If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Endometriosis, a chronic disease that afflicts millions of women worldwide, has traditionally been diagnosed by laparoscopic surgery. This diagnostic barrier delays identification and treatment by years, resulting in prolonged pain and disease progression. Development of a noninvasive diagnostic test could significantly improve timely disease detection. We tested the feasibility of serum microRNAs as diagnostic biomarkers of endometriosis in women with gynecologic disease symptoms.
The objective of the study was to validate the use of a microRNA panel as a noninvasive diagnostic method for detecting endometriosis.
This was a prospective study evaluating subjects with a clinical indication for gynecological surgery in an academic medical center. Serum samples were collected prior to surgery from 100 subjects. Women were selected based on the presence of symptoms, and laparoscopy was performed to determine the presence or absence of endometriosis. The control group was categorized based on absence of visual disease at the time of surgery. Circulating miRNAs, miR-125b-5p, miR-150-5p, miR-342-3p, miR-451a, miR-3613-5p, and let-7b, were measured in serum by quantitative real-time polymerase chain reaction in a blinded fashion without knowledge of disease status. Receiver-operating characteristic analysis was performed on individual microRNAs as well as combinations of microRNAs. An algorithm combining the expression values of these microRNAs, built using machine learning with a random forest classifier, was generated to predict the presence or absence of endometriosis on operative findings. This algorithm was then tested in an independent data set of 48 previously identified subjects not included in the training set (24 endometriosis and 24 controls) to validate its diagnostic performance.
The mean age of women in the study population was 34.1 and 36.9 years for the endometriosis and control groups, respectively. Control group subjects displayed varying pathologies, with leiomyoma occurring the most often (n = 39). Subjects with endometriosis had significantly higher expression levels of 4 serum microRNAs: miR-125b-5p, miR-150-5p, miR-342-3p, and miR-451a. Two serum microRNAs showed significantly lower levels in the endometriosis group: miR-3613-5p and let-7b. Individual microRNAs had receiver-operating characteristic areas under the curve ranging from 0.68 to 0.92. A classifier combining these microRNAs yielded an area under the curve of 0.94 when validated in the independent set of subjects not included in the training set. Analysis of the expression levels of each microRNA based on revised American Society of Reproductive Medicine staging revealed that all microRNAs could distinguish stage I/II from control and stage III/IV from control but that the difference between stage I/II and stage III/IV was not significant. Subgroup analysis revealed that neither phase of the menstrual cycle or use of hormonal medication had a significant impact on the expression levels in the microRNAs used in our algorithm.
This is the first report showing that microRNA biomarkers can reliably differentiate between endometriosis and other gynecological pathologies with an area under the curve >0.9 across 2 independent studies. We validated the performance of an algorithm based on previously identified microRNA biomarkers, demonstrating their potential to detect endometriosis in a clinical setting, allowing earlier identification and treatment. The ability to diagnose endometriosis noninvasively could reduce the time to diagnosis, surgical risk, years of discomfort, disease progression, associated comorbidities, and health care costs.
The current gold standard for the diagnosis of endometriosis is laparoscopy, which incurs costs, patient risk, and often delays in diagnosis. An accurate, noninvasive test could mitigate these issues.
Six microRNAs (miRNAs) were evaluated to predict endometriosis in patients undergoing laparoscopy (41 cases, 59 controls).
Serum levels of the miRNAs were measured by quantitative real-time polymerase chain reaction.
A random forest algorithm incorporating the miRNA levels demonstrated an area under the curve of 0.939 in an independent set of patients.
What does this add to what is known?
This study demonstrates the independent validation of a set of miRNAs that can noninvasively differentiate endometriosis from other gynecologic pathologies in a clinical setting, thus avoiding laparoscopy and reducing barriers to diagnosis.
Endometriosis, an inflammatory disorder of endometrial cell proliferation outside the uterus, affects nearly 10% of reproductive-age women, causing pain and infertility. Despite its prevalence, the average time from symptom onset to a correct diagnosis is 5–10 years.
Thus, further delay is incurred in the effort to avoid an invasive procedure, resulting in further disease progression and prolonged suffering.
The social, psychological, and economic impacts of endometriosis are manifold. Studies have described the negative effects of endometriosis on quality of life, extending to work, education, and social and intimate relationships as well as mental and emotional well-being.
Identifying and treating the disease sooner would potentially prevent complications of advanced disease such as infertility while decreasing the economic burden of untreated endometriosis. Laparoscopy is rarely undertaken early in the disease because of the associated risk to the patient and reluctance to undergo surgery without severe symptoms.
Other methods for detecting disease such as imaging and protein biomarkers have proven ineffective.
Serum microRNAs (miRNAs) are emerging as potential molecular indicators to noninvasively identify endometriosis. MiRNAs are short, noncoding RNAs that may be released into the circulation and are protected from endogenous RNase degradation because of inclusion within exosomes or association with specific protein complexes.
Using a logistic regression model and receiver-operating characteristic (ROC) analysis, an excellent area under the curve (AUC) was obtained for the combination of 3 miRNAs (miR-125b-5p, miR-451a, and miR-3613-5p).
Because this previous study evaluated miRNA expression profiles only in patients with moderate/severe (stage III/IV) endometriosis, we sought to expand the clinical utility by testing a more diverse set of patients. This study design enabled us to test whether preoperative evaluation of these miRNA biomarkers could distinguish endometriosis from other benign gynecological conditions in a diverse patient population.
Materials and Methods
Institutional review board approval was obtained from the Yale University School of Medicine (New Haven, CT). Written informed consent was obtained from patients undergoing surgery for suspected benign indications between September 2016 and October 2017. Women aged 18-49 years were included. Exclusion criteria consisted of postmenopausal state, pregnancy, critical anemia, hyperplasia, polyps, or malignancy.
Serum samples were collected from women prior to undergoing laparoscopy for suspected benign gynecological conditions, and miRNA expression analysis was performed blinded to the surgical findings. Subjects were stratified into the disease group if visual or pathology findings from surgery confirmed the presence of endometriosis or the control group if surgery revealed other benign pathology. Categorization was validated among 3 clinicians independently. Staging was done using the revised American Society of Reproductive Medicine (rASRM) classification
and independently confirmed by evaluating the surgical reports and pathology results. A separate independent data set used to validate our findings was obtained from our previously published cohort of 24 patients with surgically confirmed endometriosis and 24 controls.
Prior to surgery, blood (5-10 ml) was drawn from the subjects and collected in sterile tubes (BD, Franklin Lakes, NJ, USA). Serum was collected immediately by centrifuging at 2500 rpm for 15 min at 4° C and stored at -80°C.
Total miRNA was extracted using the miRNeasy mini kit from Qiagen (Valencia, CA) and reverse transcribed using Invitrogen NCode miRNA first-strand cDNA synthesis MIRC-50 kit (Life Technologies, Guilford, CT) according to the manufacturer's instructions. Primers (Supplemental Table) for miRNAs and human U6 small nuclear RNA were obtained from the W. M. Keck Oligonucleotide Synthesis Facility (Yale University, New Haven, CT), and universal reverse primer was obtained from Applied Biosystems (Foster City, CA). MicroRNA levels were quantified by qRT-PCR using SYBR Green and reaction conditions were followed as described.
MicroRNA expression was normalized to U6 and experiments were carried out twice independently, each in duplicate. The miRNAs evaluated in this study were selected from a large screen that identified many miRNAs altered in endometriosis. For this study we specifically used only those that showed minimal or no alteration through the menstrual cycle or in response to sex steroid hormone treatment.
Based on a power of 0.8, an alpha of 5%, an incidence of 50% and the effect size observed in our previous study,
a minimum of 52 subjects were needed to power the study. Because we were initially blinded to the diagnosis, we collected 100 samples to assure an adequate number of subjects with endometriosis.
A Student t test was used to compare the clinical characteristics of subjects in the endometriosis and control groups. Mean expression levels of serum miRNAs between the groups were compared using the Mann-Whitney U test. The Bonferroni correction was performed to adjust for multiple comparisons because 6 miRNAs were analyzed.
To evaluate the diagnostic utility of each miRNA biomarker, ROC analysis was performed, and the ROC AUC was calculated. To evaluate the performance of the prior 3-marker classifier formula developed from our previous study data set
in the current dataset (n = 100), we normalized and rescaled the 2 data sets to account for differences in qRT-PCR methodology. Each miRNA of the 2 data sets was standardized using the z-score method by setting the within data set mean to 0, SD of 1, with the following formula: miRNA standard = (miRNA – mean [miRNA])/SD (miRNA).
Imputation was performed for missing values (representing <3% of the values in the current data set) by applying the multivariate imputation by chained equations method. The prior classifier used the combination of 3 miRNAs (miR-125b, miR-451a, and miR-3613) and was derived using a logistic regression model.
After adjusting coefficients for each variable according to the rescaled data, the performance of the 3-marker classifier was evaluated by comparing the predicted outcome values with the true outcome variables and calculating the AUC.
To build and test an optimal classifier using the current data set (n = 100), we compared 2 statistical approaches: penalized regression model and machine learning with random forest. The latter was chosen because of the ability to yield an AUC of 1 in the training data set. Given that all biomarkers assessed were strongly associated with disease status, using random forest to obtain importance measures was more beneficial to rank all the variables rather than obtaining nonzero coefficients from the penalized regression model. This data set was split into training and testing sets to train and assess the performance of the classifier.
To balance upward and downward bias of prediction accuracies and to simulate a larger sample size, we applied the 0.632 stratified bootstrapping method to generate 1000 replicates. For each replicate, random forest was applied to the training set to build decision trees (n = 500) and predict the disease status in the testing set. In addition to validation of the algorithm with random subsampling, we evaluated the diagnostic performance of the classifier algorithm using an independent data set by testing the optimal random forest model in the rescaled data set previously collected (n = 48).
From the original 103 subjects, 3 were excluded because of an unexpected comorbidity including malignancy. Of the remaining 100 patients, 41 were categorized as endometriosis and 59 as controls. The demographics and clinical characteristics of the subjects are summarized in Table 1.
Table 1Patient demographics and clinical characteristics
Endometriosis (n = 41)
Control (n = 59)
34.1 ± 7.1
36.9 ± 8.2
Body mass index
28.1 ± 7.5
30.4 ± 7.5
Race, n %
rASRM endometriosis stage
Phase of menstrual cycle
Unable to determine
GnRH, gonadotropin-releasing hormone; OCP, oral contraceptive pill; rASRM, revised American Society of Reproductive Medicine.
Moustafa et al. Endometriosis diagnosis by microRNA biomarkers. Am J Obstet Gynecol 2020.
There was no statistically significant difference between age and body mass index in the study groups. Of those with endometriosis, approximately 90% of the surgeries were performed for pelvic pain and 10% for infertility. The endometriosis group consisted of varying degrees of disease as categorized by rASRM stage and divided into stage I, II, III, or IV, while the control subjects had varying benign pathologies as shown in Table 1.
The menstrual cycle phase and presence of hormonal medications for all subjects are also recorded in Table 1. In nearly half of the study subjects, the phase of the menstrual cycle could not be accurately determined based on either use of hormonal medication or a history of irregular cycles.
In this case-control study, the presence of endometriosis was not known prior to surgery. We initially included miRNAs for evaluation that showed the largest difference between disease and control populations with little to no overlap in our prior study as well as those that were menstrual cycle and sex steroid independent. Alterations of the algorithm were made with inclusion and exclusion of the previously evaluated cohort of miRNAs that demonstrated correlation with disease,
and miRNAs that did not significantly contribute to the model were ultimately excluded from the final algorithm.
Figure 1 shows the expression levels of 6 miRNAs that were prospectively measured. Among these 6 miRNAs, miR-125b, miR-150-5p, miR-342-3p, and miR-451a were significantly increased in patients with endometriosis, while miR-3613-5p and let-7b were significantly decreased. Subgroup analysis based on cycle phase for both control or endometriosis subjects showed no significant difference in miRNA expression levels between those sampled during the proliferative vs secretory phase (Figure 2).
Hormonal therapies (Table 1) included predominantly combined oral contraceptives (10, 24%) and gonadotropin-releasing hormone agonists (6, 14%). The presence of hormonal treatment did not significantly affect the average expression levels of the 6 target miRNAs tested (Supplemental Figure).
To evaluate whether the expression of these miRNAs correlates with the stage of endometriosis, we separated minimal/mild (stage I/II) endometriosis from moderate/severe (stage III/IV). Using a Kruskal-Wallis test, all 6 miRNAs were found to have significantly different variances (P < .05) between the 3 groups: control, sage I/II, and stage III/IV (Figure 3). However, after using Dunn’s multiple comparisons test for the 3 pairwise comparisons, each subgroup of endometriosis had significantly different miRNA levels compared with the control group but not between minimal/mild vs moderate/severe.
ROC analysis of individual miRNAs showed AUC scores ranging between a low of 0.68 for miR-150-5p to 0.92 for miR-342-3p (Table 2). Using machine learning with random forest approach, a new classifier algorithm was developed using the 6 miRNAs. This algorithm was validated in 2 ways: by random subsampling dividing the total data set into training and testing subsets and by testing against our previous data set, which was not used for model development (n = 48, 24 endometriosis and 24 control subjects).
This is the first study performed within a diverse population that demonstrates the ability of circulating miRNAs to reliably differentiate endometriosis from other gynecologic pathologies, with robust diagnostic performance in an independent test data set. The clinical characteristics of the current study population was reflective of real-world patients with endometriosis, including diverse racial demographics, early- and late-stage disease, varying phases of the menstrual cycle, and the presence of hormonal treatments. Evaluation of these markers among a cohort of patients with varied pelvic pathologies supports the utility of using these markers in a general population to distinguish endometriosis from other conditions.
The ROC analysis again demonstrated the significant diagnostic value of combinations of these miRNAs. In contrast to our prior work,
which excluded women undergoing hormonal treatments and included only women with moderate/severe (stage III/IV) endometriosis, the current study included cases of minimal/mild (stage I/II) disease and women using hormonal therapies. Here we sought to optimize our combination of miRNA biomarkers to reflect this more diverse and representative patient population.
The single serum miRNA biomarker with the most reproducible diagnostic performance was miR-451a. This miRNA had an AUC of 0.84 in the current study, nearly identical to the AUC of 0.835 in our prior study
an AUC of 0.8 was obtained in the current study. This performance is expectedly lower than was previously achieved with this combination because prior studies compared near-pristine controls with an advanced-disease group.
For every combination of miRNAs, we applied numerous nonmachine learning algorithm approaches and machine learning approaches and analyzed the performance results for each. In an independent data set, random forest applied to 6 of the miRNAs yielded an optimal classifier with an AUC of 0.939. Because endometriosis is not a fatal condition, optimizing the classifier for specificity (avoiding false positives) would help prevent overdiagnosis, and women could be retested if symptoms persist.
Using this strategy to define our thresholds, the current model yielded 96% specificity and 83% sensitivity. Alternatively, optimizing both values simultaneously, we achieved a sensitivity and specificity of approximately 90% each. A test with higher sensitivity and a low false-negative rate could be appropriate for the use of the biomarker panel as a screening test.
We previously selected these miRNAs based on lack of significant change through the menstrual cycle, and here we again observed no difference in the expression levels of these 6 miRNAs based on the phase of menstrual cycle (secretory vs proliferative).
Furthermore, we observed that hormonal medications alone do not significantly alter the expression levels of these miRNAs. The potential of these miRNAs to monitor response to therapy was demonstrated in our recent nonhuman primate study, in which miR-150-5p, miR-451a, and miR-3613-5p showed expression changes that correlated with the reduction of lesion volume after endometriosis treatment.
Here our study model did not allow for miRNA level evaluation at different durations of therapy; future longitudinal studies will measure how the biomarkers change over time with effective medical or surgical therapy.
this study included patients with all stages of disease. Analysis of the expression levels of each miRNA based on rASRM staging revealed that all miRNAs could distinguish stage I/II from control, and stage III/IV from control, but that the difference between stage I/II vs stage III/IV was not significant. This may reflect limitations of the rASRM staging system. The ASRM staging does not exclusively record active disease; it also reflects scarring, adhesions, and reactive damage that often lead to stage IV characterization.
MiRNA changes may be a more accurate way to measure active disease. Alternatively, because miRNAs are produced by multiple organ systems, the levels seen may represent an effect on organs outside the pelvis. The mechanisms by which alterations in miRNAs occur are poorly understood. Some miRNAs are likely made directly by the lesions, while others are altered because of the effects of endometriosis on other tissues. Endometriosis is a systemic inflammatory disease that may broadly affect miRNA production independent of stage.
Finally, the inability to detect significant differences between endometriosis stages may also be due to distinct molecular profiles of different subtypes of endometriosis that are also independent of stage; this will be investigated in future studies. Despite the inability to distinguish stage of disease, in our test the ability to capture early-stage disease may have significant clinical advantages.
Prior attempts to establish markers for this disease have been focused on nonspecific inflammatory markers, while the use of circulating miRNAs provides a disease-specific signature, unique to endometriosis. A highly sensitive and specific test will have great clinical significance in women with pelvic pain or unexplained infertility. The ability to diagnose endometriosis noninvasively could reduce the time to diagnosis, surgical risk, years of discomfort, hospitalizations and health care spending, and, ultimately, disease progression and associated comorbidities. Further studies are warranted to assess how these markers are altered by endometriosis treatment or whether unique marker profiles can provide insight into fertility or patient pain scores. Nonetheless, the combination of 6 miRNAs validated in this study yielded high AUC scores, supporting the excellent diagnostic potential of these biomarkers for endometriosis.
Our study model did not allow for the evaluation of levels over time or in response to therapy; however, ongoing longitudinal studies will measure how the biomarkers are affected by treatment. While our study is not powered to detect unique signatures for different disease phenotypes, these findings support the need for further studies to investigate this issue. Further larger prospective studies are required to evaluate the relationships between miRNA levels in serum and endometriosis and their impact on the accuracy of diagnosis, management, and outcome of the different stages of endometriosis.
Strengths and limitation of the study
Our study has many strengths compared with other contemporary studies. The study group was evaluated prospectively allowing direct correlations between levels of miRNAs and the presence of endometriosis. In addition, our study includes patients with other pelvic pathologies in the control group, accurately distinguishing between endometriosis and other potential sources for pelvic inflammation. This expands the generalizability of this test and supports its use and validity in patients with concurrent pathology. Furthermore, these results reveal that the levels of these 6 miRNAs is unaltered by hormonal therapy or cycle phase, further broadening its potential use.
The weakness of the study rests on the limited sample size of both groups, which does not allow for more detailed correlations between miRNA levels and endometriosis subtypes. This also limits the ability to assess whether certain features of endometriosis such as severity, stage, presence of endometrioma, deep infiltrating, or rectovaginal disease (which was not a specified end point of this study) drive particular shifts in miRNA expression. While we note in previous text that this model detects both mild and advanced disease, the study is not powered to detect differences between subgroups. Another possible limitation to our study is the use of surgical cases across a large number of physicians, without standardization beyond ASRM staging for documentation of endometriotic burden.
This is the first report showing that miRNA biomarkers can reliably differentiate between endometriosis and other gynecological pathologies with an AUC >0.9 across 2 independent studies. Prior attempts to establish markers for this disease have been focused on nonspecific inflammatory markers, while the use of circulating miRNAs provides an endometriosis-specific signature. A highly sensitive and specific test will have great clinical significance in women with pelvic pain or unexplained infertility. The ability to diagnose endometriosis noninvasively could reduce the surgical risk, years of discomfort, hospitalizations and health care spending, and ultimately, disease progression and associated comorbidities.
We thank Sangsoon Woo, PhD, for assistance with biostatistical analysis; Juliana Ansari, PhD, for assistance in manuscript preparation; Heather Bowerman, James Wingrove, PhD, and Marie Gaye for proofreading, and Luisa Coraluzzi, RN, for clinical research support and patient sample collection.
Supplenental TablePrimer sequences
Moustafa et al. Endometriosis diagnosis by microRNA biomarkers. Am J Obstet Gynecol 2020.