A saliva test for endometriosis reported 96.6% accuracy — in a 77% prevalence group

Endometriosis is diagnosed late, often years after symptoms begin, and definitive confirmation has historically meant laparoscopy. Any test that could shorten that path would matter to a very large number of people. A prospective multicentre external validation published in NEJM Evidence on October 28 reports diagnostic accuracy figures for a saliva-based micro-RNA signature that are, on their face, remarkable [s1].

What was tested

The ENDOmiRNA study set out to assess the diagnostic accuracy of a saliva micro-RNA signature, validate its biological reproducibility, and evaluate its clinical utility [s1].

Participants were 18 to 43 years old with signs and symptoms suggestive of endometriosis, recruited from diverse medical settings [s1]. Endometriosis was diagnosed by imaging, by laparoscopic procedure, or both [s1]. Everyone classified as a control underwent laparoscopy [s1]. Assessment of endometriosis status from the saliva signature was made blind to the patient's status as determined by imaging, laparoscopy or histology [s1].

The external validation population comprised 971 patients, including patients carried over from a prior interim analysis, with an overall endometriosis prevalence of 77% [s1].

The numbers

Accuracy — defined as the probability of correct classification for both positive and negative results — was 96.6% (95% CI 95.2 to 97.6) [s1].

Sensitivity was 97.3% (96.4 to 98.0) and specificity 94.1% (91.0 to 96.4) [s1]. Positive predictive value was 98.2% (97.3 to 98.9) and negative predictive value 91.3% (88.3 to 93.4) [s1]. The positive likelihood ratio was 16.6 (10.8 to 26.9) and the negative likelihood ratio 0.03 (0.02 to 0.04) [s1].

Among patients with surgical confirmation of the diagnosis, the saliva signature misclassified 4.6%, underestimated 2.4% and overestimated 2.2% [s1]. For imaging — transvaginal ultrasound, MRI, or both — the corresponding figures were 27.2%, 15.1% and 12.2% [s1].

That imaging comparison is the most consequential number in the paper, and also the one most likely to be quoted without its context.

The prevalence problem

Predictive values are not properties of a test. They depend on how common the condition is in the population being tested.

The validation population had 77% prevalence [s1]. That is a group of symptomatic patients referred for investigation of suspected endometriosis — by construction, an enriched population. Sensitivity and specificity, which are properties of the test, should carry across settings. Positive and negative predictive value will not.

At 77% prevalence, a specificity of 94.1% [s1] produces relatively few false positives relative to the large number of true positives, which is why PPV comes out at 98.2%. Applied to a general population of women with pelvic pain, where prevalence is much lower, the same specificity would generate a very different balance. The study does not model that scenario, and the paper's own framing is external validation of accuracy in this cohort [s1].

The reference standard

Diagnosis was established by imaging, laparoscopy, or both [s1]; controls all had laparoscopy [s1]. That is a stronger design than one where controls go unverified, because a negative laparoscopy is a real negative rather than an absence of investigation.

It is still an imperfect standard. Laparoscopy detects lesions a surgeon can see, and superficial peritoneal disease is not always visible. If the saliva signature detects biology that laparoscopy misses, some of what is scored as a false positive may not be. That would improve the test's true performance; it would also mean the reference standard cannot verify it.

Who funded it

The study was funded by Ziwig, and is registered as NCT05244668 [s1]. Ziwig developed the signature being validated.

Industry funding of a validation study of the funder's own product is common and is not by itself a reason to discount a result. It is a reason to weight independent replication heavily, and to note that the blinding described — signature assessment made without knowledge of clinical diagnosis [s1] — is the specific safeguard that matters here.

Why the diagnostic question is worth this much attention

A separate analysis published the same day, drawing on the Global Burden of Disease Study 2021, reported that the burden of endometriosis rose globally and in most countries between 1990 and 2021, even as age-standardised prevalence, incidence and DALY rates fell [s2]. Decomposition analysis attributed the increase mainly to population growth [s2].

That analysis found the highest incidence rates in the 20-24 age group, with disease burden peaking at 25-29 [s2], and projected that burden in age groups from 15 to 54 may continue rising through 2050 despite declining age-specific rates [s2]. It also found burden decreasing as the sociodemographic index rises [s2].

The population most affected is young, and the diagnostic delay falls in the years when it costs most.

What to watch

Independent validation by groups without a commercial interest in the signature; performance in lower-prevalence populations, where predictive values will differ; whether regulators in any jurisdiction evaluate the test; and whether earlier diagnosis, if achieved, changes outcomes rather than only timing.

This article describes diagnostic accuracy research and is informational only. It is not medical advice and does not recommend any test.

Sources

• [s1] Validation of a Saliva Micro-RNA Signature for Endometriosis, NEJM Evidence, 2025;4(11):EVIDoa2400195, published 2025-10-28. Funded by Ziwig; ClinicalTrials.gov NCT05244668.
• [s2] Global, regional, and national burden and trends of endometriosis, 1990-2021: an analysis of the global burden of disease study 2021 and forecast to 2050, BMC Women's Health, 2025;25(1):519, published 2025-10-28.