The search has been devised by two team members (D.N. and A.M.), who interrogated PubMed, Embase and Web of Knowledge databases (up to May 2019) with the following combination of key words: (“microRNA 141” or “miRNA-141” or “miR-141”) and (“circulating” or “serum” or “plasma”) and (“prostate”) and (“diagnosis” or “ROC curve” or “sensitivity” or “specificity”). The investigators first performed an initial screening of titles and abstracts, followed by full-text article reading, and selection of those fit for the meta-analysis. Disagreements were solved via discussion with a third member of the team (CM).

Research articles inclusion criteria were: (1) involved circulating (serum/plasma/whole blood/PBMCs) miR-141 as diagnostic biomarker for PCa, (2) contained data regarding sensitivity, specificity, and AUC, (3) involved a statistical comparison between miRNA levels in PCa patients’ samples vs. healthy controls and/or other benign prostate-associated diseases, and (4) were published in English language.

Research articles exclusion criteria were: (1) non-original research papers, such as conference abstracts, letters, and reviews, (2) non-circulating miRNAs for PCa diagnostic, (3) circulating miRNAs other than serum, plasma or whole blood (e.g., urine, seminal fluid), and (4) insufficient data to match our interests.

Our study protocol was not prospectively registered.

The following key data were extracted by two team members (D.N. and A.M.) from all the articles included in the meta-analysis: author; publication year; country; ethnicity of patients; number of PCa patients, healthy controls or with other (non-malignant) prostatic pathologies; type of specimen collected (serum/plasma/whole blood/peripheral blood mononuclear cells (PBMCs)); miRNA expression profiling method; sensitivity and specificity values; and AUC and 95% CI AUC.

QUADAS-2 was used by two investigators to assess the quality of each of the included research articles [18]. The four key domains, all assessed in terms of risk of bias, were: patient selection, index test, reference standard, flow and timing. The risk of bias is rated low if all signaling questions are “yes”; if any of the signaling questions are answered with “no”, a potential for bias exists. “Unclear” is used only with insufficient or improperly reported data to permit a judgment. Applicability concerns can be “high”, “low” or “unclear”. The results of the quality assessment were used to provide an evaluation of the overall quality of included studies and to investigate potential sources of heterogeneity.

Statistical analyses were performed by Stata MP 15.1 (StataCorp LP, College Station, TX, USA) statistical software using metandi and midas commands. p < 0.05 denoted statistical significance. Between-study heterogeneity was evaluated with the I² index and χ² test. A value of I² > 50 % and/or a p < 0.05 indicated heterogeneity. Overall accuracy of miR-141 assays, sensitivity and specificity were extracted from each study. True positives (TPs), true negatives (TNs), false positives (FPs) and false negatives (FNs) were calculated for each study.

The pooled results of positive and negative likelihood ratios (LR+ and LR−, respectively) and their 95 % confidence intervals (CI) were calculated using the random-effects model [19,20].

LR+ is the odds of a positive test result in a patient with PCa, whereas LR− is the odds of a positive result in an individual without the disease.

A pooled summary receiver operating characteristics (SROC) curve was constructed, and the AUC was calculated. The overall diagnostic accuracy of miR-141 can be estimated based on the aforementioned parameters. This plot includes an overall estimate of sensitivity and (1-specificity) across all studies, a 95% confidence region around this estimate and a 95% prediction region taking heterogeneity into account.

Forest plots of sensitivity and specificity estimates and their 95% CIs were constructed from every study [21,22].

Sensitivity analysis included: quantile plot of residual-based goodness-of fit; Chi-squared probability plot of squared Mahalanobis distances for assessment of the bivariate normality assumption; a spikeplot for checking for particularly influential observations using Cook’s distance; a scatter plot for checking for outliers using standardized predicted random effects (standardized level-2 residuals).

In order to detect the amount of heterogeneity between the studies regarding sample types, meta-regression analyses was conducted. Deeks’ funnel plot analysis was performed to explore the potential publication bias [23].

In order to estimate the post-test probabilities, clinical or patient-relevant utility of the diagnostic test was evaluated with Fagan’s nomogram [24].

After using the search string of interest in the databases, we have identified 30 articles that were relevant. In total, 17 of these studies did not respect the inclusion criteria: five articles were describing new methods of quantifying gene expression, therefore not being focused on differences in expression levels of miR-141, three papers used other specimen types, such as cell lines, urinary extracellular vesicles, two were review papers, two showed a downregulation of miR-141 in the serum of PCa patients compared to healthy controls, while all the others showed upregulation of miR-141, one did not include miR-141 in the analysis, and four did not present comparative results of PCa patients vs. healthy controls, but with different stages of PCa (such as localized vs. metastatic). Furthermore, from the remaining 13 publications, seven articles were excluded due to insufficient data, and finally, we selected six articles [17,25,26,27,28,29] for our meta-analysis study.

The six remaining studies comprised an overall number of 481 participants, of which 283 patients, 114 healthy controls and 84 patients with benign prostate pathologies. The included studies were published between 2008 and 2018 and were conducted in China (two studies), USA, Poland, Russia and Ireland. Among the six studies, three studies used serum samples, one used plasma, one used PBMCs and one used whole blood. We have especially selected only blood-based analysis to reduce heterogeneity.

All studies have detected and quantified the expression level of miR-141 by quantitative real-time polymerase chain reaction, all showing up-regulation of miR-141 in PCa patients’ samples compared to healthy/benign controls.

Figure 1 summarizes the flow diagram of study selection, and Table 1 shows the characteristics of the six studies included in this meta-analysis.

There are two studies with a high risk of bias in the selection domain, while two are considered to be at an unknown risk of bias. Furthermore, on the index domain, there is another study with a high risk of bias. Two pther studies have an unknown risk of bias, thus a sensitivity analysis can find them in a possible scenario with a high risk of bias, according to the QUADAS-2 system (Table 2). The risk of bias and applicability concerns graph for the included studies is shown in Figure 2.

The test for the heterogeneity among the studies showed significant heterogeneity (I² = 86.66 and 95.98% for sensitivity and specificity, respectively). For this reason, the random-effects model was used. Meta-analysis results showed the pooled sensitivity and pooled specificity of PCa by miR-141 were 0.78 (95% CI: 0.63–0.88) and 0.96 (95% CI: 0.28–1.00) in distinguishing PCa from control, respectively (Figure 3), with the area under receiver operating characteristic curve (AUC) of 0.87 (95% CI: 0.83–0.89) (Figure 4), thereby implying a relatively high diagnostic accuracy.

The average likelihood ratio of the positive and negative test result was calculated on the basis of the pooled estimates of sensitivity and specificity, and the results showed the LR+ and LR− of miR-141 in differentiating PCa from control patients were 21 (95% CI: 0.22–1844) and 0.23 (95% CI: 0.13–0.39), respectively. In addition, the mean DOR value was 89 (95% CI: 1.21–6547.37).

A meta-regression analysis was used to explore the potential source of heterogeneity in sensitivity and specificity based on sample types. Our analysis suggested that sample types (p < 0.001) might be the potential source of heterogeneity in specificity. Because of the small number of eligible studies, to fully elucidate the source of the heterogeneity, further studies would be needed.

Goodness-of-fit and bivariate normality analyses (Figure 5a,b) suggested that the random-effects bivariate model was robust for the calculation of the pooled estimates. No outlier was identified by influence analysis, and none appeared in the outlier detection plot (Figure 5c,d).

Fagan’s Nomogram for the likelihood ratio is shown in Figure 6.

In later steps of this meta-analysis, we repeated the pooled sensitivity and specificity analysis with each of the studies removed individually. We found that the final results were similar to the initial result, reflecting stability and credibility of the results. As shown in Figure 7, Deeks funnel plot asymmetry test was not significant (p = 0.88), and there was no significant publication bias among the studies included in this analysis.