Prognostic value of microRNAs in pancreatic cancer: a meta-analysis

Background: The prognostic impact of microRNA (miRNA) expression levels in pancreatic cancer (PC) has been estimated for years, but the outcomes are controversial and heterogeneous. Therefore, we comprehensively reviewed the evidence collected on miRNA expression in PC to determine this effect. Results: PC patients with high miR-21 (HR=2.61, 95%CI=1.68-4.04), miR-451a (HR=2.23, 95%CI=1.23-4.04) or miR-1290 (HR=1.43, 95%CI=1.04-1.95) levels in blood had significantly poorer OS (P<0.05). Furthermore, PC patients with high miR-10b (HR=1.73, 95%CI=1.09-2.76), miR-17-5p (HR=1.91, 95%CI=1.30-2.80), miR-21 (HR=1.90, 95%CI=1.61-2.25), miR-23a (HR=2.18, 95%CI=1.52-3.13), miR-155 (HR=2.22, 95%CI=1.27-3.88), miR-203 (HR=1.65, 95%CI=1.14-2.40), miR-221 (HR=1.72, 95%CI=1.08-2.74), miR-222 levels (HR=1.72, 95%CI=1.02-2.91) or low miR-29c (HR=1.39, 95%CI=1.08-1.79), miR-126 (HR=1.55, 95%CI=1.23-1.95), miR-218 (HR=2.62, 95%CI=1.41-4.88) levels in tissues had significantly shorter OS (P<0.05). Conclusions: In summary, blood miR-21, miR-451a, miR-1290 and tissue miR-10b, miR-17-5p, miR-21, miR-23a, miR-29c, miR-126, miR-155, miR-203, miR-218, miR-221, miR-222 had significant prognostic value. Methods: We searched PubMed, EMBASE, Web of Science and Cochrane Database of Systematic Reviews to recognize eligible studies, and 57 studies comprising 5445 PC patients and 15 miRNAs were included to evaluate the associations between miRNA expression levels and overall survival (OS) up to June 1, 2019. Summary hazard ratios (HR) with 95% confidence intervals (CI) were calculated to assess the effect.


INTRODUCTION
Much effort has been made over a long period of time to identify prognostic biomarkers in pancreatic cancer (PC) patients. Fortunately, a large body of literature has covered the survival of PC patients with abnormal microRNA (miRNA) expression . Among all kinds of human cancers, PC has one of the worst prognoses, with a 5-year overall survival (OS) rate of lower than 5% [170]. Despite advances in clinical treatments and new surgical techniques, the survival rate of PC patients has been low for more than 30 years [171]. PC is highly aggressive; therefore, distant metastasis and tissue invasion may occur at early stages AGING [172]. Since invasion and metastasis are the biggest obstacles to effective treatment of PC, it is imperative to explore the molecular biological mechanism leading to such invasive behavior to improve the survival time of patients. miRNAs are small noncoding RNAs involved in gene regulation [173]. In cancers, a few upregulated miRNAs can serve as oncogenes (oncomiRs) [174], and downregulated miRNAs can serve as tumor suppressors [175]. Expression profiling data analyses have revealed signatures of diagnosis and prognosis that have been employed to stratify various tumor types [174,176]. As a consequence, miRNAs have the potential to turn into clinical biomarkers for human tumors and into molecular therapeutic targets [177].
Despite comprehensive studies focused on illustrating the molecular biological mechanisms in PC, there are still challenges confronting the identification of minimally invasive and sensitive biomarkers of prognosis. Consequently, it is of vital significance to find prognostic signatures that can be conveniently and reliably applied in the clinical setting to improve the survival time of PC patients.
Increasing evidence indicates that miRNAs have the potential to act as PC prognostic biomarkers in clinical practice . Regrettably, there has not been a meta-analysis to evaluate the relationship between dysregulated miRNA expression and survival in PC patients. In view of our previous work, meta-analyses of miRNA expression and cancer patients [178,179], it is necessary to conduct the current work by searching the recently published literature about miRNAs as prognostic tools in PC tissue or blood.

Meta-analysis
An overview of the HR with 95%CI obtained from the overall comprehensive analysis for all included miRNAs is shown in Table 1. Based on the logical order of the miRNA names, the forest plot, Begg's funnel plot, sensitivity analysis and funnel plot of the merged analysis adjusted with the trim and fill method are shown in Figures 1-7. The mean NOS score of the included studies was 7.0 (5.0-8.0), indicating that their quality was adequate (Table 2).

High miR-21 levels in tissues predict poor OS (multivariate analysis)
The details are shown in Table 1 and Figure 3.

Publication bias
Begg's funnel plot was employed to estimate publication bias in the study of OS in PC patients with high tissue miR-21 levels ( Figure 4). The results showed that the P value was less than 0.01, indicating the presence of publication bias.

Sensitivity analysis
Sensitivity analysis was used to estimate whether any single study had undue influence on the OS of PC patients with high tissue miR-21 levels ( Figure 5). The outcome showed that no single investigation significantly affected the pooled HR and 95%CI.

The trim and fill method
As such (Figure 4), the trim and fill method was conducted, and the pooled HR was recalculated with assumed lost studies to assess dissymmetry in the funnel plot ( Figure 6), manifesting no publication bias AGING

Foremost findings
The current meta-analysis included 57 English articles that incorporated 15 miRNAs and 5445 patients. As the most researched miRNA, PC patients with high blood or tissue miR-21 levels had significantly poorer OS than those with low levels. It also proved true among PC patients with high tissue miR-21 levels (multivariate analysis) and pooled analysis adjusted with the trim and fill method of OS, indicating that miR-21 is a stable and useful prognostic biomarker in PC. Moreover, a few other miRNAs had significant prognostic impact on PC, including blood miR-451a, and miR-1290 and tissue miR-10b, miR-17-5p, miR-29c, miR-126, miR-155,     miR-203, miR-218, miR-221, and miR-222. Among these, blood miR-21, and miR-451a and tissue miR-23a, miR-155, and miR-218 were strong biomarkers of prognosis for PC.

Altered expression, potential targets and pathways for studied miRNAs
In addition, an overview of the 15 miRNAs with dysregulated levels, covering the validated targets and pathways, is shown in Table 3. Most of the included miRNAs showed stable expression levels, higher or lower than the control groups except miR-200c. In brief, Table 3 could support a better understanding of the molecular biological mechanisms of miRNAs in PC.

Superiorities of the meta-analysis
The present work had two strengths: (1) we looked for and found out almost all studies with OS in PC patients with dysregulated miRNA levels. In addition, the recent miRNA expression pattern is shown in Tables 4 and 5 that differentiates miRNA names and the sample types.
(2) The majority of included articles had large sample sizes (≥30, all but 4 studies [6,41,121,125]), intensifying and widening the applicability of the prognostic outcomes for PC patients.
www.aging-us.com 9389 AGING disease-free and recurrence-free survival; (4) the prognostic impact of miRNA expression levels in pancreatic cancer should be adjusted for risk factors that have an important influence on pancreatic cancer prognosis, such as age, educational level, sex, smoking, obesity, heavy alcohol intake, underlying illnesses and family history of cancer, which indicates possible mutations. However, the searched papers may not all contain the very concerned information. Therefore, the impact of bias in predicting miRNAs involved in pancreatic cancer prognosis may occur due to the lack of adjustment for risk factors in a rigorous conclusion.

Insight for future clinical and experimental studies
Notably, this study was the first meta-analysis of the associations between abnormal miRNA levels and prognosis in PC patients. This study provides direction for further clinical and experimental study: (1) joint detection of various miRNA levels could be utilized by clinical workers and other health care providers, which might extremely expand the ability to assess the prognosis of PC patients such that immediate treatment might be supplied; (2) advances and trends regarding miRNA expression levels and the survival time of PC patients could be obviously acquired by the experimental researchers mentioned in Tables 4 and 5.
In addition, miRNA molecular mechanisms could be obtained by assessing the data in Table 3; and (3) several contradictory outcomes concerning the prognostic value of miRNAs might be resolved on account of the present work.

Inclusion criteria
The inclusion criteria were as follows: (1) articles on the correlation between miRNA expression level and survival time of PC patients; (2) inclusion of estimated OS outcomes; and (3) full-text in English.

Exclusion criteria
The exclusion criteria were as follows: (1) articles without original data (reviews, letters or laboratory studies); (2) nondichotomous miRNA level; and (3) frequency of studies evaluating OS of miRNA expression level equal or less than 2 in tissue. In addition, on the condition that more than one article was published on the same subjects, the most well-rounded paper was chosen for the present work. Likewise, if both univariate and multivariate analysis of OS were covered, the latter was chosen, as this type of analysis considers interferential factors.

Quality assessment
Fei Zhao and Chao Wei confirmed all qualified studies that analyzed the prognostic value of miRNAs in PC, and Yue Zhang reevaluated undetermined information. Quality assessment for each paper was performed employing the modified Newcastle-Ottawa Scale (NOS) [180]. NOS scores were calculated according to selection, comparability, and outcome. Articles with NOS scores ≥6 were considered highquality articles [181].

Study selection
The flow chart with details of the study selection process is given in Figure 8.

Study frequency
The frequency of studies estimating the OS of PC patients with and miRNA expressions of PC patients is presented in Tables 4 (blood) and 5 (tissue), and includes the miRNA names, the frequency of included miRNAs, and the reference number.

Study characteristics
The fundamental particulars of the included literature are fully listed in Table 6. On the condition that the data were not offered in the article but just as Kaplan-Meier survival curves, the data were abstracted from the curves, and the generation of HR with 95% CI was next AGING carried out employing the software Engauge Digitizer version 4.1.

Statistical analysis
All analyses were carried out employing Stata version 13.0 (StataCorp, College Station, TX, USA). OS was the primary and unique guideline for the prognosis of PC patients with miRNAs. The HR was regarded as significant at the P <0.05 level in case of the 95% CI not including the value 1. Furthermore, a single miRNA was considered a strong candidate if its HR was over 2.
Most analyses used random-effects models other than fixed-effects models because of the dissimilarity of sample types from PC patients at dissimilar stages, cutoffs, and miRNA methods in single studies. Begg's funnel plot was used to estimate publication bias. A two-tailed P value less than 0.05 was regarded as significant. If publication bias occurred, the trim and fill method was conducted. The sensitivity analysis was employed to assess how sensitive the entire effect size was to remove the impact of single investigations. If the point estimation was outside of the 95% CI of the entire effect value after it was excluded from the entire analysis, a single study was deemed to have undue influence.

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.