The week a 200,000-woman cancer screening trial finally explained why it failed

Biotech & Longevity 5 min 80 sources

A near-perfect ovarian cancer test still saved no lives — a lesson in why a rare disease defeats even an accurate screen. Plus a prenatal blood test reads 23,000 genes, an AI flags a clot a doctor missed, and HPV shots push cervical-cancer death toward zero.

Key takeaways

A 200,000-woman trial showed a 99.8%-accurate ovarian cancer screen saved no lives, because the deadliest form moves too fast to catch in time.
The deeper lesson: a test's accuracy means little on its own — when a disease is rare, even a great test produces far more false alarms than real finds.
Prevention beat detection this week: HPV shots given at 12 or 13 cut the risk of dying from cervical cancer before 30 to almost zero.

For twenty years, the most ambitious ovarian cancer screening trial ever run gave an answer nobody wanted: screening healthy women didn’t save their lives. This week, scientists finally published the clearest account of why [3]. The lesson runs underneath half the health news in this edition — about what a test result actually means when the thing you’re hunting is rare.

The screen that worked and still failed

The UK Collaborative Trial of Ovarian Cancer Screening, or UKCTOCS, enrolled 200,000 women starting in 2001 [3]. One group got a yearly blood test for a protein called CA125, followed by an ultrasound if the protein looked off. CA125 is a “biomarker” — a substance in the blood whose level can rise with disease.

On paper the test was good. An early analysis found it caught 89% of cancers (its “sensitivity”) and correctly cleared 99.8% of healthy women (its “specificity”) [3]. Those are strong numbers. And yet, after a median 16 years of follow-up, women who were screened died of ovarian cancer at almost exactly the same rate as women who weren’t [3].

The new secondary analysis, published in the British Journal of Cancer, dug into the individual patient data to find the reason [3]. The answer is biology, not a faulty test. The deadliest and most common form — high-grade serous cancer, which causes an estimated 85% of ovarian cancer deaths — moves too fast [3]. The window in which CA125 could catch it early was under six months [3]. By the time a yearly test sees it, the cancer has usually already spread.

The researchers ran simulations: even screening continuously, the best achievable was a 15% drop in deaths [3]. To do meaningfully better, you’d need a test that spots the cancer a full year earlier than CA125 can — a test that doesn’t yet exist [3]. The conclusion isn’t “screening is pointless.” It’s that the current tools can’t catch this cancer in the brief moment when catching it would help [3].

When a near-perfect test still floods you with false alarms

Here is the part worth carrying past today. A test’s accuracy is only half the story. The other half is how rare the disease is in the people you screen.

Run that 99.8% specificity against a healthy population where ovarian cancer is uncommon, and even a tiny error rate produces a pile of false alarms relative to the true cancers found — because there are so many more healthy women than sick ones. The test isn’t broken. The math of rarity is doing the damage. This is why a screen that looks excellent in the lab can disappoint in the field, and why regulators are cautious about rolling out broad cancer-screening blood tests before efficacy trials report [3].

That caution matters now, because a wave of new “multi-cancer early detection” blood tests is heading into exactly these kinds of trials [3]. The UKCTOCS post-mortem is a warning label for all of them: a promising biomarker is not the same as lives saved.

Where genetics changes the picture — and where it doesn’t

Rarity cuts the other way when you can find the people who are genuinely high-risk. Heather Morgan, a 59-year-old from Monmouthshire in Wales, carries a BRCA1 gene mutation — an inherited fault that raises lifetime ovarian cancer risk from about 2% in the general population to 44% [51]. She wasn’t offered genetic testing after an earlier breast cancer because of where she lived; she later developed ovarian cancer and now faces a 35% ten-year survival estimate [51]. In a high-risk group, finding the mutation lets doctors act before cancer starts. The screen that fails across a whole population can be exactly right for the rare few who carry the risk.

Genetic reading is getting more powerful elsewhere too. At a European genetics conference, scientists described a maternal blood test, called non-invasive foetal sequencing, that reads fragments of a foetus’s DNA floating in the mother’s blood [42]. In a validation study of 565 pregnancies, it picked up 95–99% of the genetic variants found by amniocentesis — the needle test that carries a roughly 1-in-200 miscarriage risk [42]. The catch is the same base-rate problem in a new coat: used broadly as a screen, it could flag thousands of “variants of unknown significance” and frighten parents over findings that may mean nothing [42]. The researchers frame it as a follow-up test when an anomaly is already suspected, not a fishing expedition [42].

The case for prevention over detection

If catching cancer early is this hard, stopping it before it starts looks better and better. This week brought the strongest evidence yet for one such case. A study in The Lancet, funded by Cancer Research UK, found that girls vaccinated against HPV at 12 or 13 now have a near-zero risk of dying from cervical cancer before age 30 [71]. High-risk HPV causes 99% of cervical cancers, and the vaccine prevents about 90% of them [71]. The result is what screening kept promising and couldn’t deliver: not earlier detection, but the cancer simply not happening. The authors warn the gains could reverse, though — fewer teenagers are getting the shot [71].

Quick hits: the therapies clearing regulatory hurdles

Two gene-therapy programmes moved this week. The US Food and Drug Administration, the country’s drug regulator, told the Dutch company UniQure that three-year data from an early-to-mid-stage trial is enough to file for accelerated approval of its Huntington’s disease gene therapy — a one-time treatment for a fatal inherited brain disorder with no cure [30]. UniQure plans to file in the third quarter [30]. Separately, Intellia reported phase 3 results for a one-time CRISPR-based treatment for hereditary angioedema, a rare disorder that causes dangerous swelling attacks [4]. Phase 3 is the large, final test before approval. Both are early steps, not finish lines — “filed for approval” is not “approved,” and accelerated pathways still require the drug to prove out over time. But for diseases with nothing else, a credible shot on goal is the news.

02 · Lesson · why it matters

Why a test that's almost never wrong can still be wrong about you

A test's accuracy tells you how good the test is. It does not tell you what your result means. For that, you need to know how common the thing is in the first place.

A blood test catches 89 of every 100 ovarian cancers and correctly clears 99.8% of healthy women. Those are the numbers from a real trial of 200,000 women, and they sound like the test you’d want. The trial ran for twenty years and saved no lives. The puzzle isn’t why the test was bad. It wasn’t. The puzzle is why a near-perfect test can still mislead — and the answer is one of the most useful ideas in all of medicine, because it applies to every screen, every scan, every diagnostic alert you will ever face.

The two numbers a test gives you

When someone says a test is “accurate,” they usually mean two separate things. One is sensitivity: of the people who have the disease, what fraction does the test catch? The ovarian screen caught about 89 of every 100 real cancers. The other is specificity: of the people who are healthy, what fraction does the test correctly clear? Here the screen scored 99.8% — it wrongly alarmed only 2 healthy women in every 1,000.

Both numbers are properties of the test. They describe how the test behaves when you already know who is sick and who is well. That is the trap. When you take the test, you don’t know which group you’re in. That’s the whole reason you’re testing. So the two numbers, on their own, can’t answer the question you actually care about: my result came back positive — what are the odds I’m actually sick?

Why rarity changes everything

To answer that, you need a third number the test can’t give you: how common the disease is in the people being screened. Call it the base rate.

Picture 100,000 healthy-seeming women. Ovarian cancer is rare, so maybe 20 of them actually have an early cancer the test could find. Run the screen. It catches about 18 of those 20 — good. But it also wrongly alarms 0.2% of the 99,980 women who are fine. That’s about 200 false alarms. So of the roughly 218 women told “something looks off,” only 18 actually have cancer. Most of the alarms — more than nine in ten — are false, not because the test is sloppy, but because there were so many more healthy women to misfire on than sick women to find.

Nothing about the test changed. The same 99.8% specificity that sounds flawless produces a flood of false positives the moment you point it at a population where the disease is rare. The rarer the disease, the worse the flood. This is the base rate doing the damage, quietly, behind a number that looked like a guarantee.

The same math, turned around

Now run the identical test on a different group: women who carry a BRCA1 gene fault, where lifetime ovarian cancer risk isn’t 2% but 44%. Suddenly the room is full of real cancers, not healthy women. The same test, with the same two numbers, now produces mostly true alarms. Its positive results actually mean something.

This is why one woman in the briefing — denied a genetic test because of where she lived — represents the case where finding the rare risk early would have helped, while the 200,000-woman screen of the general public did not. It’s the same disease and could be the same test. What differs is the base rate of the people you aim it at. A screen is not good or bad in the abstract. It is good or bad for a particular population, and the population’s risk level is half the answer.

Where you stand inside this

It’s tempting to read this as a fact about cancer screening and file it away. But you are inside this math every time a test of any kind comes back about you — a routine scan, a home test, an AI alert that flags something on a chart. When the result is positive and the underlying thing is rare, the honest reading is not “I have it.” It’s “the odds I have it went up, and the next step is a better test, not a conclusion.” The first number you should ask for is not how accurate the test is. It’s how common the condition is in someone like you.

This is also why the people who run health systems move slowly on broad new screening tools, even ones that look spectacular in the lab. They’re not being timid. They know that a test which dazzles on a group of known patients can disappoint, or even do net harm, when released onto millions of healthy people — through false alarms, anxiety, and unnecessary follow-up procedures that carry their own risks. The lab number and the field result are two different things, and the base rate is the bridge between them.

What’s left when you see the whole

The cleanest version of this whole lesson is the contrast running through today’s news: a screen that caught cancers but saved no lives, sitting beside a vaccine that pushed cervical-cancer deaths in young women toward zero. Detection fought the base rate and lost. Prevention sidestepped it entirely — you can’t get a false positive for a cancer that never starts.

None of this means tests are useless or that you should distrust your doctor. It means a test result is a piece of evidence, not a verdict, and its weight depends on something the result itself never shows you. The person reading the chart, the regulator weighing a new screen, you holding a positive result — all of you are working with the same incomplete picture, each able to see only your own corner of it. Hold the result a little more loosely than the number on the box invites you to. What it means depends on how rare the thing was to begin with — and that’s the part no test can print.

03 · Lab · your turn

Read the Positive

Screen a population and watch how a rare disease turns a near-perfect test's positives into mostly false alarms.

04 · Hope · carry this

A screen that saved no lives still taught us exactly what a better one must do — and the same week, a vaccine quietly pushed a cancer's death toll toward zero. Knowing why something failed is how the next thing succeeds.