Nine Years

Endometriosis affects about one in ten women of reproductive age. It causes chronic pelvic pain, painful periods, fatigue, and often infertility. Doctors have written about it for more than a century.

And for most of that century, the average time between a woman's first symptoms and an actual diagnosis has been seven to nine years.

Nine years is not a rounding error. Endometriosis is not rare. Its symptoms are not subtle. The tools to investigate it have existed for decades. The delay exists for a less comfortable reason: the symptoms were dismissed, misread, or filed under "normal period pain," usually because the person reporting them was a woman.

What changed recently is not the biology. What changed is that a machine learning model, trained on microRNA found in saliva, can now detect endometriosis in a matter of days.

What endometriosis is

Tissue similar to the lining of the uterus grows outside the uterus, on the ovaries, the fallopian tubes, and sometimes other organs. It responds to the menstrual cycle like the real lining does. It thickens, breaks down, and tries to shed. But it has nowhere to go. The result is inflammation, scarring, and adhesions that can bind organs together.

Around 190 million women live with it. It is not curable, but it is manageable. Hormonal therapy can slow it. Surgery can remove lesions and improve fertility outcomes. Early diagnosis gives a patient more options, and gives her those options for longer.

Which is why the nine-year delay is not just a failure. It compounds.

The old pathway

Until recently, the only certain way to diagnose endometriosis was laparoscopy: surgery under general anaesthetic, a camera inserted through a small cut in the abdomen, a doctor looking for the tissue with her own eyes.

Getting to that surgery took years. A woman would see a GP. Her pain would be blamed on normal cramps, or IBS, or anxiety. A referral to a gynaecologist was not guaranteed. Even if she got one, an ultrasound or MRI could suggest endometriosis but could not confirm it. Only surgery could.

So the system worked like this: to be diagnosed, you had to be sick enough, stubborn enough, and lucky enough. Many women were not all three.

What changed

The Endotest, made by the French company Ziwig, looks for a pattern of microRNA in saliva.

MicroRNAs are short RNA molecules, roughly 22 nucleotides long. They do not code for proteins. They bind to messenger RNA and either block it or mark it for destruction. Think of them as volume knobs for genes. Humans have over 2,000 of them, and their levels shift with tissue type, cell type, and disease.

In endometriosis, the inflammation changes which microRNAs get turned up and which get turned down. And microRNAs are tough little molecules. They travel through the body in tiny protective bubbles called exosomes, and they show up in saliva. So the saliva of a woman with endometriosis carries a different microRNA signature than the saliva of a woman without it.

In 2022, Bendifallah and colleagues collected saliva from 200 women with chronic pelvic pain and sequenced every microRNA they could find. The question was which of the thousands of microRNAs actually predicted the disease.

The machine learning part

This is a hard problem. You get thousands of features from sequencing but only a few hundred patients. When features vastly outnumber samples, models tend to memorize the training data instead of learning anything real.

The study used a Random Forest. A single decision tree learns rules like "if microRNA X is high and microRNA Y is low, predict endometriosis." One tree alone will memorize. So a Random Forest builds hundreds of trees, each trained on a random slice of the data and a random slice of the features. Every tree votes. The majority wins.

The randomness is the point. No single noisy microRNA can take over the model, and the errors of individual trees cancel out. The result is a classifier that reads a pattern across many microRNAs, none of which would mean much on its own.

That is the Endotest. Spit in a tube, sequence the RNA, let the forest vote. Days, not years.

Where it stands

The test launched in France in February 2025 inside a clinical trial called Endobest. The trial aims to enroll 2,500 patients, with 25,000 women getting free access under an innovation coverage program. Each test costs €839, paid by national health insurance. More than 100 endometriosis centers are taking part.

Researchers are still cautious, and they should be. A survey of German gynecologists found opinion split on whether the test is actually useful in practice. There is a real difference between diagnostic accuracy, which asks how well a test spots the disease in a study, and clinical utility, which asks whether using it changes what happens to the patient. The second question is still open. Endobest is designed to answer it.

What nine years means

The signature was always there. The microRNA was sitting in the saliva of every one of those women the whole time. The biology did not change. Someone just finally looked.

For nine years on average, women had symptoms that were real, measurable, and often severe. The tools to investigate them existed. What was missing was the decision to take the pain seriously enough to go looking.

A model trained on spit cannot fix that history. What it can do is make the next nine years shorter, gentler, and less dependent on a woman's stubbornness in a system that was never built to believe her quickly.

That is the advance. Not that the algorithm is clever, though it is. It is that there is one less reason to wait.