Diagnostic significance of microRNAs as novel biomarkers for bladder cancer: a meta-analysis of ten articles

The relevant literature in the ExcerptaMedica Database (EMBASE), PubMed, Chinese Biomedical Literature (CBM) database, and Chinese National Knowledge Infrastructure (CNKI) web database (updated until December 31, 2014) were searched using a combination of the terms “microRNAs” or “miRNA” or “miR”, “bladder cancer” or “bladder tumor” or “bladder urothelial cell carcinoma” and “sensitivity” or “specificity”, “ROC curve” or “diagnosis”. In order to retrieve the most relevant studies, manual filtration of the references was also conducted.

The adopted literature should meet the following criteria: (a) the diagnostic value of miRNAs for BC detection must have been evaluated, (b) the authors must have used the gold standards (histopathological examinations), and (c) the studies must contain detailed information for constructing two-by-two tables, which comprise true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). And exclusion criteria are as follows: (a) abstract, review, comment, editorial, and case reports, (b) overlapping data, (c) studies investigating survival or prognosis of BC, (d) deficient data, even on asking the corresponding authors for more information, and (e) sample size <100. If more than one study used some common samples, only the most precise one with the most samples was selected.

Data extraction from the selected publications was done using a standardized table by our two authors independently. And the extracted data contains first author’s surname, publication time, and country of origin, sample size, age, type of case, and control groups, sample specimen, studied miRNAs, detection method, TP, FP, FN, and TN, and information needed for quality assessment. For studies involving more than one type of miRNAs or sample specimen, the data were extracted, and the study on each miRNA was considered as an independent study. The quality of each article was evaluated by the revised quality assessment of diagnostic accuracy studies (QUADAS-2) checklists [25]. All disagreements about the collected data were adequately debated by investigators and arrived at a final consensus.

For purpose of assessing the diagnostic value of miRNAs for BC, a bivariate meta-analysis was managed to obtain pooled negative likelihood ratio (NLR), positive likelihood ratio (PLR), sensitivity, diagnostic odds ratio (DOR), specificity, and corresponding 95% confidence intervals (CIs) [26]. Meanwhile, the bivariate summary receiver operator characteristic (SROC) curve was drawn and the area under it (AUC) was calculated to display the sensitivity and specificity of each study. The heterogeneity assumption between the studies were also measured through chi-square based Cochran’s Q and I ² tests. A P value <0.10 and an I ² value >50% were considered as significant heterogeneity [27, 28]. In order to examine the heterogeneity of origin, the sub-group analysis and meta-regression were performed according to the features (sample size, ethnicity, miRNA profiling, and specimen type). Finally, the potential publication bias of these literature was measured by Deeks’ funnel plot asymmetry test with the screening standard of P value <0.05 [29]. All P values were two sided and all statistical analyses in this meta-analysis were performed by using the STATA statistical software (version 11.0; Stata Corp, College Station, Texas, USA).

There were 370 relevant articles picked out after preliminary screening in databases. The flow chart illustrates the study filtration procedures used for assessing the diagnostic potential of miRNAs in BC (Fig. 1). After a careful search and selection, eight publications meeting the final selection criteria stood out. Of these, the publication from Jiang XM et al. [18] studied two different populations (training and validation population) and one publication (Tölle A et al.) [19] researched two kinds of specimens (blood and urine); each population and specimen was considered as a separated article. Thus, in this meta-analysis there were 10 articles including 31 studies investigating 1556 cancer cases and 1347 controls in total. Furthermore, 24 of the 31 studies analyzed were focused on single-miRNA assays, the remaining seven articles were focused on multiple-miRNA assays. The main features of each research are shown in Table 1. All articles were published between 2012 and 2014. Four of the articles were performed in Asian populations and six of the articles were performed in Caucasian populations. The method of detecting miRNAs in all articles was the reverse transcription polymerase chain reaction (RT-qPCR). For miRNA detection, six articles were performed on urine samples and four were on blood samples. As shown in Table 1, all the studies analyzed in this article met the criteria for a moderate-high quality score. The risk of bias and applicability concerns graph for the included articles is presented in Fig. 2.

The summarized estimates of the diagnostic accuracy of miRNAs that distinguish BC from controls are listed in Table 2. Thirty-one studies from 10 researches covering 38 types of miRNAs were analyzed. The pooled sensitivity and specificity of the miRNAs for the diagnosis of BC were 0.72 (95%CI 0.66–0.76) and 0.76 (95%CI 0.71–0.81), respectively (Fig. 3, Table 2). Besides, the pooled PLR, NLR, and DOR with 95%CIs for overall studies were 3.0 (2.4–3.8), 0.37 (0.30–0.46), and 8 (5.0–12.0), respectively. The SROC curve with an AUC of 0.80 (95%CI 0.77–0.84) displayed in Fig. 4 was for the overall study.

Sub-group analyses and meta-regression were conducted in order to reveal the inter-study heterogeneity attributed to ethnicity, miRNA profiling, specimens, and sample size (Table 2). The results suggested that the diagnostic accuracy of miRNA assays in Asian populations (SEN 0.83; SPE 0.84; PLR 5.2; NLR 0.21; DOR 25; AUC 0.90) was much better than that in Caucasian populations (SEN 0.68; SPE 0.74; PLR 2.6; NLR 0.44; DOR 6; AUC 0.76). The sub-groups based on miRNA profiling showed that multiple-miRNA assays were more accurate in detecting BC than single-miRNA assays, with a sensitivity of 0.87 vs. 0.65, specificity of 0.84 vs. 0.73, PLR of 5.6 vs. 2.5, NLR of 0.15 vs. 0.47, DOR of 36 vs. 5, and AUC of 0.92 vs. 0.74, respectively. Moreover, for studies on miRNA detection in the blood, the pooled sensitivity was 0.79, specificity was 0.90, PLR was 7.7, NLR was 0.23, DOR was 33, and AUC was 0.90. For studies using urine samples, the pooled sensitivity was 0.70, specificity was 0.72, PLR was 2.5, NLR was 0.42, DOR was 6.0, and AUC was 0.77. Thus, miRNA detection in the blood might have higher diagnostic accuracy than that in urine. However, the pooled estimates (SEN, SPE, PLR, NLR, DOR, and AUC) were close between studies conducted using large sample sizes (>40) with small sample sizes (≤40).

Furthermore, meta-regression analysis was also used to explore the potential sources of heterogeneity after adding eight pre-specified covariates (sample size, ethnicity, miRNA profiling, and specimen) to the bivariate model. The meta-regression analysis suggested that sample size (P value <0.01), ethnicity (P value <0.001), miRNA profiling (P value <0.001), and specimen (P value <0.01) could be the reasons of the heterogeneity in sensitivity and specificity between the studies (Fig. 5).

Deeks’ funnel plot asymmetry test was used to evaluate the possible publication bias in the included studies. Overall, the funnel plots showed symmetry and the slope coefficient was associated with a P value of 0.11 (Fig. 6). For the ethnicity, miRNA profiling, and specimen-based sub-group analysis, the slope coefficient showed P values >0.05 for all (not shown). All results confirmed that there was no obvious publication bias.