Effectiveness of urine fibronectin as a non-invasive diagnostic biomarker in bladder cancer patients: a systematic review and meta-analysis

The systemic review and meta-analysis in this manuscript were performed in accordance with the recommendations of the Quality of Reporting of Meta-analysis (QUOROM) consensus guidelines [25]. A computerized literature search was carried out in PubMed, ISI Web of Science, EMBASE, Cochrane library, and China Biology Medicine disc (CBM) on March 2, 2017, with the following key words and/or medical subject headings: “urine fibronectin” or “fibronectin” or “Fn” plus “bladder cancer” or “bladder tumour” or “urinary bladder neoplasms” or “transitional cell carcinoma,” without language restriction. Besides, references of relevant reviews and retrieved articles were checked for additional articles that were omitted through the database searches. Two investigators (XT and YS) did the searches independently.

We search the full-text articles that investigated the effectiveness of urine fibronectin for detection of human beings with bladder cancer. In our meta-analysis, we included researches that met the following criteria: (1) studies utilized urine Fn as a diagnostic test for human bladder cancer patients, (2) the bladder tumor was confirmed by pathology, (3) the urine samples were collected before the final treatments, (5) studies reported the sensitivity (Sen) and specificity (Spe) of urine Fn with their 95% confidence intervals (CIs), gave the number of true positive (TP)/false positive (FP)/false negative(FN)/true negative(TN), or provided sufficient information to calculate them, (6) studies gave clear cut-off value or cut-off criteria. The exclusion criteria were as follows: (1) insufficient data for analysis; (2) case reports, conference reports, retrospective design, reviews, letters, or editorials; (3) duplicate data published in other studies; and (4) full text unavailable.

Two investigators independently extracted the following data from each article: name of the first author, publication year, study design, study location, total sample size, mean age, number of bladder cancer cases enrolled, number of NMIBC cases, detection method of urine Fn, cut-off criteria, sensitivity, specificity, and area under curve (AUC) of the receiver operating characteristic curve (ROC). The variables including TP, FP, FN, and TN results were also collected or calculated from the data in each article. If there was any disagreement between the two investigators, a discussion among all of the authors would be carried out to resolve it. Quality assessment of the selected studies was conducted using the quality assessment of diagnostic accuracy studies-2 (QUADAS-2), a revised quality assessment tool for systematic reviews of diagnostic studies to evaluate bias in the study [26].

The data analysis was performed by STATA version 13.1 (Stata Corporation, College Station, TX, USA) with the midas and metandi modules as well as Meta-Disc 1.4 (XI Cochrane Colloquium, Barcelona, Spain), which were widely used for meta-analysis of diagnostic studies [27, 28]. RevMan 5.3 (Cochrane Collaboration, Oxford, UK) was used to do the QUADAS-2 assessment. The Sen, Spe, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and diagnostic score, with corresponding 95% CIs, were pooled in order to estimate the diagnostic power of urine Fn in the diagnosis of BCa. We also calculated the summary receiver operating characteristic (SROC) curves and area under the curve (AUC) of urine Fn with 95% confidence contour. To better judge the test performance, hierarchical summary receiver operator curve (HSROC) was constructed by using the metandi module in STATA, which is a flexible method for meta-analysis of diagnostic test accuracy evaluations [29] and is widely used in high-quality researches [30]. Besides, Z test was used to judge the diagnostic performance of urine Fn and the combined method (Fn+Cyto). Heterogeneity in the meta-analysis was appraised using Cochran Q test (with p value) and the I² index, with statistically significant heterogeneity set at P < 0.05 and I² > 50% [31]. To further analyze the sources of heterogeneity, Spearman correlation coefficient test was conducted to rule out the heterogeneity caused by the threshold effect. Moreover, sensitivity analysis, univariable meta-regression, and subgroup analyses were performed. To make better clinical decisions, we calculated post-test probability of bladder cancer through combining the pre-test probability with the likelihood ratios using the Fagan’s nomogram. Deeks’ funnel was used to investigate the publication bias, and plot P < 0.05 indicated significant asymmetry [32]. Begg’s rank correlation test was performed as a supplement to estimate the publication bias.

The primary search identified 741 articles from all databases. After excluding the duplicate publications, non-relevant literatures, and articles that did not meet the inclusion criteria, eight manuscripts published between 2000 and 2013 were considered eligible for meta-analysis (Fig. 1) [17‐24]. Among the studies, five of them reported the diagnostic value of both urine Fn itself and urine Fn combined with urine cytology (Fn+Cyto) [18, 20, 22‐24]. So, we additionally selected those five studies to further evaluate the diagnostic performance of the combination of these two methods (Fn+Cyto).

Characteristics of the included studies are shown in Table 1. All studies were carried out from 2000 to 2013 and varied in sample sizes (from 122 to 355), and a total of 744 bladder cancer patients pathologically confirmed at biopsy were included. Of these studies, two were conducted in Asia, two in Europe, and four in transcontinental regions. The pathological types of bladder tumor in three studies [18, 22, 23] consisted of both bladder transitional cell carcinoma (BTCC) and squamous cell carcinoma (SCC) and the rest only included BTCC. In all, three studies gave the volumetric urine Fn concentrations (measurement unit: μg/L), three only reported the Fn cut-offs adjusted by urine creatinine (Cr) (measurement unit: ng/mgCr or ng/μmolCr) in order to eliminate the bias caused by concentrated or attenuated urine samples, and Eissa [23] chose another calibrator—bovine serum albumin (BSA) rather than frequently used urine Cr in one of his studies (ng/mgBSA). Different assay methods were used while the enzyme-linked immunosorbent assay (ELISA) was the most frequently selected. The ROC curves together with respective AUC of urine Fn were reported in six studies. Moreover, we also listed the outcome variables including TP, FP, FN, and TN in Table 1.

We used the QUADAS-2 tool to do the study quality assessment, and the results were shown in Additional file 1: Figure S1. The bar graph (Additional file 1: Figure S1A) indicated that the overall quality of the included studies was moderately high. The potential risk of bias mainly came from the patient selections of some studies. To be more precise, the non-random or inconsecutive inclusion of patients might slightly reduce the study quality of this meta-analysis. Additional file 1: Figure S1B listed the performance of each research in the assessment.

In our meta-analysis, based on the included studies, the overall pooled sensitivity was 0.80 (95%CI = 0.77–0.83) and the pooled specificity was 0.79 (95%CI = 0.73–0.84), with a DOR of 15.18 (95%CI = 10.07–22.87) and a diagnostic score of 2.72 (95%CI = 2.31–3.13), shown in Fig. 2. The sensitivity and specificity of urine Fn were both high, indicating that this biomarker could not only correctly detect the patients with bladder cancer but effectively avoid the FP cases. The pooled PLR was 3.85 (95%CI = 2.94–5.04), and the NLR was 0.25 (95%CI = 0.21–0.30), shown in Fig. 2e, f, respectively. We also showed the SROC (midas module in STATA) and HSROC (metandi module in STATA) results in Fig. 3a, b. The AUC of SROC was 0.83 (95%CI = 0.79–0.86), which was far greater than 0.70—a reference value of the useful risk predictor in diagnostic tests [33]. The Q* index of this SROC, which referred to the Sen of the point where its Sen was equal to Spe, was 0.798, which further verified the diagnostic accuracy. From both the SROC and HSROC, we could tell that the 95% confidence region of the summary operating point were quite narrow, which illustrated that the confidence level of this summary operating point as a sample to estimate the overall population was sufficiently high. The HSROC also showed a rather restricted 95% prediction region, which was the confidence limit for an individual predicted value of the summary point of urine Fn Sen and Spe. In this study, the pooled pre-test probability was 0.47. Fagan’s nomogram (Fig. 4) of this meta-analysis showed that for patients who were under suspicion for bladder cancer, the probability of developing this disease increased to 77% when urine Fn testing was positive and reduced to 18% when urine Fn testing was negative.

There was no significant heterogeneity between studies as evidenced by a Q test p = 0.061 and an I² index = 54%. Specifically, there was nearly no heterogeneity in Sen (Q test p = 0.14, I² = 36.81%) while an obvious heterogeneity was observed in Spe (Q test p < 0.05, I² = 66.45%). Spearman correlation coefficient of these eight articles was − 0.542 (p = 0.183), suggesting there was no significant threshold effect in this meta-analysis. In order to analyze the sources of the heterogeneity in detail and have a better view of the impact of various study characteristics on the diagnostic efficacy of urine Fn, meta-regression along with subgroup analysis were conducted.

We considered the (1) urine Fn units of measurement (μg/L or not), (2) the pathological types of bladder cancer (BTCC only or BTCC and SCC), (3) the study objects (primary tumor or residual tumor), (4) the assay methods (ELISA or not), and (5) the proportion of NMIBC in all bladder cancer cases (recorded as > 50% or no evidence) as the potential sources of heterogeneity. First, we used the above-mentioned five covariates to do an univariable meta-regression, and the results were shown in Additional file 2: Figure S2. Except for the covariate “study objects,” the p values of other four covariates were all < 0.05 in both Sen and Spe, indicating that those four covariates were the main sources of the heterogeneity. The pooled Sen, Spe, DOR, PLR, NLR, and AUC with their 95% CIs for each subgroup were listed in Table 2 and the respective I² were also calculated.

For measurement units, studies using urine Fn concentration (μg/L) significantly reduced the I² index of Spe to less than 70%, without obviously increasing the heterogeneity of Sen. And the diagnostic efficacy of urine Fn using its concentration as the measurement unit was significantly higher than that using other measurement units, with an AUC of 0.88 (95%CI = 0.80–0.94) versus 0.80 (95%CI = 0.76–0.83). As for pathological types, studies based on both BTCC and SCC have a higher Sen with a lower Spe than the studies which only included BTCC cases. Moreover, the “BTCC&SCC” subgroup had an improved AUC [0.89 (95%CI = 0.86–0.92)] than the overall pooled AUC [0.86 (95%CI = 0.83–0.89)], indicating that the urine Fn can be used as a biomarker for not only BTCC. It might also be efficient for detecting SCC. Besides, the AUC of “NMIBC > 50%” were larger than 0.85, which means the Fn might have a very high diagnostic accuracy especially in NMIBC. To evaluate the credibility and consistency of the results, we performed sensitivity analysis by omitting included studies one by one. With sequential removal of each single study, the overall results were essentially unchanged, which further confirmed the robustness of our outcomes.

Among the 11 eligible studies, five also reported the diagnostic value of urine Fn combined with urine cytology (Fn+Cyto) [18, 20, 22‐24]. So, we also collected the pooled meta-analysis of bladder cancer patients with the combined testing methods (Fn+Cyto), and the results were summarized in Table 3. In order to better compare the diagnostic efficacy of urine Fn alone and the combined method, we further conducted Z tests of these two testing methods and the Z values together with respective p values were also listed in Table 3. Among those parameters, the combined method (Fn+Cyto) was significantly more sensitive than urine Fn alone, with a sensitivity of 0.86 (95%CI = 0.82–0.90) versus 0.79 (95%CI = 0.77–0.82) (p < 0.01). Moreover, the AUC of the combined method [0.89, 95%CI = (0.86–0.92)] was also significantly larger than that of urine Fn alone [0.82, 95%CI = (0.78–0.85)] (p < 0.01), indicating that the combination of Fn and cytology had a much better diagnostic performance than urine Fn alone. However, the Spe and NLR of urine Fn were higher than those of the combined method (p < 0.05).

Deeks’ funnel plot asymmetry test was used to do the publication bias analysis, which showed a symmetric funnel plot in Fig. 5, meaning that there was no significant publication bias among the included studies (Deeks’ test p = 0.44). Besides, the Begg’s test was performed and the z was 1.36 and p value was 0.174, which also proved that no evidence of publication bias for this meta-analysis was observed.