Diagnostic Accuracy of Liquid Biomarkers for the Non-Invasive Diagnosis of Endometrial Cancer: A Systematic Review and Meta-Analysis

Simple Summary Endometrial cancer is common amongst women and rates are increasing annually. The diagnosis of this condition for women with bleeding after the menopause is invasive and often painful with many more women undergoing investigation than needed. A simple, non-invasive blood or urine test for the diagnosis of endometrial cancer is being sought. This review summarizes the current research on blood and urine tests and their diagnostic accuracy for detecting endometrial cancer. Whilst many blood and urine tests have been assessed there is currently no test that has a similar accuracy to biopsy of the uterine lining. However, this review demonstrates that there are some potential candidates which need to be explored by larger studies and on bigger groups of women. Abstract Endometrial cancer rates are increasing annually due to an aging population and rising rates of obesity. Currently there is no widely available, accurate, non-invasive test that can be used to triage women for diagnostic biopsy whilst safely reassuring healthy women without the need for invasive assessment. The aim of this systematic review and meta-analysis is to evaluate studies assessing blood and urine-based biomarkers as a replacement test for endometrial biopsy or as a triage test in symptomatic women. For each primary study, the diagnostic accuracy of different biomarkers was assessed by sensitivity, specificity, likelihood ratio and area under ROC curve. Forest plots of summary statistics were constructed for biomarkers which were assessed by multiple studies using data from a random-effect models. All but one study was of blood-based biomarkers. In total, 15 studies reported 29 different exosomal biomarkers; 34 studies reported 47 different proteomic biomarkers. Summary statistic meta-analysis was reported for micro-RNAs, cancer antigens, hormones, and other proteomic markers. Metabolites and circulating tumor materials were also summarized. For the majority of biomarkers, no meta-analysis was possible. There was a low number of small, heterogeneous studies for the majority of evaluated index tests. This may undermine the reliability of summary estimates from the meta-analyses. At present there is no liquid biopsy that is ready to be used as a replacement test for endometrial biopsy. However, to the best of our knowledge this is the first study to report and meta-analyze the diagnostic accuracy of different classes of blood and urine biomarkers for detection of endometrial cancer. This review may thus provide a reference guide for those wishing to explore candidate biomarkers for further research.


Introduction
Endometrial cancer (EC) is the most common form of uterine cancer and arises from the lining of the uterus, known as the endometrium. In 2020, GLOBOCAN the World Health Organizations' International Agency for Research on Cancer recorded 417,367 new cases of uterine cancer and 97,370 deaths making this the fourth most common cancer of women and the third leading cause of cancer mortality in females after breast and lung cancer [1]. Incidence in the UK has increased by around 55% since the 1990s with 9700 women diagnosed per year in 2016-2018 [2]. The rise in cases of EC diagnosed annually is set to increase globally by 30% by the year 2040 [3]. This is likely due to the advancing age of the population and rising rates of obesity. Currently if diagnosed and treated at International Federation of Gynecology and Obstetrics [FIGO] Stage I or II, EC 5-year survival rates are around~92% and 75%, respectively, whereas advanced stage III and IV ECs have 5-year survival rates of 48% and 15%, respectively [4][5][6]. Strategies for early diagnosis are therefore critical.
Symptoms suspicious for EC are post-menopausal bleeding, unscheduled bleeding on hormone replacement therapy (HRT), persistent intermenstrual or irregular bleeding, hematuria or abnormal vaginal discharge. Although EC is less commonly diagnosed in pre-menopausal women, in the UK, 6.5% of women diagnosed with EC between 2015-2017 were less than 50 years old [7]. In post-menopausal women with bleeding, transvaginal ultrasound scan (TVS) is performed as a triaging test to assess endometrial thickness (ET). An ET < 4 mm is associated with an <1% risk of EC [8,9]. For those with an ET > 4 mm, an endometrial biopsy is recommended as the gold standard diagnosis of EC [8]. However, in symptomatic perimenopausal or pre-menopausal women with risk factors for EC, the National Institute of Clinical Excellence recommend hysteroscopy and targeted biopsy in the first instance as TVS has limited value in women who are still menstruating [10].
Whilst the gold standard of EC diagnosis, endometrial biopsy has an almost 99% accuracy, the current technique for triaging with ultrasound lacks specificity resulting in more than 50% of patients needing invasive biopsy [11]. Furthermore, failure to obtain a biopsy in the outpatient setting is common and occurs in around one third of women, often due to sampling failure or pain during the investigation [12]. In these patients, repeat investigations are needed, often under general anesthetic which is not without associated risks and costs [13]. A simple, easy to administer, non-invasive test that could triage women with EC for diagnostic biopsy whilst safely reassuring healthy women, would vastly improve patient care over the current model. The ideal detection tool would be simple to perform, non-invasive to obtain, accurate in reassuring women without disease and cost effective to allow implementation as a screening program within the primary care setting.
'Liquid biopsy' has the potential to offer this solution and refers to the sampling and analysis of non-solid biological tissue for its tumoral elements [14]. These elements might be circulating tumor cells (CTCs), cell-free tumor DNA (ctDNA), as well as extracellular vesicles (EVs), microRNAs (miRNAs), mRNA, long noncoding RNAs (lncRNAs), small RNA, circulating cell-free proteins, and tumor-educated platelets (TEPs) [14]. Many body fluids can be used for liquid biopsy, however the most non-invasive and widely investigated fluids are blood and urine. Multiple approaches have been employed in the search for diagnostic cancer biomarkers. Advances in areas such as proteomics, metabolomics and genomic sequencing have increased the scope for individual or panels of biomarkers to be discovered. However, biomarkers must overcome several hurdles before they are implemented into clinical practice; discovery, validation, and verification [14]. An ideal biomarker should be accurate and reproducible between laboratories. Its clinical validity should be reported in terms of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV). However, between given studies, it is difficult to compare the performance of one or multiple biomarkers unless they share a fixed false positive rate. For this reason, the area under the curve (AUC) value of a receiver operator characteristic (ROC) curve plotting Sensitivity over the False Positive Rate (1-Specificity) gives an objective comparison between test performance with one value and is therefore the best tool of comparison between studies analyzing the same marker [15,16].
The aim of this systematic review is to evaluate studies assessing blood and urinebased biomarkers as a replacement test for endometrial biopsy or as a triage test to inform the decision to perform endometrial biopsy. Specific objectives are to provide summary estimates of the diagnostic accuracy of blood or urinary biomarkers for the diagnosis of EC compared to endometrial biopsy and to assess the diagnostic utility of biomarkers that could differentiate between benign and malignant endometrium.

Materials and Methods
The study was carried out according to the Cochrane Collaboration recommendations as well as the Preferred Reporting Items for Systematic Review and Meta-analysis of Diagnostic Accuracy Studies (PRISMA-DTA) [17]. This study was registered with PROSPERO, registration number: CRD42020202191 [18].

Literature Search
Systematic literature searches were carried out in the NICE Healthcare Databases Search tool using CINAHL, EMBASE, Medline and Pubmed. The database was searched from 2000 to January 2022. The search was restricted to English language papers, full text articles and non-review articles. The search strategy included the following key words: (endometrial cancer [Title/Abstract]) AND (biomarker [Title/Abstract]. Cross-referencing of key texts and grey literature (Google Scholar) searches were also carried out.
Titles and abstracts were screened for the eligibility of the study and evaluated by two different operators (RK, SW). Relevant articles were retrieved in full-text and assessed against the inclusion criteria. Inclusion criteria were as follows; firstly, studies evaluating the diagnostic accuracy of either blood or urine biomarkers for EC detection; second, those studies diagnosing EC using endometrial histopathological assessment as the reference test; third, those studies reporting AUC as well as sensitivity and specificity and fourth those studies comparing a control group cohort without EC. Reviews, letters, conference reports, and duplicated publications were excluded to make sure only primary publications of original studies were included. Studies not reporting diagnostic accuracy with AUC and studies where biomarker performance was reported with only grouped markers were excluded from further analysis. Studies solely reporting HE4 or CA125 where multiple systematic reviews and meta-analyses have already been published were excluded from further review [19][20][21][22].

Data Extraction and Quality Assessment
Data was extracted using a standard form that included methods, basic study population characteristics, inclusion and exclusion criteria and accuracy of diagnostic tests used.
The risk of bias was evaluated by three independent authors (RK, SW, JC) in each study using the Diagnostic Precision Study Quality Assessment Tool (QUADAS-2) as recommended by the Cochrane Collaboration [23]. Any discrepancies between authors were discussed and the opinion of a third reviewer was sought. The assessment was carried out with use of the Review Manager Software (Version 5.4, The Cochrane Collaboration, London, UK, 2020).
Each study was deemed to be at 'low', 'high' or 'unclear' risk for each of four domains. Studies were deemed as having low methodological quality when they were at high or unclear risk of bias or when there was high concern regarding applicability in at least in one domain. The original signaling question, 'Was a case-control design avoided?' was amended to 'Was a two-gate design avoided?' in agreement with the Cochrane study of Nissenblat et al., 2016 [24]. Diagnostic accuracy studies are cross-sectional in nature, comparing an index test with the reference standard in the same group of participants. Study investigators measure the parameters at a single point in time and classify the groups by the outcome of the reference standard test. Therefore, unlike in epidemiological studies, Cancers 2022, 14, 4666 4 of 28 the terminology 'cohort' and 'case-control' is less informative so 'single-gate' and 'two-gate' design was subtitled. This question was included because a two-gate design has more potential to introduce selection bias and overestimate accuracy of a given biomarker [25].

Statistical Analysis
For each primary study, a diagnostic accuracy of different biomarkers was assessed by sensitivity, specificity, likelihood ratios and area under ROC curve with their 95% confidence intervals. It is common that primary studies report sensitivity and specificity for a studyspecific cut off. AUC cannot be generated where source data is not available and thus it was necessary for inclusion criteria for this to be reported. Where the AUC 95% confidence interval (CI) was not reported in the text, study authors were contacted individually for missing data. Where no response was received the decision to calculate the standard error of the mean using non-parametric Wilcoxon's statistics was assessed individually [15,26]. If study numbers of cases and control patients reached less than 60 and no 95% CI for AUC was reported, the reliability of this method was deemed suboptimal and hence the paper was excluded from meta-analysis. A weighted summary AUC was calculated with the assumption of non-homogeneity and non-normality of empirical sensitivity and specificity as reported by Zhou et al. [15]. Forest plots of summary statistics were constructed using the data from the random-effect models. The heterogeneity of the studies was established by using Cochran's heterogeneity statistics Q, which was calculated as the weighted sum of squared differences between the individual study effects and pooled effect across the studies [27]. I 2 , assesses the percentage of variation across individual studies that is due to heterogeneity rather than chance and does not depend on the number of studies included in the meta-analysis [27]. Values (I 2 ) of 0-40% might be insignificant, 30-60% considered as moderate, 50-90% substantial and 75-100% considerable. Publication bias was evaluated by the Egger regression and Beggs' correlation tests with funnel plots [27]. Analyses were carried out with MedCalc software (MedCalc Software 20.112, Ostend, Belgium).

Results
The literature search of MEDLINE, EMBASE and PUBMED yielded 1951 citations. A total of 56 studies were included in the systematic review. Publications were then grouped by biomarker category into publications relating to exosomal biomarkers, proteomic biomarkers, metabolomic biomarkers and circulating tumor materials (CTMs). Proteomic biomarkers were further sub-categorized. Those biomarkers that were reported in studies on two or more occasions were included for meta-analysis. Those biomarkers reported by one study only were included in the systematic review and tabulated in order of test performance statistics. A flow diagram of the literature review is shown in Figure 1 in accordance with PRISMA guidance [28].

Assessment of Quality and Heterogeneity of Studies
The methodological quality of studies was assessed using the QUADAS-2 for exosomal, proteomic, metabolomic and studies discussing circulating tumor material. Data was inputted and tabulated graphically as seen in  The key area of concern for multiple studies was the patient selection domain where a two-gate study design was chosen.

Assessment of Quality and Heterogeneity of Studies
The methodological quality of studies was assessed using the QUADAS-2 for exosomal, proteomic, metabolomic and studies discussing circulating tumor material. Data was inputted and tabulated graphically as seen in  The key area of concern for multiple studies was the patient selection domain where a two-gate study design was chosen.

Meta-Analysis of Exosomal Biomarkers
15 studies reported 29 different exosomal biomarkers and studied a total of 5527 patients of which 2530 had a diagnosis of EC and 2456 were non-cancerous control group patients. All biomarkers were obtained by blood sample. The control groups consisted of healthy women with normal endometrium as well as those with benign endometrial lesions such as polyps and fibroids. Only studies of micro-RNA 21, 27a, and 223 were suitable for meta-analysis after meeting the inclusion criteria [29][30][31][32][33][34][35]. None of the included studies reported positive predictive values (PPV) or negative predictive values (NPV) ( Table 1).

Meta-Analysis of Exosomal Biomarkers
15 studies reported 29 different exosomal biomarkers and studied a total o tients of which 2530 had a diagnosis of EC and 2456 were non-cancerous con patients. All biomarkers were obtained by blood sample. The control groups co healthy women with normal endometrium as well as those with benign endo sions such as polyps and fibroids. Only studies of micro-RNA 21, 27a, and 223 able for meta-analysis after meeting the inclusion criteria [29][30][31][32][33][34][35]. None of the studies reported positive predictive values (PPV) or negative predictive values ( ble 1).

Micro RNA-21
Three studies examined the diagnostic accuracy of MiRNA 21 as a biomarker for the diagnosis of EC ( Figure 5, Table 2). The study by Gao et al., 2016 examined test performance against two control cohorts, one with healthy endometrium and one cohort with benign endometrial changes [29]. Sensitivity and specificity varied between the studies (0.640-0.850 and 0.760-0.920, respectively). The summary weighted AUC was deemed excellent at 0.825 (95%CI 0.735-0.915, p < 0.001). There was considerable heterogeneity between studies as seen by a Q-test score of 13.4 and I 2 of 77.6%.  (Table 3) [32,33]. The study by Wang examined test performance against a control cohort with both normal and benign endometrial changes. The study by Ghazala included a control cohort with normal endometrium only. The summary weighted AUC was outstanding and calculated at 0.925 (95%CI 0.801-1.000, p < 0.001) ( Figure 5). There was considerable heterogeneity between studies as seen by both the Q test and I 2 test (10.3 and 90.3%, respectively).

Protein Based Biomarkers
There were 35 studies reporting 47 different proteomic biomarkers (after the exclusion of CA-125 and HE-4 as previously described) and a total of 3526 patients of which 1483 had a diagnosis of EC and 2043 were non-cancerous control group patients . A total of 34 studies were assessing blood-based biomarkers and one study a urine derived biomarker [64]. The control groups comprised of healthy women with normal endometrium as well as those with benign endometrial lesions such as polyps and fibroids. The summary of the diagnostic accuracy and performance of those biomarkers is outlined in Table 5. For the purpose of meta-analysis, proteomic markers were classed as cancer antigens (CA-15.3, CA-19.9, CA-72.4, CEA), hormones (leptin, visfatin, prolactin) and other proteomic markers (G-CSF, YKL-40 and DJ-1).    (Table 6) [48,55,70]. In the study of Unuvar et al., 2020, the diagnostic accuracy of CA 15-3 was assessed against healthy controls with normal endometrium as well as those with benign endometrial changes [48]. The summary weighted AUC showed good performance reporting 0.608 (95%CI 0.536-0.681 p < 0.001) ( Figure 6). Tests for heterogeneity between studies showed insignificant heterogeneity (Q = 1.271, I 2 = 0.00%; 95%CI: 0.00-69.53).

Metabolomic Biomarkers
There were 4 studies reporting the diagnostic accuracy of 23 different metabolomic biomarkers in a total of 669 patients of which 190 had EC and the remainder were noncancerous control group patients. All biomarkers were obtained by blood sample. The control groups comprised of healthy women with normal endometrium as well as those with benign endometrial lesions such as polyps and fibroids. Two out of four studies reported findings on endometroid adenocarcinoma only. One study did not specify the histological subtype of EC. The summary of the diagnostic accuracy and performance of those biomarkers is outlined in Table 16.   1 Tested by electrospray ionisation tandem mass spectrometry (ESI-MS/MS). 2 Tested by mass spectrometry. 3 Tested by gas-chromatography mass-spectrometry. 4 Tested by the ultra-performance liquid chromatography coupled with triple-quadruple tandem mass spectrometry (UPLC-TQ/MS). AUC: Area Under the Curve; CI: Confidence Interval; PPV: positive predictive value; NPV: negative predictive value.

Circulating Tumor Related Material Biomarkers
There were three studies reporting on circulating tumor related materials. Two studies reported the diagnostic accuracy of circulating cell-free DNA (cCFDNA) and one study reported on Survivin expressing circulating tumor cells (CTC). All biomarkers were obtained by blood sample. No studies were eligible for meta-analysis. The performance of these biomarkers is reported in Table 17.

Discussion
Endometrial cancer is one of the most common malignant tumors in females, and the primary symptom is abnormal vaginal bleeding or discharge. The frequency with which this symptom occurs and the invasive nature of endometrial biopsy, means that at present the triage of women with suspected EC is suboptimal for women and clinicians alike. The optimum biomarker for EC would have a high sensitivity and specificity for detecting EC compared to benign and healthy controls. It would be utilizable for both pre-menopausal and post-menopausal women. Those women at low risk of disease could be reassured without the need for secondary care interventions such as imaging and biopsy. The ideal receiver-operating characteristic area under curve (AUC) would be close to 1, with a minimum of 0.7 to indicate clinical utility as a biomarker. An accurate diagnostic biomarker utilized in primary care could reduce the number of women referred for painful and costly investigations and indeed might also be used for consideration of screening of high-risk groups such as women with Lynch Syndrome or those with multiple risk factors. Hence, the aim of this study was to determine which biomarkers have been assessed for their diagnostic accuracy to date.
At present, the biomarkers most assessed for their diagnostic accuracy of EC are serum CA125 and HE4 which have been reviewed by multiple studies and meta-analyses. The performance of CA 125 is poor in terms of sensitivity and specificity making it unsuitable for use [20,21]. HE4 may have utility as a diagnostic tool, however it has been assessed by several meta-analyses and each of these point to considerable heterogeneity within the data [21,86,87]. Similarities in the data seem to suggest that HE4 may have high specificity but a lower sensitivity than would be needed as a primary care screening tool.
There have been promising results yielded from studies assessing biomarker panels and specifically those using spectroscopic techniques [88,89]. The study by Paraskevaidi et al. assessed the diagnostic accuracy of infrared spectroscopy as a method of detecting EC with an overall diagnostic accuracy of 0.83 [88]. Spectroscopic techniques do not allow for analysis of a single biomarker because peaks may be formed by multiple biological entities. As such, these studies were not eligible for inclusion into this review; however, they are simple techniques yielding promising results.
Obesity is accompanied by changes in expression of adipose factors that act both locally and systemically. With the known link between obesity, insulin resistance and EC, adipokines and hormones such as leptin have become the focus of intense investigation. In EC, leptin activates STAT3 proteins, which increase their activity in the process of oncogenesis by stimulating proliferation, promoting angiogenesis and avoiding the control of the immune system. There is a positive correlation between leptin levels and body mass index (BMI). Adiponectin is predominantly secreted by visceral adipose tissue and is the most abundant adipokine, with circulating concentrations inversely correlated with adiposity. This review included three studies of leptin and one study of adiponectin [50,51,53,61]. The diagnostic accuracy of leptin is moderate with a summary ROC of 0.757. Adiponectin assessed in one study in tandem with Visfatin showed that adiponectin was inversely associated with EC risk [61]. The AUC of this adiponectin alone was 0.814 and the adiponectin to Visfatin ratio 0.838 suggesting some potential as diagnostic biomarkers. However, similarly to other hormones such as FSH, LH, estradiol and prolactin which are also raised markedly in EC patients, these may be markers of the major risk factors, obesity and polycystic ovarian syndrome rather than EC itself, thus showing poorer specificity than required of a screening biomarker. A meta-analysis of the effect of adipokines in obesity driven cancers using mendelian randomization supports this conclusion as they failed to find any causal role between adipocytokines and EC and other obesity driven cancers [90].
Exosomes represent a wide group of membrane-bound lipid particles that originate from the plasma membrane or the endosomal system and are secreted from cells. Exosomes released from both healthy and cancer cells, are abundant in body fluids and mediate cell-to-cell communication by shuttling DNA, RNA, lipids, metabolites, and proteins. In this way, exosomes are implicated in numerous physiological processes but also participate in the formation of the tumor microenvironment and cancer progression. Of all the biomolecules contained in exosomes, miRNAs seem to have the most clinical utility in EC diagnosis. MiRNAs are small non-coding single stranded molecules that regulate all hallmarks of cancer as defined by Hanahan and Weinberg, including proliferation, invasion, angiogenesis, as well as influencing cancer cells chemosensitivity [91].
With regard to the review of exosomal biomarkers, 15 studies were eligible for inclusion, and these reported on 29 different exosomal biomarkers and a total of 5527 patients. Amongst this group 2530 had a diagnosis of EC and 2456 were non-cancerous control-group patients [29][30][31][32][33][34][35][36][37][38][39][40][41][42][43]. Only Mi-RNA 21, 27a, and 223 were suitable for meta-analysis as they met the inclusion criteria and were reported by two or more studies. The summary ROC score was 0.825 (95% CI 0.735-0.915), 0.925 (95%CI 0.801-1.00), and 0.813 (95%CI 0.735-0.890) for MiRNA-21, MiRNA 27a and MiRNA-223, respectively. In comparison to HE4, the relatively high performance of MiRNAs is promising. However, it must be noted that MiRNA 27a included in a study by Ghazala et al. reported an AUC of 1.00 making it the perfect test in terms of sensitivity and specificity, positive and negative likelihood ratio [33]. This has not been replicated and has likely overestimated the performance of this biomarker on meta-analysis. The only other study by Wang et al., reported much lower AUC of 0.813 [32]. This may potentially be explained by the fact that Ghazala et al. used a healthy control cohort whereas Wang included a mixed cohort of those women with PMB who had both benign disease and normal endometrium. Both cohorts were relatively small in size with under 75 patients in each study and thus results must be interpreted with caution.
To the best of our knowledge this is the first study to report and meta-analyze the diagnostic accuracy of different classes sole biomarkers obtained by non-invasive biopsy for detection of EC and thus it may provide a reference guide for those wishing to explore candidate biomarkers for research or those wishing to assess the current evidence for implementation into current practice.
However, there are numerous limitations to this review and the evidence summarized within it. The main limitation is that there were a low number of small, heterogeneous studies for the majority of the evaluated index tests. It was not apparent from many studies whether a 'test-phase' had been conducted prior to the biomarker validation phase which is the point at which the diagnostic accuracy should be assessed. This may undermine the reliability of the summary estimates from the meta-analyses and is likely to have contributed to the marked variability in AUC and sensitivity and specificity seen for most index tests. For the vast majority of biomarkers, no meta-analysis was possible. The decision to meta-analyze using two studies or more was conducted based on numbers of study participants in each study and not solely on the number of studies conducted. However, it is evident that there was significant heterogeneity amongst the biomarkers analyzed. Formal assessments of heterogeneity, such as Eggers's test were deemed likely too unreliable given the low number of studies in most evaluations. A significant limitation of this review was also that the studies varied with respect to the control group used, the type, stage and grade of endometrial cancer, the age of the cohorts assessed and the cut-off thresholds for index tests. Additionally, most of the included studies evaluated the diagnostic cut-off thresholds using a ROC analysis without any subsequent validation in an independent cohort. Lack of validation of the diagnostic data in conjunction with the low number of studies for the majority of the presented tests contributed to the low quality of evidence presented in this review. A standardized methodology for fluid biospecimen collection, processing and storage was published in 2014 and would likely improve the quality of studies if adopted for use by future studies.
The variation in the selection of the case and control groups with inclusion of participants that may not be reflective of the EC population is also a limitation in this systematic review. The recent change towards molecular classification has shifted clinicians away from describing EC cell types as just serous, endometrioid or other histological cell types however some studies made no attempt to sub-classify their EC population other than to report grade and stage. Subgroup analysis of data was considered but due to the size of the studies and relative paucity of studies discussing the same biomarker this would have yielded no more reliable information.
Most studies do not report whether they attempted to reduce selection bias by consecutively enrolling participants. More than two-thirds of the included studies (38/56, 68%) had a two-gate design and included a wide group of participants who underwent surgery for various indications. Inclusion of healthy asymptomatic individuals or participants with other pathological conditions represents a potential selection bias with regard to the control group, which could have biased the test outcomes. The majority of studies involved a normal healthy control group which was either age-matched or in some cases non-age matched. This is likely to overestimate diagnostic accuracy of a biomarker more than in studies where a one-gate design is used and patients with a presenting symptom such as PMB go on to be part of the case or control group.
We suggest cautious interpretation of the presented results. Although studies demonstrated diagnostic potential for a number of tests, the level of heterogeneity, wide confidence intervals and risk of bias in many studies included in this review undermine reliability of the presented results, and hence these data are insufficient to confidently inform clinical practice at this stage. Additional biomarkers, reported in individual studies, displayed diagnostic estimates that qualified for either replacement or triage tests; however, there were not enough data for a meaningful recommendation on the use of any of these tests. As with all biomarkers and index tests, the next phase of validation will be to assess diagnostic accuracy amongst a non-symptomatic, low risk, low prevalence cohort in order to assess its performance as a true screening test.

Conclusions
In conclusion this systematic review and meta-analysis has sought to summarize the existing literature on the performance of non-invasive biomarkers for suspected EC triage and diagnosis. There is a clear need for a biomarker with a low false positive rate that can be used in primary care to reduce the number of unwarranted, invasive investigations. It has become clear from review that many biomarkers are still at discovery phase, rather than validation phase and are thus not at a stage where diagnostic accuracy should be assessed. There is also a high degree of heterogeneity between studies, this is likely due to studies reporting on different types of EC. Given that the dualistic Bohkman classification of type-1 and type-2 EC fails to adequately differentiate between EC or provide useful estimates of prognosis, it is likely studies reporting biomarker performance stratified by EC cancer subtype will demonstrate different biomarkers detecting different subtypes of EC. At the very least with more literature available, subgroup meta-analysis will become possible. Furthermore, it is clear from the review of current literature that two-gate study design may inflate the diagnostic accuracy of the biomarkers studied. Healthy control group patients are not reflective of the patient population undergoing evaluation of suspected EC. Until there is a diagnostic biomarker identified that is likely to be capable of high-performance triage amongst the symptomatic population, studies at the biomarker discovery or validation phase should consider inclusion of both healthy and symptomatic control cohorts.
There is wide recognition that an accurate non-invasive test for EC triage is likely to confer several advantages over the current standard of ultrasound scan and endometrial biopsy. These potential advantages include a reduction in cost (both in direct medical costs and in time off work), reduced discomfort, shorter recovery times and a reduction in the rate of serious complications associated with surgery. Another benefit of an accurate, non-invasive diagnostic test for EC is the prospect of early diagnosis and timely therapeutic interventions to minimize disease progression. Whilst this review highlights several methodological issues with the current body of evidence, there are some promising findings; exosomal compounds, in particular MiRNAs have shown moderate to good performance in the limited available data, but perhaps more reassuringly less heterogeneity between studies. Similarly, amongst proteomic compounds serum YKL-40 and DJ-1 have good to excellent performance and a next step might be to consider larger scale prospective evaluation of these biomarkers in order to determine their utility.