Introduction

When endometrial tissues (glands and stroma) are found outside the uterus, the condition is known as endometriosis. It is one of the most common diseases in women of childbearing age, and the incidence rate has increased significantly in recent years [1,2,3]. Endometriosis has a variety of clinical manifestations, the most common of which are infertility, pelvic pain, a pelvic mass, and sexual discomfort [4]. Due to the shortage of effective and simplified biomarkers, the fertility of patients with symptoms can be greatly reduced, and they may suffer for many years before their disease is confirmed. A 6–11-year delay often occurs between the onset of the disease and the final diagnosis. The symptoms of endometriosis may show many commonalities with those of a variety of diseases. On average, women consult seven gynecologists before they are diagnosed and before they begin treatment [5]. Histologically, endometriosis is a benign disease, but it is characterized by malignant features, such as hyperplasia, infiltration, metastasis, and a high recurrence rate. This disease is classified as a cancer-like entity by the World Health Organization Histologic Classification of Ovarian Tumors [6,7,8]. According to Angiolo Gadducci et al., a 2- to 3-fold higher risk of ovarian endometrioid and clear cell carcinoma has been reported in women with endometriosis [9]. If no medical intervention occurs, the condition will continue to progress. The current diagnosis primarily depends on an imaging examination, as serological markers such as CA-125 have poor specificity and sensitivity [10, 11]. To date, laparoscopic examination together with biopsy for the histological confirmation of suspicious lesions is the gold standard for diagnosis, but these are invasive and expensive. Therefore, due to the delayed diagnosis, a noninvasive diagnostic approach is urgently needed.

MicroRNAs (miRNAs) are a class of endogenous noncoding RNAs 17-25 nucleotides in length that are capable of participating in a variety of physiological and pathological processes such as cell growth and apoptosis at posttranscriptional levels, and they play an important role in gene regulatory networks [12]. MiRNAs are present extensively in body fluids, and many miRNAs in certain diseases are abnormally expressed and are easy to detect, and thus, they may become novel biomarkers for early noninvasive diagnosis of diseases [13]. Numerous studies have shown that miRNAs regulate various aspects of endometriosis, such as angiogenesis and cell migration [8, 14]. MiRNAs have also been found to be abnormally expressed in endometriotic lesions, tissues, serum, and other body fluids, and their diagnostic value has recently received much attention worldwide [15]. For example, in 2009, RO Burney et al. evaluated the differential expression of miRNAs in the eutopic endometrium of individuals with endometriosis and those without endometriosis. MiR-21-5p was found to be one of the most differentially expressed miRNAs [16]. In addition, Zong wen Liang et al. found that many miRNAs are abnormally expressed in endometriosis, including miR-200c and miR-638 as well as let-7 [17]. The increasing numbers of miRNAs, such as the let-7 family and the miR-200 family, have been identified and reported to show abnormal serological levels in patients with endometriosis, and such miRNAs can be used as novel serological markers [18, 19]. Therefore, we designed a meta-analysis to verify whether circulating miRNAs can serve as potential diagnostic markers for endometriosis.

Results

Search Results and Characteristics of the Eligible Studies

We searched the PubMed, Embase, and Cochrane library databases up to July 30, 2018, to select articles that met the criteria. In all, 111 studies were selected, of which 34 studies were duplicates. Thirty-eight articles remained after removing letters, reviews, editorials, nonclinical studies, and irrelevant articles by a primary screening of the titles and abstracts. Finally, eight eligible articles with 23 records, 329 patients, and 277 controls were included in our meta-analysis. The characteristics of the included studies are presented in Table 1.

Fig. 1–2
figure 1

Details of QUADAS-2 quality assessment of each included study (QUADAS-2 tool)

Full size image
Table 1 Characteristics of eligible studies
Full size table

Quality Assessment of the Articles

The QUADAS-2 study quality assessments are shown in Fig. 1–2. Most articles included in the current meta-analysis satisfied most of the items in the QUADAS-2 guidelines, which suggests that the overall quality of the included studies was moderate to high.

Diagnostic Accuracy of the Serum miRNAs

The data calculated based on the bivariate model are as follows: the pooled sensitivity was 0.86 (95% CI 0.79–0.90); the pooled specificity was 0.88 (95% CI 0.80–0.93) (Fig. 3–5); the pooled likelihood ratio (PLR) was 7.05 (95% CI 4.20–1.84); the pooled negative likelihood ratio (NLR) was 0.16 (95% CI 11–0.24) (Fig. 3–5); the pooled diagnostic odds ratio (DOR) was 43.46 (95% CI 19.36–97.57) (Fig. 3–5). In addition, the summary receiver operating characteristic (SROC) curve was plotted, and the area under the fitted curve (AUC) was 0.93, which suggests an outstanding diagnostic accuracy of the overall miRNAs (Fig. 6).

Fig. 3–5
figure 2

Forest plots for studies on overall microRNAs (miRNAs) used in the diagnosis of endometriosis among 23 records included in the present meta-analysis

Full size image
Fig. 6
figure 3

Summary receiver operator characteristic (SROC) curves based on miRNAs detected in blood samples. AUC area under the curve, SENS sensitivity, SPEC specificity

Full size image
Fig. 7
figure 4

Bivariate boxplot showed the existence of heterogeneity

Full size image
Fig. 8
figure 5

Diagram of (a) goodness-of-fit, (b) bivariate normality, (C). influence analysis, and (d) outlier detection. Goodness-of-fit and bivariate normality showed that random effects bivariate model is suitable. Influence analysis identified that studies of record nos.6, 9, and 12 were most dominant studies in weight. Outlier detection showed that record no. 6 may be the reason heterogeneity

Full size image
Fig. 9
figure 6

Univariable meta-regression and subgroup analyses for sensitivity and specificity of miRNAs for diagnosis of endometriosis. Pbsize30, the sample size of patient group was bigger than 30. Cbsize30, the sample size of control group was bigger than 30. Bbsize 30, the sample size of patients and control groups were bigger than 30. Predesign, prospective study. Fulverif, verification of all by same method. Consel, consecutive selection of subjects. Fast, prior fasting of at least 6 h. Subjdescr, adequate description of study subjects. Prop, blood samples were collected during proliferative phase

Full size image
Fig. 10
figure 7

Forest plots for studies on overall miRNAs used in the diagnosis of endometriosis among records (outlier record was excluded)

Full size image
Fig. 11
figure 8

Deek’s funnel plot indicates no significant publication bias (p = 0.04 < 0.1)

Full size image
Fig. 12
figure 9

Fagan’s nomogram for calculation of post-test probabilities

Full size image
Fig. 13
figure 10

Assessment of the clinical applicability of microRNAs (miRNAs) for diagnosis. Summary of positive likelihood ratio and negative likelihood ratio for diagnosis of endometriosis. LRN likelihood ratio negative, LRP likelihood ratio positive, LLQ left lower quadrant, LUQ left upper quadrant, RLQ right lower quadrant, RUQ right upper quadrant

Full size image

Most of the research fell in the middle area, and several studies fell outside, as evaluated by a two-variable box plot (Fig. 7). The I [2] values of the sensitivity and specificity were 82.43 and 86.46, respectively (Fig. 3–5). All the above findings demonstrate significant heterogeneity among the studies.

In addition, the Spearman correlation coefficient was − 0.230 (p = 0.302), with no evidence of a threshold effect. After eliminating the heterogeneity caused by the threshold effect, a sensitivity analysis was performed. As shown, the goodness-of-fit and the bivariate normality indicated that the bivariate model was suitable for this study. The impact analysis showed that the Cook’s distances of record nos. 6, 9, and 12 were all greater than 1, which indicates that these studies were the most dominant studies in weight, and record no. 6 ranked first. Anomaly testing demonstrated that record no. 6 was an outlier and was likely to be the main reason for heterogeneity (Figs. 8 and 9). After eliminating record no. 6, the I [2] values of sensitivity and specificity were reduced by 4.14% and 2.25%, respectively (Fig. 10). To further explore the reason for heterogeneity, a subgroup analysis and meta-regression were performed. The results showed that racial differences, individual miRNAs, test group samples, and fasting status had an impact on sensitivity, which was homogeneous in a certain subgroup. Furthermore, the sample size, whether blood samples were collected during the proliferative phase and whether the study was prospective had impacts on the sensitivity, which was statistically significant. The sample size and whether the test was prospective or retrospective had an impact on specificity (Fig. 9). The results of the statistical analyses are shown in Table 2. Moreover, Deek’s funnel plot revealed no publication bias in the meta-analysis (p = 0.04) (Fig. 11).

Table 2 Statistical results of meta-regression
Full size table

Discussion

Endometrial tissue that has migrated outside the uterus changes periodically with the fluctuation of ovarian hormones. Recurrent hemorrhage and slow absorption cause proliferation and adhesion of surrounding fibrous tissue. According to site differences, the condition can be divided into two groups: endopelvic and extrapelvic manifestations [27]. Most lesions occur in the pelvic cavity, a condition known as pelvic endometriosis, which often occurs on the ovary, in the peritoneum and in the recto-vaginal pouch. Extrapelvic locations include abdominal wall scars, the ureter, the bladder, the nasal mucosa, and other sites [27,28,29,30,31]. Due to the variety of locations and symptoms, the diagnosis can be difficult, which leads to a delay in treatment. Presently, the diagnosis of endometriosis mainly depends on the following methods: blood markers (e.g., CA-125, NLR) [32, 33], urinary markers (e.g., NNE, VDBP, urine peptide, CK19) [34], pelvic examination (e.g., pelvic tenderness, pelvic mass), clinical symptoms (e.g., continuous aggravation of dysmenorrhea, dyspareunia, chronic pelvic pain, infertility) [35], imaging examination (e.g., pelvic ultrasound and MRI) [36,37,38,39,40,41], and surgery (e.g., laparoscopic examination, laparoscopic surgery, and laparotomy) [42]. Vaginal ultrasound and abdominal ultrasound are important tools that are used to identify chocolate cysts of the ovary and rectal-vaginal septum lesions, and their sensitivity and specificity are above 90% [43, 44]. However, in some cases, ultrasound examinations may not reveal positive findings, such as mild to moderate lesions, minimally changed ovarian endometriosis, infertility and chronic pelvic pain likely caused by endometriosis, sexual pain likely caused by endometriosis, or nodules causing pelvic tenderness [45]. The diagnosis for these early and atypical patients may be missed, which results in an increasing severity of the condition. Until the typical symptoms are diagnosed via ultrasound, the patient experiences fertility issues and pain [46]. Therefore, the use of circulating miRNAs as noninvasive serum markers can increase the rate of diagnosis. Early diagnosis and early intervention are therefore crucial to improve female fertility, reduce pain, and improve quality of life.

In recent years, with the development of evidence-based medicine, meta-analyses have been widely applied in the field of medicine. The greatest advantage of a meta-analysis is that it can include homogeneity tests and quantitative merger analyses on multiple similar studies to improve the demonstrated benefits of preliminary conclusions [47, 48]. A meta-analysis is cheaper and even more feasible than large-scale clinical trials and therefore plays an important role in disease analysis, treatment, risk assessment, interventions, preventive measures, and health decision-making. This study had strict literature inclusion and exclusion criteria. After a quality evaluation, it was found that all the included documents were of medium to high quality, which establishes a solid foundation for the reliability of further data analysis. The accuracy of the diagnostic tests was evaluated by plotting the SROC curve followed by the AUC was calculated. The closer the AUC value is to 1, the better the diagnostic assay functions. The AUC in this study was 0.93, which indicates that miRNAs have excellent diagnostic value for endometriosis [49, 50].

We also focused on clinical applicability. In evidence-based medicine, likelihood ratios are used for assessing the value of performing a diagnostic test, and the sensitivity and specificity of the tests are used to determine whether a test result usefully changes the probability that a condition (such as a disease state) exists. The PLR of 7.05 and the NLR of 0.16 (Fig. 4) indicated that patients with endometriosis have a 7.05-fold higher chance of testing positive based on the circulating miRNAs than the controls, and 16% of individuals with endometriosis will have a negative result. Although a DOR of 1 suggests that miRNAs failed to differentiate between endometriosis and control cases, the DOR of 43.46 in our study showed that miRNAs are outstanding diagnostic biomarkers for endometriosis (Fig. 5). We also generated likelihood ratio plots to help determine the clinical applicability of the miRNA test (Figs. 12 and 13). A PLR > 10 and an NLR < 0.1 indicate high diagnostic accuracy. The summary point in this paper falls in the lower right quadrant. Based on the current research, the clinical value of miRNA for endometriosis is still limited, and further research is needed. We found that the miRNAs of record nos. 5, 6, 8, 9, and 21 had high diagnostic accuracy and clinical applicability. From the Fagan’s nomogram, we found that, when a pre-test probability of 50% was specified, the post-test probability of positivity increased to 88% with a positive likelihood ratio of 7, and the post-test probability of negativity decreased to 14% with a negative likelihood ratio was 0.16 (Fig. 12). These outcomes suggest a stable value of circulating miRNAs in the diagnosis of endometriosis.

This study also has some limitations. In this investigation, significant heterogeneity was observed. Subgroup and sensitivity analyses were performed, but the results only partially explained the causes of the heterogeneity. The cutoff values are different in the included studies, which may be one of the reasons for heterogeneity. Moreover, studies with positive results are easier to publish than those with negative results, which may exaggerate the accuracy of the overall diagnosis. In addition, we included only literature published in English, which may have affected the research results.

Materials and Methods

Study Protocol

The research protocol was drafted at the beginning of this project according to the Cochrane Collaboration format [51].

Search Strategy and Study Selection

A comprehensive search was performed to identify all studies that assessed the diagnostic accuracy of plasma- or serum-based miRNAs for endometriosis; PubMed, EMBASE, and Cochrane Library were searched up to June 1, 2018. Keywords included the following: “plasma” or “serum” or “blood,” “miRNA,” “endometriosis,” and “biomarker.” Both Medical Subject Headings (MeSH) and freestyle words were used as search terms. The reference lists of relevant studies and reviews were also searched for potentially eligible studies. (For details, see the supplemental material.)

The Criteria and Screening Process

The inclusion criteria for the primary studies were as follows: (1) the study was a diagnostic study of serum miRNAs; (2) the subjects included endometriosis patients and control groups; (3) all patients with endometriosis were diagnosed by the gold standard method (laparoscopic exam); and (4) sufficient information was available to construct 2 × 2 tables that consisted of true positives (TPs), false positives (FPs), true negatives (TNs), and false negatives (FNs). Articles were excluded based on the following criteria: (1) written in a language other than English; (2) not conducted in humans; (3) reviews, letters, and meeting records; (4) studies with insufficient data. The studies included in our meta-analysis were independently assessed by two investigators. All the selected studies were managed using EndNote X9.

We used a search strategy to retrieve the target documents from the database. We eliminated duplicate articles and then removed noncompliant documents by title, abstract, and full-text screening.

Data Extraction and Quality Assessment

The following data were extracted: first author’s name, year and country of publication, number of patients and number of controls, miRNA type being tested, and specificity and sensitivity values of the tested miRNAs. QUADAS-2 guidelines were used to evaluate the quality of the included studies. The quality assessment was managed by Review Manager 5.3 (Cochrane Collaboration, Copenhagen, Denmark).

Data Analysis

We extracted TP, FP, FN, and TN values from each study and calculated the pooled sensitivity, specificity, PLR, NLR, DOR, and 95% confidence intervals. The diagnostic value of miRNAs was assessed by plotting the SROC curve and its AUC. Deeks’ funnel plot was used to evaluate the publication bias. Chi-squared and I2 tests were used to assess heterogeneity between studies. If p < 0.1 or I2 > 50%, the heterogeneity was considered significant between studies. If heterogeneity existed, meta-regression, subgroup, and sensitivity analyses were designed to explore the sources of heterogeneity. In addition, the Spearman correlation coefficient was calculated to determine whether the heterogeneity in the meta-analysis was related to the threshold effect; when p > 0.05, there was no threshold effect. The data analyses were performed using Meta-Disc statistical software version 1.4 (XI Cochrane Colloquium, Barcelona, Spain) and STATA software (version 14.0, StataCorp, MIDAS module).

Conclusions

Taken together, our research findings suggest that circulating miRNAs are potential noninvasive diagnostic biomarkers for endometriosis. In regard to clinical applicability, large-scale, high-quality prospective, and retrospective studies should be conducted to further explore the clinical value of circulating miRNAs for endometriosis. The subgroup analysis suggests that multiple miRNA combinations may be more accurate than individual miRNAs in distinguishing individuals with and without endometriosis. Finally, the results may be affected by a patient’s menstrual cycle.