Multicenter evaluation of blood‐based biomarkers for the detection of endometriosis and adenomyosis: A prospective non‐interventional study

To evaluate blood‐based biomarkers to detect endometriosis and/or adenomyosis across nine European centers (June 2014–April 2018).


| INTRODUC TI ON
The long interval between the onset of endometriosis and/or adenomyosis symptoms and receiving a diagnosis (4-12 years) 1-7 may result in disease progression, prolonged pain, and impaired fertility. 7,8dometriosis severity can be classified using the revised American Society of Reproductive Medicine (rASRM) system, into four stages. 91][12] Such surgery carries inherent risks and extended patient recovery times; 13 moreover, detecting endometriosis during laparoscopy is often dependent on the surgeon's experience and lesion accessibility. 8,14Adenomyosis is challenging to diagnose as it is based on the patient's symptoms combined with transvaginal ultrasound (TVUS) and/or magnetic resonance imaging (MRI); 15 accurate diagnosis requires highly experienced staff and appropriate equipment, 16 meaning that formal diagnosis often takes place via histologic evaluation of uterine tissue following hysterectomy. 17,18ood-based biomarker testing may facilitate earlier endometriosis/adenomyosis diagnosis compared with current diagnostic techniques.Earlier detection may allow timely implementation of clinical management strategies and fertility procedures.It is unlikely that this testing would replace the need for diagnostic laparoscopy, but it could be implemented to help guide clinical decision making around further diagnostic procedures.
The primary objective was to evaluate suitable blood-based biomarkers, or biomarker combinations, for detection of suspected endometriosis and/or adenomyosis with high sensitivity.S1).All participants provided written informed consent.dometriosis" group, including participants with endometriosis with or without adenomyosis; and the "adenomyosis only" group.The "all endometriosis" group was further stratified by rASRM stage.The "all symptomatic controls" group was stratified into two sub-groups: "pathologic symptomatic controls", including participants with pelvic pain and/or unexplained subfertility with pathologic findings of a condition other than endometriosis/adenomyosis (e.g.fibroids); and "pathology-free symptomatic controls", which included participants with pelvic pain and/or unexplained subfertility, but no pathologic findings on laparoscopic investigation.

| Measurement of blood-based biomarkers
Samples were combined into one large panel and subsequently randomized before measurement.For each biomarker, measurements for all samples were performed in one batch.A panel of biomarkers was examined, including 17 autoantibodies identified from previous internal discovery analysis, alongside seven selected clinical variables (Table S2).Protein biomarkers were chosen due to their association with endometriosis, adenomyosis, and/or pelvic inflammation, consistent with existing literature, or disease pathway analyses with a focus on the involvement of the neurogenic inflammation pathway. 22Statistical analyses are described in the Methods S1.

| Statistical analysis
The highest-ranking single biomarkers for discriminating between cases and controls for each comparison were identified by plotting receiver operating characteristic curves for all biomarkers analyzed and calculating area under the curve (AUC) values.Biomarkers were assessed within menstrual cycle phase subgroups in a comparison between participants with confirmed hormone intake versus no hormone intake, described in the Methods S1.Wilcoxon P values were calculated for biomarkers to evaluate the statistical significance between cases and controls; P values were corrected for multiple testing by applying a false discovery rate correction (referred to as q values throughout). 23,24To find the top-performing two-and three-biomarker combinations (including clinical variables) for distinguishing between cases and controls, all combinations were systematically evaluated, and their performances were assessed in extensive double (two-tier) cross-validation employing logistic regression (Figure S1).The sensitivity of each top-performing biomarker or biomarker combination (set at 60% specificity, based on unpublished findings of an advisory board hosted by Roche Diagnostics International Ltd in October 2012) was also recorded.The statistical significance of the endometriosis stage dependency of each top-performing biomarker was assessed by linear regression, from which the P value of the t test (i.e.testing whether the stage regression coefficient is different from 0) was reported using an α of 0.05 for the cut-off for significance.For all analyses, R version 3.6.3was used; multivariate analyses employed the mlr R-package. 25 rare cases, some biomarker measurements were not available due to measurement errors or insufficient sample volumes; see Tables S8-S12 for available biomarker measurements for each comparison.In such cases, participants with missing values were removed; no imputation was performed.The multivariate workflow was performed on the complete data matrix (i.e. the subset of participants with all biomarker values present).

| Participant demographics and clinical characteristics
In total, 1051 women were recruited; 966 women were eligible to participate; data from 919 participants were analyzed; and 47 participants were excluded (Figure 1).Tables S3-S6 provide an overview of participant demographics and clinical characteristics by study center.
Of the 919 participants analyzed, 741 had confirmed endometriosis and/or adenomyosis ("cases") and 178 were symptomatic controls ("all symptomatic controls"; Figure 1).Of the participants with confirmed endometriosis and/or adenomyosis, 617 participants had endometriosis with or without adenomyosis ("all endometriosis cases") and 124 participants had adenomyosis only ("adenomyosis only cases").Table S7 provides an overview of the surgical findings from participants with endometriosis and/or adenomyosis.
Of the symptomatic controls ("all symptomatic controls"), 131 of the participants were symptomatic with pathologic findings other than endometriosis/adenomyosis ("pathologic symptomatic controls") and 47 were symptomatic controls without pathologic findings ("pathology-free symptomatic controls").
Results from the hormone intake and menstrual cycle phase subgroup analyses are detailed in the Methods S1.

TA B L E 1
Top-performing biomarkers, autoantibodies, and clinical variables identified.Note: Not all biomarkers could be measured in all samples as there were measurement errors or insufficient sample volume remaining.
b FDR-corrected Wilcoxon P value; a q-value <0.05 was considered statistically significant.
The top-performing bivariate biomarker combination for this comparison was CA-125 and S100-A12; there was an AUC improvement of ~2% (AUC 0.707, 95% CI 0.631-0.784;Figure S2d) and a sensitivity improvement of ~1% (sensitivity was 73.0%), when compared with S100-A12 or CA-125 alone.No trivariate biomarker combination had a higher AUC than this bivariate combination for this comparison.

| Biomarker analysis across endometriosis stage
The CA-125 concentration differences between "all endometriosis cases" and the "all symptomatic controls" and "pathology-free symptomatic controls" significantly increased (P < 0.001) with endometriosis stage (Figure S4).For "all endometriosis cases" versus "all symptomatic controls", CA-125 had an AUC of 0.583 (95% CI 0.533-0.633)for detecting endometriosis Stages I/II and an AUC of 0.795 (95% CI 0.748-0.843)for Stages III/IV.For "all endometriosis cases" versus "pathology-free symptomatic controls", CA-125 had an AUC of 0.587 (95% CI 0.497-0.677)for detecting endometriosis Stages I/II and an AUC of 0.798 (95% CI 0.732-0.864)for Stages III/IV.For S100-A12 and sFRP-4, there was no significant trend in stage dependency (Figures S5 and S6).For "all endometriosis cases" versus "pathology-free symptomatic controls", CA-125 was a stronger predictor of endometriosis than S100-A12 in the Stages III/IV group, but weaker in the Stages I/II group (Figure S4).

| DISCUSS ION
This study examined the diagnostic accuracy of selected bloodbased biomarkers, alongside clinical variables, in samples from women with pelvic pain and/or unexplained subfertility to detect suspected endometriosis and/or adenomyosis.CA-125, S100-A12, and sFRP-4 were the top-performing statistically significant biomarkers across comparisons.The combination of CA-125 and S100-A12 performed better than either marker alone in distinguishing between "all endometriosis cases" and "pathology-free symptomatic controls".No other multivariate combination of biomarkers, or combination of biomarkers and clinical variables, outperformed the best univariate marker in all other comparisons.CA-125 showed a gradual and significant (P < 0.001) increase by disease stage in "all endometriosis cases" versus "all symptomatic controls" and "pathology-free symptomatic controls".
Currently, this is the most robust multicenter, prospective study investigating biomarkers to identify endometriosis/adenomyosis.This contemporary analysis examined a well-characterized participant population attending routine clinical appointments and used standardized data and sample collection protocols.While a large panel of blood-based biomarkers was studied, it was not possible to conduct measurements for all autoantibodies potentially associated with endometriosis that have been identified in the literature. 26,27This study supports the idea that endometriosis/ adenomyosis are highly heterogeneous conditions encompassing a wide range of disease subgroups, making it difficult to identify a single biomarker to diagnose/rule out endometriosis and/or adenomyosis.
9][30][31] As such, the use of CA-125 as a diagnostic or screening tool is not recommended by National Institute for Health and Care Excellence or European Society of Human Reproduction and Embryology guidelines. 10,11,32Nevertheless, several studies reported that CA-125 serum concentration increases with endometriosis stage and is particularly high in women with dense pelvic adhesions or ovarian endometriomas. 28,30,33,34These studies support our findings that CA-125 showed moderate performance at detecting all stages of endometriosis, but performed better for Stage III/IV.Notably, the performance of CA-125 was not sufficient for diagnostic or screening purposes. 35 our knowledge, sFRP-4 and S100-A12 have not previously been investigated as blood-based biomarkers for the diagnosis of endometriosis/adenomyosis. In a small study, endometriotic cells collected from participants with endometriosis had significantly increased S100-A12 expression versus endometrial stromal cells from healthy participants. 36Our results support these findings, S100-A12 showed a moderate performance at diagnosing adenomyosis/endometriosis in populations of women with symptoms but no pathologic findings.This is of biologic relevance because it suggests that altered S100-A12 serum levels may indicate pelvic inflammation associated with endometriosis/adenomyosis; however, the AUC values are insufficient for diagnostic/screening purposes.
In our study, sFRP-4 was the strongest predictor of adenomyosis in women with symptoms versus the other biomarkers.Our results support the findings of Delaney et al., 37 who reported that, when compared with healthy myometrium, sFRP-4 was overexpressed in uterine fibroids, which are benign myometrial lesions that commonly co-exist with adenomyosis. 38Although these findings are of biologic relevance, the AUC values are insufficient for diagnostic or screening purposes.
The development of simple, non-invasive tools to improve diagnosis of endometriosis is currently a topic commanding a great deal of research interest.In addition to blood-based biomarkers, recent publications have reported investigations into the use of salivary microRNA signatures 42 and urinary biomarker measurements. 43wever, large-scale, confirmatory studies will be needed to better establish and validate the accuracy and utility of such tests before they can be considered for clinical use.
In conclusion, despite the high standard of data and sample collection and the large number of participants, a single biomarker or F I G U R E 2 Protein biomarker concentration differences and ROC curves of (a) CA-125, (b) S100-A12, and (c) sFRP-4 for "all endometriosis cases" versus "all symptomatic controls"."Other endometriosis" indicates when the surgeon did not/could not stage the endometriosis during laparoscopy, or the participant had deep endometriosis with or without adenomyosis.Measurement errors or insufficient sample volume meant that not all biomarkers could be measured in all samples.In the box plots: the black line shows the median value, the yellow cross shows the mean value, the box shows the interquartile range, and the whiskers show the range.AUC, area under the curve; CA-125, cancer antigen 125; CI, confidence interval; ROC, receiver operating characteristic; S100-A12, S100 calcium-binding protein A12; sFRP-4, secreted frizzled-related protein 4.
This multicenter, prospective, non-interventional study (June 2014-April 2018) was conducted at nine European centers.The study complied with the International Conference on Harmonization guidelines for Good Clinical Practice and the principles of the Declaration of Helsinki.Ethical approval was obtained for each study center (Table Eligible participants were recruited during routine clinical visits.Inclusion criteria were: women aged 18-45 years experiencing relevant symptoms (i.e.chronic pelvic pain and/or unexplained subfertility) with clinical suspicion of endometriosis and/or adenomyosis undergoing laparoscopy or laparotomy.Further enrolment criteria are described in the Methods S1.Surgical findings were recorded in accordance with the World Endometriosis Research Foundation Endometriosis Phenome and Biobanking Harmonization Project (EPHect) protocols, using standardized data and sample collection techniques. 19-21Participant disposition is shown in Figure 1.Participants were defined as "cases" (endometriosis, endometriosis and adenomyosis, or adenomyosis only) or as "all symptomatic controls" (no endometriosis or adenomyosis).Endometriosis diagnosis required histologic confirmation; adenomyosis diagnosis was made by either TVUS or visualization during surgery.Participants with endometriosis/adenomyosis ("cases") were stratified into two subgroups: the "all en-

F
I G U R E 1 Participant disposition.a With or without adenomyosis.

Variable Ranking a N Mean difference/fold change (log 2 /linear) Significance (Wilcoxon q value b ) AUC Sensitivity at 60% specificity
, % "All endometriosis cases" versus "all symptomatic controls"