The Diagnostic Value of MicroRNAs as a Biomarker for Hepatocellular Carcinoma: A Meta-Analysis

Background Recently, the role of microRNAs (miRNAs) in diagnosing cancer has been attracted increasing attention. However, few miRNAs have been applied in clinical practice. The purpose of this study was to evaluate the diagnostic efficacy of miRNAs for hepatocellular carcinoma (HCC) at early stages clinically. Methods A literature search was carried out using PubMed, Web of Science, and EMBASE databases. We explored the diagnostic value of miRNAs in distinguishing HCC from healthy individuals. The quality assessment was performed in Review Manager 5.3 software. The overall sensitivity and specificity and 95% confidence intervals (CIs) were obtained with random-effects models through Stata 14.0 software. And heterogeneity was assessed using Q test and I2 statistics. Meta-regression and subgroup analyses were conducted based on the sample, nation, quality of studies, and miRNA profiling. The publication bias was evaluated through Deeks' funnel plot. Results A total of 34 studies, involving in 2747 HCC patients and 2053 healthy individuals, met the inclusion criteria in the 33 included literature studies. In the summary receiver operating characteristic (sROC) curve, AUC was 0.92 (95% CI, 0.90–0.94), with 0.84 (95% CI, 0.79–0.88) sensitivity and 0.87 (95% CI, 0.83–0.90) specificity. There was no publication bias (P=0.48). Conclusion miRNAs in vivo can be acted as a potential diagnostic biomarker for HCC, which can facilitate the early diagnosis of HCC in clinical practice.


Introduction
Hepatocellular carcinoma (HCC) accounts for more than 90% of primary liver cancers. It is one of the most common malignant tumors in the world and the third leading cause of cancer-related death [1], with an increasing incidence rate in the United States [2]. However, HCC is often diagnosed at an advanced stage, leading to limited treatment [3,4] and poor prognosis, with a median overall survival of 6-20 months and 5-year survival rate of 3% [5].
HCC is often confirmed by pathological biopsy [6] and immunohistochemistry [7]. However, these methods have high invasiveness, leading to limited clinical application for early HCC screening. e alpha-fetoprotein (AFP) is the most common serum marker in the clinic for HCC routinely screening. However, it is not accurate for the diagnosis of HCC [4]. Given the cutoff value of AFP was 20 μg/L [8], less than 400 μg/L [9], the AFP has limited diagnostic efficacy.
Besides, the imaging techniques are also usually used to early screen HCC patients, including ultrasound (US), computed tomography (CT), and magnetic resonance imaging (MRI) [10]. However, the diagnostic accuracy of the imaging techniques mainly relies on the size of the nodules, and these techniques are insensitive to small HCC nodules [10]. Because of the limited diagnostic value in the previous methods, it is urgent to discover a biomarker for the diagnosis of early HCC. Recently, many studies were focused on the role of microRNAs (miRNAs) in HCC.
miRNAs are highly conservative noncoding RNA, with 19∼25 nucleotides [11,12], resulting in mRNA degradation or inhibiting transcript through combining with mRNA [13,14], which play an essential role in the formation mechanism of tumor [12,15,16]. Moreover, it is characterized by stable in serum or plasma [17], laying a foundation for serum/plasma miRNAs in the diagnosis of tumors. In recent years, miRNAs have higher diagnostic accuracy in various cancers, such as glioma [18], prostate cancer [19], breast cancer [20], renal cell carcinoma [21], colorectal cancer [22], non-small-cell lung cancer [23], and HCC [24,25]. When it comes to the diagnosis of HCC, Shaker et al. demonstrated that the expression levels of miR-221 and miR-101-1 could be used as noninvasive biomarkers for the diagnosis of early HCC from HCV patients [26]. Zekri et al. showed that miRNA panels might distinguish early HCC from liver cirrhosis (LC), chronic hepatitis C (CHC), and healthy individuals combining with the AFP [27]. And miR-26a was identified as a promising biomarker for the diagnosis of early HCC by Zhuang et al. [28]. However, the diagnostic efficacy of miRNAs in HCC was different.
e race of participants, study design, sample type, the types of miRNAs, the size of the sample, and the background of HCC might be the main cause of inconsistent results in different studies. erefore, it was imperative to systematically and comprehensively analyze the differences among these studies. And this article aimed at estimating the overall diagnostic efficacy of circulating miRNAs in HCC.

Study Search Strategies.
We searched the related studies about the diagnosis of miRNAs for HCC in the Web of Science, EMBASE, and PubMed databases.
ere was no limit to the published time, languages, and sample source, with the deadline of searching articles (October 08, 2019). To verify the effectiveness of the study, we also manually searched the review papers and other relevant references. e search terms were found in the website of http://fmrs. metstr.com/index.aspx. "Carcinoma, hepatocellular," "microRNAs," and "diagnosis" were inputted in http:// mesh.metstr.com/, where the entrance words were obtained for literature retrieval.

Study Selection.
Inclusion and exclusion criteria to screen the literature were developed. A study can be included if it met the following criteria: (1) the studies were focused on the expression of miRNAs between HCC patients and healthy controls (HCs); (2) the difference in miRNA expression levels was statistically significant; (3) the data in studies must be complete, including sample size of two groups and sensitivity and specificity evaluation indexes to calculate the value of true positive (TP), false positive (FP), false negative (FN), and true negative (TN) or TP, FP, FN, and TN were directly given in the studies; (4) the purpose of studies was related to the diagnosis of HCC. In addition, the exclusion criteria were described below: (1) the studies were review, systematic evaluation, or meta-analysis; (2) the studies were duplicate; (3) the studies were focused on animal studies and cell culture, without case-control groups of humans; (4) the studies were the abstract of literature, letter to editors, or meetings; (5) the studies lacked complete information; and (6) the papers concentrated only on the survival, treatment, and prognosis of HCC, without involving in the diagnosis of HCC.

Quality Assessment.
e qualitative evaluation of quality assessment was performed using QUADAS-2 tools for diagnostic studies, which included four domains: Patient Selection, Index Test, Reference Standard, and Flow and Timing [29,30]. And two reviewers (Yao Jiang and Yiqin Li) performed the assessment separately. When encountering the divergence on the same literature, we invited a third individual (Jimin He) to discuss and solve the problem together.

Data Extraction.
e data were extracted by two reviewers (Yao Jiang and Yiqin Li). e following contents need to be extracted: the first author, publication year, country, sample size, age ± standard deviation (SD), proportion of males, miRNAs categories, area under the curve (AUC), sensitivity, specificity, detection method, internal reference, and cutoff value. Besides, we pooled multiple groups of miRNAs in a single study using Meta-disc 1.4 software (https:// meta-disc.software.informer.com/1.4/). Finally, the data were obtained in each study on the basis of the same specimen source, including the value of TP, FP, FN, and TN.

Statistical Analysis.
Meta-disc 1.4 and Stata 14.0 software were used for all statistical analysis. And P value less than 0.05 was considered statistically significant. Pooled sensitivity and specificity statistical indicators were analyzed using a random-effects model. e overall diagnostic efficacy was evaluated by the summary receiver operating characteristic (sROC) curve. e threshold effect was investigated based on the Spearman correlation coefficient and P value. And the heterogeneity was assessed using I 2 and chi-square test. When the value of I 2 >50% and P value <0.05, the heterogeneity exits. e value of I 2 is 0-40%, 40-70%, and 70-100%, which indicate the low, medium, and high heterogeneity, respectively [31]. In addition, meta-regression and subgroup analyses were further applied to explore the potential sources of heterogeneity. e AUC was an index of diagnostic efficacy, the values ranging from 1.0 to 0.5. e closer the AUC is to 1.0, the better the diagnostic efficacy. At last, Deeks' funnel plot was used to analyze the potential publication bias [32,33].

Study Selection.
A total of 2410 related studies were found through literature retrieval, 477 of which were duplicated literature. Ultimately, 33 papers were selected for meta-analysis according to the inclusion and exclusion criteria.
e flowchart of study selection is shown in Figure 1.

Study Characteristics.
e 33 papers included 34 eligible studies, which were published between 2010 and 2019. e data involved in 2747 patients with HCC and 2053 HCs. All the studies were published in English. e characteristics of the included studies are shown in Table 1. And the diagnostic efficacy of miRNAs for HCC in the included studies is shown in Table S1. 2 BioMed Research International

Quality Assessment.
According to the QUADAS-2 tools, we provided an overview of the quality assessment for these studies. In the Index Test aspect, there existed a high risk of bias and applicability concern due to presetting the threshold. Figure 2 shows the details of the quality assessment form.

Comprehensive Analysis.
e threshold effect was evaluated by the Spearman correlation coefficient (− 0.537), with the P value of 0.001, showing that the threshold effect existed. en we analyzed the pooled sensitivity and specificity of miRNAs, with 0.84 (95% CI, 0.79-0.88) sensitivity, 0.87 (95% CI, 0.83-0.90) specificity (Figure 3), and 0.92 (95% CI, 0.90-0.94) AUC in sROC curve ( Figure 4). Heterogeneity was found, with I 2 of 88.13% in sensitivity and 82.22% in specificity, indicating that the heterogeneity was significant. Subsequently, subgroup analyses were conducted to explore the possible sources of heterogeneity.

Subgroup Analyses.
We divided these studies into four subgroups, including sample, nation, quality, and miRNA profiling. In the nation subgroup, we divided the studies into three groups: China, non-China, and Egypt groups. We found the studies on the Egypt people had superior diagnostic efficacy, with 0.91 (95% CI, 0.79-0.96) sensitivity, 0.92 (95% CI, 0.84-0.97) specificity, 0.97 (95% CI, 0.95-0.98) AUC, 12.0 (95% CI, 5.2-27.5) positive likelihood ratio (PLR), 0.10 (95% CI, 0.04-0.24) negative likelihood ratio (NLR), and 119 (95% CI, 24-592) diagnostic odds ratio (DOR), showing miRNAs had better diagnostic ability for HCC in Egypt. e studies were divided into low-or highquality subgroup according to the result of quality assessment.  Table 2 shows the detailed results of subgroup analyses.  specificity, PLR, NLR, and DOR were more than 70% ( Figure S2), we explored the source of heterogeneity. e P values were 0.0410, 0.9808, 0.3906, and 0.5372 in quality, sample, nation, and miRNA profiling subgroups, respectively. We also separately analyzed the impact of quality on the meta-analysis. e variable of quality had a P value with 0.0093, demonstrating the quality of studies was the main source of heterogeneity (Table 3). Meanwhile, we also performed meta-regression analysis using Stata 14.0 software. However, the difference in quality was not statistically significant in Figure 6.

Publication Bias.
In order to evaluate the underlying publication bias, Deeks' funnel plot was designed in Stata 14.0 software. e P value was 0.48, indicating the probability of publication bias was fairly small (Figure 7).

Discussion
Early HCC patients are usually asymptomatic [67], which made the diagnosis of HCC more difficult. When the HCC patients have obvious symptoms, such as liver pain, jaundice, refractory ascites, progressive weight loss, fever, cachexia, or very serious complications (hepatic encephalopathy), indicating that early HCC may progress into advanced stages [68,69], then the treatments will be limited. e radical hepatic resection in early HCC is one of the most effective treatments [70]. However, only 30% to 40% of HCC patients can perform radical treatment at the time of diagnosis [63]. erefore, the early diagnosis of HCC is rather important for improving the five-year survival rate of HCC patients. In recent years, miRNAs have been found to be potential biomarkers for the diagnosis of HCC [44,63]. However, the conclusions are inconsistent. erefore, we conducted this study to evaluate whether miRNAs can be used as diagnostic biomarkers for early HCC.
In this study, the overall sensitivity, specificity, and AUC were 0.84, 0.87, and 0.92, respectively, indicating that the overall accuracy was high using the circulating miRNAs as diagnostic biomarkers for HCC. In addition, the 6.5 PLR showed better diagnostic efficacy for distinguishing HCC patients from healthy individuals. e 0.18 NLR showed miRNAs had the probability of excluding the participants without HCC. e value of DOR (36, 95% CI: 20-64) showed high diagnostic efficacy in 34 studies. Subsequently, we analyzed the main source of heterogeneity and divided these studies into four subgroups. It was reported that the circulating miRNA concentrations might be associated with different ethnic groups [71]. en, we found the studies in Egypt had better diagnostic accuracy than China. And Shaker et al. found the incidence of HCC in Egypt is overall increasing, from 4% in 1993 to 7.3% in 2003 [67]. Besides, the incidence rate of HCC among the cirrhosis patients in Egypt was approximately 21% [41], which might play an important role in the diagnosis of HCC. In addition, Yang, Y et al. found the miRNAs had higher diagnostic efficacy in Asians with 22 DOR than Caucasians [72]. erefore,       BioMed Research International multiple-central studies are needed to verify our findings. Furthermore, the miRNAs were reported that they were differentially expressed in plasma and serum [73]. en we explored the diagnostic efficacy of miRNAs in serum and plasma for HCC. Intriguingly, the serum and plasma subgroup had the same AUC and 95% CI. Like our study, Yang et al. also found the diagnostic efficacy was not statistically significant in the two sample types through a meta-analysis [72]. We speculated that serum-or plasma-derived miRNAs might have little difference on the diagnosis of HCC. However, due to the lack of consensus on whether plasma or serum is more suitable for sample detection, there exist limitations to analyzing the expression level of miRNAs in both plasma and serum [74]. Subsequently, we found the high-quality studies had better diagnostic efficacy than low-quality studies, showing the quality of studies  was one of the most essential factors influencing the overall heterogeneity. And in the meta-regression, we found the difference in quality was statistically significant using Meta-disc 1.4 software, with P value <0.05. Besides, we also showed that miRNA panel had a higher diagnostic value than multiple miRNAs or single miRNA subgroup. Hung et al. showed that miRNA panel had 84.8% sensitivity and 50.0% specificity, better than single miRNA (miR-122 and let-7b) [75]. Zhou et al. also reported that miRNA panel (miR-122, miR-192, miR-21, miR-223, miR-26a, miR-27a, and miR-801) can distinguish early HCC from healthy individuals with 82.5% sensitivity, 83.5% specificity, and 0.888 AUC [63]. And Zhang et al. also showed that 3-miRNA panels (miR-92a-3p, miR-107, and miR-3126-5p) had better diagnostic accuracy with 0.975 AUC [76]. Ning et al. also demonstrated miRNA panel (miR-155, miR-96, and miR-99a) had 0.931 AUC, higher than single miRNA [48]. Meanwhile, Pascut et al. provided comprehensive profiling of miRNome in HCC patient blood and serum, which provided useful molecular markers for the diagnosis of HCC [77].
Our study had some advantages compared with previous studies. Firstly, we integrated multiple single miRNAs or   [34]. Secondly, the subgroup classifications of HCC based on the different background, such as chronic hepatitis B, chronic hepatitis C, other types of nonviral hepatitis, and liver cirrhosis, were not conducted because some studies lack the detailed information. Furthermore, these studies lacked united internal reference in RNA quantification [79]. In the included articles, most of the studies used RNA U6 (also called snRNA U6 or U6), while some studies used miR-16, miR-39, or other miRNAs as internal reference.
In conclusion, our results have shown miRNAs in vivo can be acted as a potential diagnostic biomarker for HCC, which can promote the diagnosis of early HCC in clinical practice. Additionally, we also found miRNA panel in serum or plasma may have better diagnostic efficacy than single miRNA. In addition, more high-quality and multiple-central studies are needed to verify our findings. 0.94 Specificity (95% CI) * p < 0.05, * * p < 0.01, * * * p < 0.001 Univariable meta-regression and subgroup analyses Figure 6: Meta-regression for subgroups. Univariate meta-regression and subgroup analyses of sensitivity and specificity. * , * * , and * * * show the statistically significant difference.

Data Availability
e data used to support the findings of this study are included within the article.

Disclosure
Hualin Tao and Yongcan Guo contributed equally to this work. e funding agencies had no role in study design, analysis, interpretation of results, decision to publish, or preparation of the manuscript. e content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agency.

Conflicts of Interest
e authors declare that they have no conflicts of interest.