Researchers at Mayo Clinic and collaborating institutions have developed a two-marker methylated DNA assay capable of detecting endometrial cancer with 96% sensitivity and 82% specificity using self-collected vaginal fluid samples from patients presenting with abnormal uterine bleeding. The findings, published in Gynecologic Oncology, represent a promising methodological step forward — though these performance figures come from a case-control validation study with enriched cancer prevalence and cannot be translated directly into clinical rule-out use. Several critical questions, including negative predictive value in a real-world population, must be answered before the assay could move toward clinical implementation.
Clinical Context: Why a New Diagnostic Tool Is Needed
Abnormal uterine bleeding affects approximately 1.4 million women annually in the US, yet only 5% to 10% of cases involve underlying endometrial cancer or atypical endometrial hyperplasia. Despite this relatively low pretest probability, current American College of Obstetricians and Gynecologists guidance recommends endometrial sampling in patients aged 45 years and older with abnormal uterine bleeding and in younger patients with risk factors.
The existing diagnostic toolkit has well-documented limitations. Office endometrial biopsy carries approximately 80% sensitivity for endometrial cancer detection, and dilation and curettage reaches roughly 90%. Although both remain standard diagnostic tools, neither is perfectly sensitive, and both carry procedural burdens. Nearly one in four patients who undergo office biopsy describe the procedure as "horrible" or "excruciating," and dilation and curettage requires anesthesia and carries surgical risk. Transvaginal ultrasonography, while valuable for evaluating structural causes of bleeding such as fibroids, is not reliable as a rule-out test for endometrial cancer in the abnormal uterine bleeding setting. Further, transvaginal ultrasonography performance data derive predominantly from European populations, raising generalizability concerns — particularly for aggressive endometrial cancer histologies, such as serous carcinoma, that disproportionately affect Black women and may not present with a thickened endometrial stripe.
These gaps define the clinical need this assay aims to address.
Study Design: Two Sequential Independent Experiments
This was not a prospective diagnostic-accuracy study in an unselected abnormal uterine bleeding population. It was a multicenter case-control marker optimization and validation study with enriched cancer prevalence — approximately 50% endometrial cancer in each experimental cohort. That design is appropriate for this stage of development, but it means the 96% sensitivity and 82% specificity figures cannot be translated directly into clinical rule-out performance. The paper itself identifies prospective validation in a real-world abnormal uterine bleeding population and determination of negative predictive value as essential future steps.
The broader study involved 15 institutions across two sequential experiments, while the independent marker selection phase enrolled participants from 14 institutions. Eligible participants included patients aged 45 years and older undergoing evaluation for abnormal uterine bleeding or postmenopausal bleeding, and patients aged 18 years and older with biopsy-confirmed endometrial cancer or atypical endometrial hyperplasia. All vaginal fluid samples were self-collected using unscented tampons — importantly, collection occurred in office or clinical settings, not at home. This distinction matters: the eventual intended-use case of home-based self-collection has not yet been evaluated. The current study builds on earlier pilot work in which a 28-marker panel discriminated endometrial cancer from benign endometrium with an area under the receiver operating characteristic curve of 0.88 using tampon-collected vaginal fluid.
DNA methylation — an epigenetic modification that alters gene expression and is commonly detectable early in carcinogenesis — served as the biomarker target. Bisulfite-converted DNA from tampon-collected vaginal fluid was assayed using long-probe quantitative amplified signal technology.
Marker Reduction Study
Beginning with 19 candidate methylated DNA markers identified through prior tissue-based discovery and vaginal fluid validation work, the investigators used random forest modeling and variable importance analyses in a cohort of 374 patients with benign endometrium, 256 with hysterectomy-confirmed endometrial cancer, and 37 with hysterectomy-confirmed atypical endometrial hyperplasia. The 19-marker model achieved 91% sensitivity and 87% specificity in an independent test set within this experiment. Notably, observed specificity in the independent test set slightly exceeded the training-set specificity target of 80%, a pattern that can occur in smaller validation cohorts. Nine markers with the highest variable importance were carried forward.
Marker Selection Study
This independent experiment enrolled 150 patients with benign endometrium and 150 with endometrial cancer, plus 31 with atypical endometrial hyperplasia, at 14 institutions. Three additional markers were reintroduced after re-mining tissue-based discovery data specifically to improve detection of atypical endometrial hyperplasia and low-grade endometrial cancer, bringing the candidate pool to 12 markers. A logistic regression model built in a randomly selected training set was then applied to an independent test set of 45 benign endometrium and 45 endometrial cancer samples.
The use of independent sample sets across both experiments reduces — but does not eliminate — concerns about overfitting, and the results should be understood as promising validation-stage evidence rather than established clinical performance.
Key Results
In the independent test set of the marker selection study, the final two-marker panel achieved 96% sensitivity for endometrial cancer, 82% specificity, and an area under the receiver operating characteristic curve of 0.97. Only two of 45 endometrial cancer cases were misclassified as benign — both were grade 1 endometrioid tumors, the most common and generally most indolent subtype.
Sensitivity by histologic subtype varied considerably and must be interpreted in the context of very small subgroup sizes. For grade 1 endometrioid tumors, the assay achieved 93% sensitivity. Among grade 2 endometrioid tumors, sensitivity was 100%. For grade 3 endometrioid tumors, serous carcinomas, and clear cell carcinomas — represented by just one, two, and one case respectively in the test set — sensitivity was reported as 100% in each subtype, but with confidence intervals as wide as 5% to 100%.
Those subgroup sizes are too small to support reliable subtype-specific conclusions. The study's strong overall sensitivity signal remains encouraging, but subtype-specific performance estimates should be considered preliminary until validated in substantially larger cohorts.
Clinical covariates including age, body mass index, smoking status, and menopausal status did not significantly affect assay performance in stratified analyses.
Atypical Endometrial Hyperplasia Performance: A Meaningful Limitation
Importantly, the assay was optimized primarily for detection of established endometrial cancer rather than precursor lesions such as atypical endometrial hyperplasia.
When applied to patients with a confirmed final diagnosis of atypical endometrial hyperplasia that was not upgraded to cancer, the assay achieved only 43% sensitivity. This remains a major limitation and should shape how clinicians interpret the assay's current capabilities. Atypical endometrial hyperplasia is clinically important for early detection and prevention strategies, but performance at this level means the current assay cannot yet reliably identify precursor disease. A negative result in a patient with known or suspected atypical endometrial hyperplasia should not be used clinically to exclude concurrent or evolving malignancy.
Within this context, one exploratory finding is worth noting. Among patients initially diagnosed with atypical endometrial hyperplasia on endometrial sampling who were subsequently found to have endometrial cancer at hysterectomy, the two-marker model correctly identified five of seven such "upgraded" cases. The authors suggest this raises the possibility that vaginal fluid–based methylated DNA marker testing may detect malignancy that preoperative tissue sampling misses. The finding is hypothesis-generating, but the sample size is small, confidence intervals are wide, and portions of the atypical endometrial hyperplasia dataset in the broader marker-selection analysis were not fully independent from earlier study phases. It should be understood as an interesting signal requiring prospective confirmation, not evidence of reliable atypical endometrial hyperplasia or precursor detection.
What the Study Cannot Yet Tell Us: The Negative Predictive Value Gap
For a test designed for eventual triage or rule-out use, negative predictive value is the most clinically actionable metric. It is also the one this study cannot yet provide.
Negative predictive value depends on both test performance and disease prevalence in the tested population. Because this was a case-control study with an artificially elevated cancer prevalence of roughly 50%, negative predictive value estimates derived from these data would not be valid in a real-world abnormal uterine bleeding population where endometrial cancer prevalence is approximately 5% to 10%.
In practical terms, clinicians still cannot answer the question patients will ask most directly: "If the test is negative, can I safely avoid a biopsy?"
Establishing negative predictive value in a prospective, real-world cohort is one of the primary objectives of the ongoing ENVISION study and is a prerequisite for any clinical rule-out application.
Where This Might Eventually Fit in the Diagnostic Pathway
Elle Kielar-Grevstad, PhD, described the envisioned clinical role of the assay in an interview with this publication:
The test would function as a triage tool for patients presenting with abnormal uterine bleeding, used early in the diagnostic pathway to help determine who may benefit from further workup. A negative result could help identify patients at low likelihood of underlying malignancy, while a positive result would support proceeding to endometrial biopsy for definitive diagnosis. This would complement existing approaches rather than replace them. Biopsy would remain the tool for diagnosis, and ultrasound would continue to play a role in evaluating structural causes of bleeding.
She added:
There are recognized limitations with current tools, especially when it comes to ruling out disease across diverse patient populations and cancer subtypes. A highly sensitive, molecular-based approach could provide an additional data point to support clinical decision-making and reduce unnecessary invasive procedures.
This triage framing is a reasonable vision for where the assay could eventually fit, but it remains contingent on prospective validation. The current evidence does not establish that the assay can safely reduce biopsies in routine abnormal uterine bleeding care.
Additionally, a specificity of 82% means roughly one in five cancer-free patients would screen positive and proceed to biopsy. That tradeoff may still prove clinically acceptable for a triage-oriented rule-out test if future studies demonstrate sufficiently robust negative predictive value in real-world abnormal uterine bleeding populations.
Equity Considerations
The study enrolled participants across 15 institutions, but the investigators acknowledge that the cohort was still predominantly White. This limitation is particularly relevant because aggressive endometrial cancer histologies — including serous carcinoma and carcinosarcoma — disproportionately affect Black women and are precisely the subtypes that current transvaginal ultrasonography-based evaluation may miss.
A molecular triage tool with high sensitivity across histologic subtypes could theoretically provide particular value in this population, yet the current evidence base does not adequately represent it.
According to Kielar-Grevstad, the ongoing ENVISION study is enrolling across 50 US sites and is designed to evaluate performance across variation by race, ethnicity, and geographic location. The investigators note that if methylated DNA markers are found to vary by race, they would consider repeating marker reduction and selection steps or modifying the analysis algorithm to include race as a variable.
Marker Identity, Proprietary Panel, and Conflicts of Interest
Clinicians should be aware that the two markers comprising the final panel — identified in the paper as Marker 21 and Marker 1 — are proprietary and are not publicly named. Because the markers are not publicly identified, outside investigators cannot independently assess their genomic identity or biological rationale from the publication alone.
Conflicts of interest are substantial and should inform how readers weigh the findings. Five co-authors, including lead author Jamie N. Bakkum-Gamez, MD, and senior author John B. Kisiel, MD, are inventors on Mayo Clinic intellectual property licensed to Exact Sciences and may receive royalties. Multiple additional co-authors are listed in the paper's disclosures as Exact Sciences employees, and several hold Exact Sciences stock. Exact Sciences provided research funding and salary support for the study.
These relationships do not invalidate the findings. The study design was methodologically rigorous, with sequential independent sample sets and transparent disclosure reporting. However, the financial relationships are material considerations for physicians evaluating an emerging commercial diagnostic technology.
The assay is not currently cleared by the US Food and Drug Administration or commercially available.
What ENVISION Will Need to Show
According to Kielar-Grevstad, the ongoing ENVISION study is designed to address several critical questions before clinical implementation could be considered.
Can the test reliably rule out cancer in the intended-use population? Establishing robust negative predictive value in a large prospective cohort of patients with abnormal uterine bleeding is the threshold requirement for any triage or rule-out application.
Will performance hold across a diverse patient population, including variation by race, ethnicity, geographic location, menopausal status, and cancer subtype?
Can sensitivity for earlier disease and atypical endometrial hyperplasia improve beyond the 43% observed here?
How will the assay perform when embedded into real-world clinical workflows, and how will clinicians and patients act on results?
And critically, can home-based collection be validated? The current study evaluated only office-based self-collection. ENVISION will need to demonstrate comparable performance with home collection before the assay could realistically function as a true previsit triage tool.
"Together, these data are intended to move beyond proof of concept and establish whether the test can be used safely, reliably, and equitably in clinical practice," Kielar-Grevstad said.
Bottom Line for Clinicians
This study demonstrates that a two-marker methylated DNA panel can detect endometrial cancer in tampon-collected vaginal fluid with high sensitivity in a selected case-control validation cohort. The results are promising and methodologically thoughtful: a parsimonious panel derived through sequential marker reduction and selection in independent sample sets, achieving strong overall performance in a multicenter cohort.
The assay's signal in identifying cancers missed on preoperative tissue sampling is also worth watching.
At this stage, however, the evidence does not establish that the assay can safely reduce biopsies in routine abnormal uterine bleeding care. That question depends on prospective validation in a real-world, diverse abnormal uterine bleeding and postmenopausal bleeding population with robust negative predictive value data.
Several limitations remain important: negative predictive value in the intended-use population has not yet been established; sensitivity in atypical endometrial hyperplasia was limited at 43%; subgroup sizes for aggressive histologies were too small to support reliable subtype-specific conclusions; the study population was predominantly White; only office-based self-collection has been evaluated; and the markers remain proprietary.
Physicians should follow the ENVISION study closely. If that study establishes robust negative predictive value across a diverse, prospective, real-world cohort, the case for incorporating this assay into the abnormal uterine bleeding diagnostic pathway will be substantially strengthened.
For now, this is a promising diagnostic innovation that still requires prospective intended-use validation before it could support routine clinical rule-out use.
Disclosures: Elle Kielar-Grevstad, PhD, is Director of Cancer Diagnostics Products at Abbott; she was an Exact Sciences employee and stockholder at the time the study was conducted. This article incorporates original interview material with Dr. Kielar-Grevstad conducted by this publication.
Source: Gynecologic Oncology