Qing Jiang, Department of Urologic Surgery, Second Affiliated Hospital of Chongqing Medical University, No. 76, Lin Jiang Road, Yu Zhong District, Chongqing 400016, China. e-mail: firstname.lastname@example.org
Study Type – Diagnosis (systematic review)
Level of Evidence 1
What's known on the subject? and What does the study add?
In recent years, more attention has focused on the role of narrow band imaging (NBI) in bladder cancer detection and NBI technology has spread rapidly. It is an important method for diagnosing new or recurrent bladder cancer. But its diagnostic accuracy is still uncertain.
This paper summarizes the diagnostic accuracy of NBI in bladder cancer and compares NBI with white light imaging. The results show that NBI cystoscopy significantly improves the detection accuracy in bladder cancer, compared with white light imaging. However, some limitations still exist. Multicentre randomized studies are recommended to determine whether the visual advantages of NBI can translate into real therapeutic benefit for individual patients.
• To assess the test performance and clinical effectiveness of narrow band imaging (NBI) cystoscopy compared with white light imaging (WLI) cystoscopy in people suspected of new or recurrent bladder cancer.
• Literature on NBI cystoscopy in the diagnosis of bladder cancer was searched in PubMed, EMBASE, Cochrane Library, MEDLINE and CNKI, with hand searching of relevant congress abstracts and journals.
• The literature was selected according to inclusion and exclusion criteria. The Meta-DiSc1.4 software was used to review management and analysis.
• Eight studies including 1022 patients assessed test performance.
• On a per-person analysis, the pooled sensitivity, specificity, positive likelihood ratio, negative likelihood ratio and diagnostic odds ratio of NBI and WLI were respectively 0.943 (95% CI 0.914−0.964) and 0.848 (95% CI 0.803−0.885), 0.847 (95% CI 0.812−0.878) and 0.870 (95% CI 0.831−0.903), 7.038 (95% CI 3.357−14.754) and 6.938 (95% CI 2.052−23.465), 0.054 (95% CI 0.012−0.237) and 0.181 (95% CI 0.091−0.361), and 185.32 (95% CI 45.714−751.26) and 42.931 (95% CI 8.088−227.88).
• The area under the curve and Q* of NBI and WLI were respectively 0.9781 and 0.8944, and 0.9337 and 0.8253.
• For the characterization of carcinoma in situ, the pooled sensitivity, specificity, positive likelihood ratio, negative likelihood ratio and diagnostic odds ratio of NBI were 0.927 (95% CI 0.878−0.960), 0.768 (95% CI 0.730−0.802), 4.545 (95% CI 2.820−7.325), 0.125 (95% CI 0.051−0.304) and 48.884 (95% CI 15.642−152.77) on a per-person analysis.
• The area under the curve and Q* were 0.9391 and 0.8763.
• NBI is an effective method for the identification of abnormal lesions including carcinoma in situ and can provide higher diagnostic precision of bladder cancer than WLI.
Bladder cancer is the second most common malignancy of the genitourinary system. It is estimated that more than 70 530 people are diagnosed each year in the USA . Approximately 80% of patients have non-muscle-invasive bladder cancer (NMIBC, pTa, pT1, carcinoma in situ[CIS]) at first diagnosis . NMIBC has characteristics of high risk of recurrence and probability of grade and stage progression in patients after transurethral resection. For a long time, white light imaging (WLI) cystoscopy was the standard method in NMIBC diagnosis . However, WLI cystoscopy cannot effectively identify small bladder cancer such as CIS or small papillary tumours, resulting in residual tumour and recurrence after transurethral resection. WLI cystoscopy is not the best method for early diagnosis and follow-up. Therefore, novel endoscopic imaging techniques are urgently needed to improve the detection of bladder cancer. Narrow band imaging (NBI) is an optical image enhancement technology that depends on the use of two short wavelength light beams of 415 nm (blue) and 540 nm (green). The light penetrates the superficial bladder tissues and is strongly absorbed by haemoglobin, so it greatly increases the visibility of capillaries and delicate tissue surface structures by reinforcing contrast of the bladder surface of tissues. NBI technology has been a hot research topic in recent years, and some studies report the test performance of NBI. The objective of this review is to analyse the diagnostic accuracy (sensitivity and specificity) of NBI in the diagnosis of bladder cancer.
We searched the literature in PubMed, EMBASE, Cochrane Library, MEDLINE and CNKI (publications up to April 2012) using search terms ‘narrow band imaging’, ‘NBI’, ‘bladder cancer/bladder tumour’, ‘superficial bladder cancer’, ‘non-muscle-invasive bladder cancer’ and ‘urothelial carcinoma’ without language restriction. We also hand searched the relevant congress abstracts and journals. Two of the authors (C.J.Z. and Y.L.L.) independently searched the databases and reviews; case reports and duplicated studies were excluded.
Studies were included if (i) NBI was performed as a diagnosis modality for bladder cancer; (ii) at the same time sensitivity and specificity were provided or a 2 × 2 contingency table could be constructed; (iii) absolute numbers of true positive, false positive, true negative and false negative cases or their equivalent were reported or could be calculated from the literature; (iv) pathology from surgical treatment or biopsy was used as the gold standard for the diagnosis of lesions. Patients enrolled in the trials consisted of clearly identified different groups of patients (with and without disease) by the gold standard.
Studies were excluded if (i) they were reviews, editorials, opinions, animal models or case reports; (ii) in addition to NBI cystoscopy and WLI cystoscopy, other instruments were used as the diagnostic modality; (iii) patients were undergoing the procedure without pathological confirmation of lesions.
EXTRACTION OF DATA AND QUALITY EVALUATION
The data from included studies were independently extracted by two reviewers (C.J.Z. and Y.L.L.). The following variables were extracted: first author, year of publication, type of study, number of biopsies and patients, baseline patient characteristics and diagnostic purposes. The true positives, false positives, false negatives and true negatives were extracted using the pathology as gold standard. All data were extracted both on a per-patient and a per-lesion basis wherever available.
The included papers were assessed using the Quality Assessment of Diagnostic Studies (QUADAS) . The tool is based on an 11-item questionnaire recommended by the Cochrane Handbook, with each item having the response ‘yes’, ‘no’ or ‘unclear’. Ultimately a quality score is calculated to infer the quality of research. Quality assessment of studies was done by two reviewers independently (C.J.Z. and Y.L.L.) and differences of opinion were resolved by mutual agreement.
We performed data analysis using the Meta-DiSc version 1.40 , provided by the Cochrane Collaboration. Sensitivity, specificity, likelihood ratio (LR) and diagnostic odds ratio (DOR) were used as measures of the diagnostic accuracy of NBI and WLI in characterizing bladder cancer, including CIS. We formulated forest plots of the summary measures of accuracy and examined the heterogeneity of the summary measures of sensitivity and specificity with a random-effects model. The summary receiver operating characteristic (SROC) curves were drawn to describe the joint distribution of true positive rate and false positive rate . The area under the curve (AUC) is an overall summary measure index of the diagnostic. A perfect test will have an AUC close to 1 and a poor test has AUC close to 0.5 . The Q* index is defined by the point where sensitivity and specificity are equal, which is the point closest to the ideal top-left corner of the SROC space . LRs state how many times more likely particular test results in patients with disease are than in those without disease. If positive LRs are greater than 10 and negative LRs are less than 0.1, they are considered to provide convincing diagnostic evidence, whereas those >5 and <0.2 give strong diagnostic evidence . Heterogeneity could be due to variation in thresholds, the base situation of patients, the test method and the study quality among studies. To explore this source of variation, it is useful to plot sensitivity and specificity on an ROC plane. If such a threshold effect exists, the points will show a curvilinear pattern and the pooled summary estimates from meta-analyses are meaningless. Combining study results in these cases involves fitting an ROC curve rather than pooling sensitivities and specificities or LRs. Cochran's Q test, which has a chi-squared distribution with k− 1 degrees of freedom , is used to test the heterogeneity. The I2 index, which describes the percentage of total variation across studies that is due to heterogeneity rather than chance , is another approach to quantify the effect of heterogeneity. Values of I2 of 25%, 50% and 75% may be considered to represent low, moderate and high inconsistency.
SELECTION OF STUDIES
An initial search identified 75 potentially relevant studies. After reading the titles and abstracts, 51 studies were excluded because of irrelevance. With further screening of full texts, 12 studies were excluded. The reasons for study exclusion were as follows. Six of the 12 studies had no data available on true negatives or false positives for calculating sensitivity and specificity; five were reviews of the literature or editorials; four were duplicated studies; and one was animal experiments. Finally, eight studies [11–18] were available for the meta-analysis, published between 2008 and 2012 and including 1022 patients. The literature screening process is shown in Fig. 1. The characteristics of each study are shown in Table 1.
Table 1. Characteristics of studies included in the meta-analysis
Mean age (range)
No. of patients
N/A, not available (insufficient information provided).
QUADAS quality assessment of the included studies is shown in Fig. 2. All the eligible studies scored >9 points (11 is full points), indicating that they were of good quality. Of the eight studies, six enrolled patients prospectively. Three studies were performed in the USA [11–13], one was conducted in Japan , three were carried out in China [14,17,18] and one was performed in The Netherlands .
DIAGNOSTIC ACCURACY OF NBI AND WLI ON A PER-PERSON BASIS
Five studies including 759 patients were analysed for NBI on a per-person basis. The pooled sensitivity and specificity of NBI were 0.943 (95% CI 0.914–0.964) and 0.847 (95% CI 0.812−0.78) respectively (Fig. 3). The pooled positive LR was 7.038 (95% CI 3.357−14.754) and the pooled negative LR was 0.054 (95% CI 0.012−0.237). The pooled DOR was 185.32 (95% CI 45.714−751.26) using a random-effects model. The AUC was 0.978 (SE 0.015) with Q*= 0.934 (SE 0.027) (Fig. 4), indicating a high level of diagnostic accuracy for NBI.
For WLI, three studies including 648 people were included in the pooled diagnostic assessment of performance; the pooled sensitivity and specificity were 0.848 (95% CI 0.803−0.885) and 0.870 (95% CI 0.831−0.903) (Fig. 5). The pooled positive LR was 6.938 (95% CI 2.052−23.465) and pooled negative LR was 0.181 (95% CI 0.091−0.361). The pooled DOR was 42.931 (95% CI 8.088−227.88). The AUC was 0.894 (SE 0.076) with Q*= 0.825 (SE 0.080) (Fig. 6).
DIAGNOSTIC ACCURACY OF NBI AND WLI ON A PER-LESION BASIS
In the pooled estimates for biopsy level analysis, based on evidence from four studies including 341 people and 1195 lesions, the pooled sensitivity and specificity of NBI were 0.949 (95% CI 0.929–0.964) and 0.548 (95% CI 0.503−0.593) respectively (Fig. 7). The pooled positive LR was 2.083 (95% CI 1.259−3.446) and pooled negative LR was 0.113 (95% CI 0.074−0.173). The pooled DOR was 23.046 (95% CI 9.230−57.546) using a random-effects model. The AUC was 0.903 (SE 0.067) with Q*= 0.835 (SE 0.072), indicating a high level of diagnostic accuracy for NBI (Fig. 8).
For WLI, four studies involving 341 people and 1195 lesions were included in the pooled diagnostic assessment of performance; the pooled sensitivity and specificity were 0.751 (95% CI 0.717−0.784) and 0.719 (95% CI 0.678−0.758) (Fig. 9). The pooled positive LR was 2.488 (95% CI 1.175−5.270) and pooled negative LR was 0.418 (95% CI 0.283−0.619). The pooled DOR was 5.879 (95% CI 2.408−14.352). The AUC was 0.768 (SE 0.056) with Q*= 0.708 (SE 0.047) (Fig. 10).
DIAGNOSTIC ACCURACY OF NBI IN CIS
Four studies including 719 people were used to analyse the characterization of CIS by NBI on a per-person basis. The pooled sensitivity and specificity were 0.927 (95% CI 0.878−0.960) and 0.768 (95% CI 0.730−0.802) respectively (Fig. 11). The pooled positive LR was 4.545 (95% CI 2.820−7.325) and pooled negative LR was 0.125 (95% CI 0.051−0.304). The pooled DOR was 48.884 (95% CI 15.642−152.77) using a random-effects model. The AUC was 0.939 (SE 0.033) with Q*= 0.876 (SE 0.041) (Fig. 12).
There are significant heterogeneities among studies both for NBI and WLI analysis, as shown in Figs 3, 5, 7, 9 and 11. But the heterogeneities were not caused by threshold effects except for the analysis for WLI on a per-lesion basis. Because of the small number of studies, we did not use meta-regression or subgroup analysis. We performed SROC to summarize the diagnosis accuracy.
WLI cystoscopy is the gold standard method in bladder cancer diagnosis and recurrence monitoring. However, it usually overlooks clinically important papillary tumours and CIS. Repeat transurethral resection performed only 2–6 weeks after the initial resection demonstrates that incomplete resection is frequent . Residual tumour is found in 30%–44% of patients resected up to 8 weeks after the original transurethral resection [20,21]. Recent advances in imaging technology offer the ability to improve the diagnostic accuracy of bladder tumours. Fluorescent cystoscopy and NBI cystoscopy are two new imaging techniques that can enhance the diagnostic accuracy. However, fluorescent cystoscopy, which costs too much, has many defects. It is complex to operate and not popular in clinical work. Also the latest meta-analysis  showed that fluorescent light guided transurethral resection of bladder tumours was not superior to conventional WLI in diagnostic accuracy and had no significant effect on short-term recurrence-free survival and progression-free survival. However, NBI is now widely used in gastrointestinal endoscopy and has already been shown to be superior to conventional WLI endoscopy [23,24]. In the field of urology, Bryan et al. was the first person to use NBI cystoscopy to follow 29 patients with known recurrences of bladder cancer. He found that NBI cystoscopy could provide a much clearer view of bladder cancer and improve the detection rate of recurrent disease in patients with NMIBC. The study  revealed that NBI cystoscopy could easily discover non-muscle-invasive high grade bladder cancer. Retrospective studies [27,28] and randomized prospective studies [29,30] showed that NBI-assisted transurethral resection could significantly reduce the residual tumour and recurrence risk.
The results of our meta-analysis showed that NBI cystoscopy has a high diagnostic precision of bladder cancer. The pooled sensitivity, specificity, positive LR, negative LR and DOR were 0.943, 0.847, 7.038, 0.054 and 185.32, respectively. The AUC was 0.98 (SE 0.0148) with Q*= 0.9337 (SE 0.027). NBI can improve the detection rate of bladder cancer. As we know, the diagnosis and treatment of CIS is particularly important for the control of tumour progression, because of the high potential to recur and spread into bladder muscle and to increase risks of disease-specific mortality. Our study showed that the sensitivity and negative LR in the diagnosis of CIS were 0.927 (95% CI 0.878−0.960) and 0.125 (95% CI 0.051−0.304) respectively. So, NBI is valuable for the diagnosis of CIS.
In our study, the specificity of NBI cystoscopy was lower than that of WLI cystoscopy. In our opinion, there were two main reasons for the result. The first reason might have been because patients with recurrent bladder cancer were enrolled, and therefore the false positive rate (1 − specificity), influenced by inflammatory lesions after intravesical therapy, was decreased especially in NBI [14–16]. In addition, because of the subjectivities of doctors, different doctors may make different judgements about the same lesion, because at present there is no agreement with a unified and accepted judging standard. The specificity of NBI cystoscopy, however, increases the number of NBI cystoscopy negative normal tissues, but this could not conceal the advantage of accurate detection of small lesions. We evaluate a new diagnosis technique not only by the sensitivity and specificity but also by the value prediction and LR. Although the sensitivities of NBI were more than 90%, the negative LRs were low, indicating that NBI cystoscopy is still an effective technique for identifying abnormal lesions; it is useful for excluding a diagnosis of bladder tumour and increases the chance of an effective therapeutic outcome.
As usual, most reviews have some limitations, as did ours. First, there may be some bias caused by the designs of the studies included. WLI cystoscopy and NBI cystoscopy were performed subsequently by the same urologist in two of the studies [11,15]. This may produce some bias which enhanced the tumour detection of NBI. Nevertheless, this problem had already been tackled in other studies [14,16–18]. These studies showed that NBI cystoscopy detected more lesions, including CIS, with mapping between NBI and WLI in a randomized imaging sequence modality by two urologists. In particular, randomized controlled trials  revealed that NBI cystoscopy significantly improved the diagnostic accuracy in NMIBC and the study of 126 patients  found that NBI cystoscopy was associated with fewer patients having tumour recurrences and a longer recurrence-free survival time (3 years). Another limitation of the design was that some studies excluded bladder intravesical BCG therapy patients. A previous study  showed that NBI appears to better identify patients who have suspected residual tumour on follow-up WLI at 3 months after BCG therapy. But more high quality and multicentre studies are still needed for further investigation of this advantage. Second, it is challenging to assess the publication bias of meta-analysis in diagnosis. It is difficult to interpret funnel plots for the small sample of studies included. So the publication bias was not assessed. Third, there was obvious heterogeneity between studies. The random-effect model was used to summarize the effects of NBI; however, we cannot perform meta-regression and subgroup analysis to identify the sources of heterogeneity as a limitation of the primary studies. Fourth, it was a pity that the true positive, false positive, true negative and false negative cannot be extracted for calculating sensitivity and specificity in six related studies [25,26,28,31–33]. Some useful information may have been missed. Finally, we only included the English and Chinese literature, and therefore language bias may exist. So, more clinical evidence is needed to confirm the findings in this review.
In conclusion, the existing evidence shows that NBI is an effective method for the identification of abnormal lesions including CIS and NBI can provide a high diagnostic precision method to WLI. More multicentre randomized studies should be launched to compare the impact of NBI and WLI transurethral resection on residual tumour and the recurrence/progression of bladder tumours. Prospective studies are recommended to determine whether the visual advantages of NBI can translate into real therapeutic benefit for individual patients. As only a few studies with small study populations were available, we believe that more results with high quality trials should be provided to update this study.