A meta-analysis of narrow band imaging for the diagnosis and therapeutic outcome of non-muscle invasive bladder cancer

Objectives To assess the additional detection rate (ADR) of within-patient comparisons of Narrow band imaging (NBI) and white light cystoscopy (WLC) for non-muscle invasive bladder cancer (NMIBC) detection and compare the impact of NBI and WLC on bladder cancer recurrence risk. Methods We searched relevant studies from PubMed, Embase, Medline, Web of Science and the Cochrane Library database for all articles in English published beforeJuly26th, 2016. Pooled ADR, diagnostic accuracy, relative risk (RR) and their 95% confidence intervals (CIs) were calculated. Results Twenty-five studies including 17 full texts and eight meeting abstracts were included for analysis. Compared to WLC, pooled ADR of NBI for NMIBC diagnosis was 9.9% (95% CI: 0.05–0.14) and 18.6% (95% CI: 0.15–0.25) in per-patient and per-lesion analysis, respectively. Pooled ADR of NBI for carcinoma in situ (CIS) diagnosis was 25.1% (95% CI: 0.09–0.42) and 31.1% (95% CI: 0.24–0.39) for per-patient and per-lesion analyses, respectively. The pooled sensitivity of NBI was significantly higher than WLC both at the per-patient (95.8% vs. 81.6%) and per-lesion levels (94.8% vs. 72.4%). In addition, NBI significantly reduced the recurrence rate of bladder cancer with a pooled RR value of 0.43 (95% CI: 0.23–0.79) and0.81 (95% CI: 0.69–0.95) at month three and twelve, respectively. Conclusions NBI is a valid technique that improves the diagnosis of NMIBC and CIS compared to standard WLC either at per-patient or per-lesion level. It can reduce the recurrence rate of bladder cancer accordingly.


Conclusions
NBI is a valid technique that improves the diagnosis of NMIBC and CIS compared to standard WLC either at per-patient or per-lesion level. It can reduce the recurrence rate of bladder cancer accordingly. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

Introduction
Bladder cancer (BC) is the fourth most common malignancy in men and ninth in women [1,2].The incidence of BC is rapidly increasing in underdeveloped countries. Approximately 80% of diagnosed bladder tumors are non-muscle invasive bladder cancer (NMIBC) [3]. White light cystoscopy (WLC) is the standard imaging tool to identify suspicious lesions, detect cancer and tumor recurrence in bladder. Once a lesion is identified, transurethral resection (TUR), the mainstay of treatment for NMIBC, will be performed to assess histopathologic grade and stage. Despite its central role, WLC has several well recognized limitations. It is difficult to visualize non-papillary bladder cancer using WLC, such as carcinoma in situ (CIS), and small, or satellite tumors [4].In addition, bladder cancer may be incompletely resected because of understaging [5]. These limitations of WLC contribute to the high risk of cancer persistence and high recurrence rate (approximately 61% at year one and 78% at year five) [6,7]. Due to the high prevalence, high recurrence rate, and the need for long-term cystoscopic surveillance, BC has a tremendous impact on healthcare infrastructure and costs [8].
NBI is a valid technique that can improve bladder cancer detection. NBI filters out the red spectrum of white light, resulting in blue (415 nm) and green (540 nm) bands that can differentially penetrate mucosa to enhance visualization of mucosal vasculature and highlight neoplastic neoangiogenesis of urothelial tumors. There is a commercially available NBI system (Olympus Corp, Tokyo, Japan) used to detect BC. Urologists can change the optical setting on these devices to toggle between WLC and NBI [9].To date, an increasing number of studies, which focused on evaluating the additional detection rate (ADR) of NBI for BC compared with WLC, have been published with a variety of findings. However, there was only one metaanalysis included seven studies (data search up to April 2012)compared the detection rate of NMIBC between NBI and WLC [10]. After the previous meta-analysis, many relevant original studies were published. It is necessary to update the pooled ADR of NBI for BC compared with WLC with the latest evidences. Besides, there is still lack of evidence from the direct comparison of NBI and WLC for same patients, and so it is still unclear whether there is any significant advantage in the clinical use of NBI compared with WLC.
To achieve a comprehensive analysis in order to guide rational use of NBI based on the latest evidence, we performed a meta-analysis to assess the ADR of within-patient comparisons of NBI and WLC for NMIBC detection and compare the impact of NBI and WLC on bladder cancer recurrence risk.

Selection criteria
Studies were included if they met following criteria:(1) they evaluated ADR and/or recurrence rate of NMIBC; (2) they provided effective comparison groups (NBI vs. WLC); (3)they reported sufficient data including detected and total number of NBI and WLC at a per-patient or per-lesion level (a lesion was defined as a biopsy specimen or a biopsy location), or total number of subjects and recurrence rate of NBI and WLC during the follow-up period of at least three months, or provided sufficient data for their estimation. When there were multiple publications from the same population during an overlapping time period, only the study with the largest series patients was included.
Studies were excluded if: (1) they were reviews, editorials, opinions, animal models or case reports; (2)they only evaluated the ADR and/orrecurrence rate of NBI or WLC for NMIBC; (3) no sufficient data of ADR and/or recurrence rate could be extracted; (4) patients were undergoing the procedure without pathological confirmation of lesions.

Data selection and extraction
Citations were merged in EndNote version X7 (Thomson Reuters) to facilitate management. Two authors (Li JD and Ma SJ) evaluated all retrieved articles by title and abstract in an unblinded standardized manner, to determine whether a paper met the inclusion criteria. Studies that fulfilled the inclusion criteria after full-text screening were finally included in quantitative synthesis. We extracted relevant data from each eligible study for first author, study year, country of origin, study setting, number of enrolled patients or lesions, sex ratio, detected and total number of NBI and WLC, recurrence, and total number of patients in NBI-TUR and WLC-TUR groups. Data for per-patient and per-lesion analyses were extracted separately whenever available. Data extraction was by two authors (Li JD and Ma SJ) independently and consensus was reached on all items.

Quality assessment
The quality and risk of bias of included diagnostic studies were assessed using Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) [12]. The QUADAS-2 tool consists of four key domains including patient selection, index test, reference standard, flow and timing. Risk of bias was judged as "low", "high", or "unclear".

Statistical analysis
ADR was defined as the additional number (patients or lesions) of NBI detected divided by the total number (patients or lesions) of NBI and WLC detected. The between-study heterogeneity was estimated using the I 2 statistic. Significant heterogeneity was defined as I 2 exceeding 50%. Pooled results of ADR, relative risk (RR), sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR) and corresponding 95% confidence intervals (CIs) were calculated using the fixed effects model (Mantel and Haenszelmethod) when heterogeneity was not significant (I 2 <50%). Otherwise, a random-effects model was applied. Forest plots were constructed for visual display of pooled results if necessary. Publication bias was examined using Egger's linear regression text and "trim and fill" method. Meta-regression was applied to detect the potentially important covariates exerting substantial impact on betweenstudy heterogeneity. Statistical analyses were conducted using Meta-Disc software (version 1.4; Unit of Clinical Biostatistics, Ramony Cajal Hospital, Madrid, Spain) [13] and STATA 12.0 (Stata Corp LP, College Station, TX, USA).

Quality assessment
Results of the assessment of study quality are shown in S1 Table. In three studies[22,26,28], risk of bias in patient selection was unclear, as unable to determine whether the patients in these studies were continuously enrolled or not. One study[24] had a high risk of bias in the patient selection as included patients had confirmed NMIBC before endoscopy. As for the index test item of NBI in applicability concerns, three studies [15,16,32] scored a high risk of bias, in which NBI was followed by WLC to identify positive lesions and investigate whether any additional bladder was available.

Heterogeneity analysis
Between-studies heterogeneity of ADR in per-patient and per-lesion analysis was explored by meta-regression. We included four variables: (1) study design (prospective or retrospective); (2) study type (full text or abstract); (3) study center (single or multiple) and (4) number of patients (<50or !50) or lesions (<100 or !100).The meta-regression analysis did not reveal any factor that contributed to the heterogeneity.

Discussion
This meta-analysis synthesized published evidence about the ADR of within-patient comparisons of NBI and WLC for NMIBC diagnosis and therapeutic outcome. Our results indicated that NBI increased NMIBC detection by 9.9% at the per-patient level and 18.6% at the perlesion level. The pooled sensitivity of NBI was significantly higher than WLC both at the perpatient (95.8% vs. 81.6%) and per-lesion levels (94.8% vs. 72.4%). In addition, NBI significantly reduced the recurrence rate of BC after TUR with a pooled RR value of 0.43 and 0.81 at month 3 and 12, respectively. Compared with the previous meta-analysis conducted by Li et al. [10], the pooled ADRs of NBI for NMIBC diagnosis in our study were lower both at per-patient level [17% (95% CI:0.10-0.25) vs. 9.9% (95% CI: 0.05-0.14)] and per-lesion level [24% (95% CI:0.17-0.31) vs. 18.6% (95% CI: 0.15-0.25)].Although the pooled ADRs of NBI for NMIBC diagnosis in our study were lower than the previous meta-analysis [10], the results in our meta-analysis may more moderate and that could reflected the value of NBI in clinical practice as more original studies and patients were included in our study.
Because of the higher sensitivity of NBI, more tumors can be detected. Herret al. [15] reported a higher number of identified tumors per patient visualized on NBI cystoscopy (3.4) than WLC cystoscopy (2.3). Similar results were shown in a study conducted by Bryan et al.
[26], which indicated that NBI identified 2.6 tumors per patient while WLC identified only  1.9. However, several other studies showed that the number of tumors identified per patient by NBI and WLC was similar [21,22,39].
Photodynamic detection (PDD) is another new diagnostic and imaging tool for establishing a diagnosis of bladder cancer. Although sufficient evidence indicates that when compared to WLC, both PDD and NBI can improve diagnosis and reduce the recurrence rate of bladder cancer [10,[40][41][42],few trials directly compare the two techniques. A preliminary study conducted by Yoshio et al. [43]firstly reported the comparison of PDD and NBI in the same patients with flat urothelial lesions suspicious of CIS of the bladder. The results indicated that the sensitivity and specificity of PDD and NBI were similar (91.6% vs 62.5 and 82.7% vs. 87.9%).A network meta-analysis [44], assessing the therapeutic outcome of TUR in patients with NMIBC assisted by PDD or NBI, showed that resection using NBI and PDD did not differ significantly in terms of cancer recurrence rate [Hexaminolevulinic acid (HAL)-based PDD Higher sensitivity may bring a higher false-positive rate (FPR) and result in additional biopsies. The reported FPR of NBI ranged from 21.8% to 50% at per-patient level [14,15,39] and ranged from 13.6% to 39.1% at per-lesion level. [14,18,19,25,45,46] Cauberget al. [19] found FPR of NBI was significantly higher than WLC (31.6% vs. 24.5%, p<0.01) at per-lesion level. Additionally, a similar result was found by Katsunoriet al. [18] (29.1% vs. 13.8%, p<0.01). A previous meta-analysis [10] including four studies showed a slightly higher false positive detection rate (FDR) of NBI than WLC in tumor level without significant difference. Our results showed the specificity of NBI was significantly lower than WLC at per-lesion level in withinpatient comparisons. However, another meta-analysis [47] showed no significant specificity difference between NBI and WLC (84.7% vs. 87.0%). Whether the FDR and specificity of NBI are indeed different to WLC needs to be further explored.
Although NBI provides a subjective impression of abnormal areas of bladder mucosa without the use of dyes, it does not appear to have a significant associated learning curve. [26,48] Bryan et al. [26] reported that a new NBI user demonstrated a significantly improved detection rate of urothelial carcinoma (UC) with NBI compared to WLC with an ADR of 35%. When compared to an experienced user, there was no significant difference in the excess number of detected UC (p = 0.74). Herr et al. [48] also found no significant difference between new and experienced users of NBI in the detection rate of recurrent bladder tumors.
Several studies showed that NBI significantly reduced the recurrence rate of BC. [15,16,19,39] Koheiet al. [16] reported that the recurrence rate at month 12 in the NBI-TUR group was significantly lower than in the WLC-TUR group (21.1% vs. 39.7%, p = 0.016). A metaanalysis [44]including four studies showed that NBI-TUR was superior to WLC-TUR, with an RR of 0.47 (95% CI: 0.31-0.72). However, in this meta-analysis, the author did not show the short-term or long-term impact of NBI-TUR on the recurrence rate of BC. In our results, the pooled recurrence rates of NBI and WLC at month 3 and 12 were 4.6% vs. 16.7% and 26.0% vs. 38.6%, respectively. Pooled RRs for NBI at month 3 and 12 were 0.43 (95% CI: 0.23-0.79) and 0.81 (95% CI: 0.47-0.77), respectively. Limited by the small number of patients in included studies, the value of NBI as an aid to TUR in reducing the long-term recurrence rate needs further evaluation in randomized controlled trials.
Even though several studies have shown that NBI objectively improves the detection of primary and recurrent BC, there are still some controversies. One controversial area is potential observer bias. In instances where WLC and NBI were performed subsequently by the same urologist, the increased detection rate by NBI may result from the ''second look" inspection of the bladder. Herret al. [15] reported that subtle tumors recognized first with NBI also became visible when the image was switched back to WLC in several cases. In order to address observer bias, Shen et al. [21] performed NBI and WLC to detect BC in a randomized imaging sequence modality. According to the randomization protocol, the bladder was mapped using WLC then NBI, or vice versa during the same observation period. The result showedthat NBI still identified significantly more additional tumors than WLC, confirming that a "second look" did not compromise the superiority of NBI over standard WLC for detecting primary NMIBC including CIS lesions.
This meta-analysis had four limitations. First, there was significant heterogeneity for some major results. Different inclusion and exclusion criteria and observer experience bias might contribute to such heterogeneity. Although we used a random-effects model, there was still some influence on the final results. Second, quality assessment showed that not all the included studies were high quality, as some indices were labeled as high risk bias diagnostic studies, which might lead to some bias in the final statistical results. Third, studies included to evaluate the diagnostic accuracy of NBI may be uncomprehensive. Since the main objective of this study was to perform an analysis of within-patient comparisons of NBI and WLC, studies that only reported the performance of NBI were excluded. Fourth, the limited number of patients and lesions of CIS in the pooled analysis mean that the results of NBI for CIS detection should be interpreted with caution.

Conclusions
In conclusion, our meta-analysis indicated that NBI improved the diagnosis of NMIBC and CIS compared to standard WLC either at the per-patient or per-lesion level. This diagnostic test could reduce the recurrence rate of BC accordingly.