.86–12.63) = 0.08; however, NBI was significantly more accurate than assessment with white light alone in the colon, relative DOR 7.78 (95% CI 2.01–30.05), = 0.009.

Discussion

Principal findings

This study suggests that NBI endoscopic analysis of epithelial lesions in the gastrointestinal tract and lung has a high level of diagnostic precision for neoplasia. The overall sensitivity of NBI for neoplasia was 0.94 (0.92–0.95) and overall specificity was 0.83 (0.80–0.86). Results were similar for the colon, oesophagus and lung. In terms of overall accuracy, there was no significant difference between the use of microvessel measures and pit pattern assessment. Accuracy was significantly higher in studies with more than 50 patients. Although there was significant study heterogeneity, there was a high overall DOR for each organ (range 61.19–69.74) that was not affected by study quality. The high point accuracy, as assessed by sensitivity and specificity, with narrow confidence limits (precision), confirms that NBI is likely to be an effective clinical tool during endoscopy in routine practice to target biopsies or therapy, potentially reducing cost, time and risk.

Comparison with other studies

The use of microvascular networks as a marker of neoplasia is a novel feature of NBI.13, 14, 17, 18, 35, 41 Epithelial neoplasia also leads to changes in surface structure, which can be detected by NBI, described as the mucosal pattern or ‘pit pattern’. High levels of accuracy for the differentiation of neoplastic from non-neoplastic lesions using mucosal pattern in both the oesophagus and the colon has been achieved with magnification chromoendoscopy in expert hands;4, 8–11 however, chromoendoscopy has not been widely adopted by endoscopists as it requires additional time, training and the use of dye-spray catheters and mucolytics for the best results.8, 11 The accuracy reported for NBI for colon overall is closely comparable to that seen in previous studies using expert magnification chromoendoscopy.4, 8, 11 This impression is confirmed in the meta-regression analysis of colonic studies where NBI and chromoendoscopy were both used. Accuracy was almost identical with no significant difference seen. Accuracy of NBI in the oesophagus was also similar to expert chromoendoscopy;9, 10 however, as no included oesophageal study used NBI and chromoendoscopy, a head-to-head comparison with meta-regression is not possible. Nevertheless, this suggests that NBI may be considered an equivalent to chromoendoscopy for lesion characterization in both the oesophagus and colon as form of ‘electronic dye-spray’.

Study limitations

This study has a number of limitations. There is no comparison with a white light control group where lesions were assessed without NBI for all studies; however, studies by Chiu et al., Su et al. and Machida et al. have looked at this issue in the colon and meta-regression confirms significantly lower diagnostic accuracy without the use of NBI.14, 38, 42 The situation in the oesophagus is less clear with one study, published after study selection and not included in our analysis, suggesting that white light images with magnification were as accurate at predicting neoplasia as magnified NBI images.46 Comparison with chromoendoscopy is also limited to colonic studies with only four studies reporting data for both chromoendoscopy and NBI. In the metaregression analysis, there was no significant difference in accuracy, consistent with the overall NBI data showing equivalent accuracy to chromoendoscopy. NBI may be preferred to chromoendoscopy as it seems to have a relatively short learning curve compared to the 200–300 lesions for chromoendoscopy.11 We were unable to assess learning curve formally, but accuracy was higher in studies with more than 50 patients, suggesting that a learning curve exists, but may be relatively short.

Narrow band imaging itself has disadvantages: a new video processor and light source need to be purchased, adding to capital cost; the endoscopic image is darker requiring closer range examination of the mucosa; stool appears brick red with NBI, making examination impossible without meticulous bowel preparation; optical magnification was used in all but two studies, leading to further potential capital cost to purchase ‘zoom’ instruments, which may have a wider diameter and be less manoeuvrable; there are limited head-to-head data comparing NBI, chromoendoscopy and high definition white light examination in each organ; inflammation leads to higher tissue microvascular density. This may mimic the increased microvessel density that NBI classification systems use to differentiate neoplastic (microvessels present) from non-neoplastic tissue (microvessels absent), a problem already seen in colitis, and a potential confounder in inflamed BO and in the lung, leading to false positive results.32, 47 The current series of studies with one exception use the Lucera (red–green–blue trichromatic illumination) system, which anecdotally gives higher contrast than the colour chip based Excera system. Therefore, these results can probably only be generalized to Lucera and further data must be awaited before firm conclusions can be drawn regarding the diagnostic precision of Excera-based NBI. Furthermore, the Lucera system available commercially uses the 260 series light sources and processors and it is unknown if the performance is identical to prototype systems used in studies, which may include 240 series technology. Two studies used high definition endoscopes, but both used review of still video images recorded to nonhigh definition media and therefore the role of high definition endoscopy in NBI characterization, which might increase accuracy further, cannot be assessed.22, 40

Studies of diagnostic accuracy also need to consider specific biases: incorporation bias is unlikely as most studies state that histopathology was interpreted without knowledge of the result of NBI examination, although this is not explicit in four studies; verification bias is not possible as all included lesions underwent both the experimental test (NBI) and reference test (histopathology). Bias caused by selective publication is possible and we have not actively sought unpublished data. The assessment of publication bias is challenging in studies of diagnostic accuracy as comparison is usually to a reference standard rather than examining the outcomes between two tests and therefore standard techniques such as funnel plots cannot be used; however, sensitivity analyses did not show a difference between high and low quality studies in terms of accuracy. Furthermore, small studies were significantly less accurate than larger studies, the reverse of what would be expected if small studies, which may initially overestimate accuracy because of low sample size, had been preferentially published.

This study did not look at the ability of NBI to differentiate between dysplasia and carcinoma. This is potentially important as the endoscopic approach is different if resection is considered.12 In the colon, there is evidence that vessel thickness and particularly vessel irregularity are predictive of invasive lesions, although data are currently limited.48 Kudo type V (nonstructural) pit pattern seen with NBI may also be predictive of invasion in the colon, but currently there is only a single case report addressing this issue.12

There was also significant study heterogeneity; but, in view of the variety of organs examined and novel classification systems used, this is not surprising and may decrease over time as classification systems become standardized and endoscopists familiar with them.

Clinical implications

The high diagnostic accuracy of NBI suggests that it is likely to be a useful tool in accurately targeting biopsies or therapy towards neoplasia. A step further would be to consider whether for low risk lesions NBI might be accurate enough to replace conventional histopathology either completely, with the lesion not being sent for histological assessment (colon) or to increase the likelihood of dysplasia in biopsies, with a reduction in random sampling (BO and lung).49 Critical to introduction into clinical practice will be data on inter- and intra-observer variability. Currently, very limited data exist with no data on intra-observer variability. One small study in the colon suggests that inter-observer agreement between two observers was higher using microvessel measures, weighted kappa 0.64 (substantial agreement), than NBI pit patterns, weighted kappa 0.48 (moderate agreement).22 In BO, inter-observer agreement between 12 observers using white light combined with NBI had kappa values of 0.38–0.42 (fair-to-moderate agreement).46

Colon.  This is particularly relevant in the colon where a majority of neoplastic polyps, 80%, are diminutive (5 mm or less in size) and very rarely contain advanced histology (high-grade dysplasia or villous elements, <2%) and are only reported to contain cancer in case reports where the lesion almost always has a depressed morphology.50–52 An accuracy of 90–95% has been suggested as the required level to consider clinical introduction of such ‘optical biopsy’.22 Our data suggest that this level of accuracy is possible for colonic lesions with narrow confidence limits and that appropriately trained and accredited endoscopists might safely stop sending diminutive colonic lesions routinely for histopathology; however, further data from large, multi-centre, randomized studies, involving non-academic endoscopists in routine clinical practice are needed to clarify the potential clinical benefits and cost savings that appear possible. The clinical utility of NBI in the colon only appears to relate to characterization of previously identified lesions, as there are currently no randomized data to suggest that NBI improves adenoma detection.53, 54

Barrett’s oesophagus.  In BO, the situation is somewhat different in that low-grade dysplasia was considered a negative finding in this study and only high-grade dysplasia and cancer were considered positive. This choice of dichotomous endpoint has been commonly used in oesophageal studies as the histopathological interpretation between these two end points is much more consistent, the likelihood of a true invasive lesion is high and the clinical approach, usually to offer oesophagectomy, or endoscopic resection/ablation in selected cases, is clear,26 whereas for low-grade dysplasia, further more intensive endoscopic surveillance is warranted. Nevertheless, a finding of low-grade dysplasia is potentially important and so the diagnostic values presented here should be interpreted with this in mind. On the basis of our data, areas characterized as neoplastic with NBI have a high likelihood of containing neoplasia or carcinoma, which will help target biopsies to clinically important areas, but it is probably too soon to abandon nontargeted ‘Seattle protocol’ biopsies. Furthermore, although NBI appears to give clearer mucosal images than white light, this may not improve diagnostic accuracy beyond that offered by white light, with potentially additional information even being distracting, when assessed with still images.46 A formal, prospective, head-to-head clinical trial of diagnostic accuracy would clarify this area.

Lung and duodenal ampulla.  Data for the lung and duodenal ampulla are relatively limited. In the lung, sensitivity was good, but specificity poor, with one trial assessing a specific sub-type of bronchogenic dysplasia.17 This makes the results more difficult to generalize, but suggests that NBI may be helpful to exclude neoplasia as the technique has a high negative predictive value, potentially reducing biopsy numbers. The results in a small study examining the duodenal ampulla were good, but need to be replicated in further studies.18 Assessment of ampullary lesions is something of a niche area, but might have particular relevance for the surveillance of patients with familial adenomatous polyposis, where peri-ampullary biopsies are routinely taken to detect dysplasia and may be associated with a small risk of causing pancreatitis.55 Generalization of the overall result to NBI performance in these two organs may be inappropriate given the relatively small contribution each organ makes to the overall result.

Future directions

As results were similar between upper and lower gastrointestinal tract and the lung, the prospects for NBI being used successfully for lesion characterization in other endoscopically accessible organs, such as the bladder and oro-pharynx seem promising. This may relate to the fact that abnormal microvascular networks are a common and early feature of neoplasia on epithelial surfaces and NBI is able to highlight these networks.56–58 Future studies will need to be methodologically rigorous and ideally randomized, comparing current best tools, e.g. chromoendoscopy or high definition white light with NBI and make comprehensive assessment of inter- and intra-observer variability. Terminology, particularly relating to microvessel patterns, will need to be standardized and further mechanistic studies will need to be performed. Most importantly, the effect of introducing NBI based lesion characterization on patient management and outcomes will need to be assessed, if such technologies are to be adopted widely.

Conclusions

In conclusion, NBI at endoscopy has a high diagnostic precision for epithelial neoplasia across multiple organs, using either mucosal pattern or simpler microvascular based measures. Use of NBI is likely to improve targeting of endoscopic biopsies or therapy to neoplastic lesions. These finding should encourage further studies on the use of NBI for lesion characterization in other organs such as the bladder and oro-pharynx and to assess cost effectiveness of NBI targeted biopsies and therapy in routine clinical practice.

Acknowledgements

Declaration of personal interests: No author has a conflict of interest; however, Drs East, Saunders and Bergman have used NBI equipment on loan from Olympus Medical Systems Corp., Tokyo, Japan, to conduct NBI based research. Declaration of funding interests: No funding was received for this study.

Ancillary