FDA Advisers Split Six to Four on Whether the Cancer Test Works

26 SEP 2026 — A United States advisory panel took three votes on GRAIL's Galleri blood test on 23 September. Effectiveness carried six to four, safety passed unanimously, and benefits against risk passed seven to two with an abstention.

The results of the trial behind the test were published in full the day before, and it did not meet the endpoint it was designed to meet.

What the panel actually voted on

The Molecular and Clinical Genetics Devices Panel of the Medical Devices Advisory Committee has ten voting members, and it took three separate votes rather than one.

The spread between the three is the useful part. Safety asks whether the test harms the people who take it; effectiveness asks whether it does what it claims. The panel was unanimous on the first and closest to deadlocked on the second, and it is the second that maps onto what the trial was built to measure.

GRAIL notes that the Food and Drug Administration is not bound by the committee's recommendations, though it takes them into consideration. The company submitted its premarket approval application on 29 January 2026 and expects a decision in the coming months. The test was designated a Breakthrough Device in 2018.

The trial results, now peer reviewed

In August this desk reported that the trial supporting the test had missed its primary endpoint, ahead of the hearing. At that point the results were a conference presentation. They are now published in the New England Journal of Medicine and Nature Medicine.

Queen Mary University of London, which ran the trial with the Cancer Research UK Cancer Prevention Trials Unit, states the outcome plainly. After three rounds of screening there was "no significant difference between the groups in the incidence of stage III and IV cancers combined across 12 prespecified cancer types".

That was the primary endpoint. NHS-Galleri randomised 142,250 people aged 50 to 77 in England and screened them annually for three rounds.

Two secondary findings run the other way. Stage IV cancers alone were diagnosed less often in the screened group, and more cancers were caught at stages I and II. Neither carries the same weight as the measure declared in advance.

The harms figures we lacked in August

Our August report noted that no false-positive or overdiagnosis figures from the trial were available. The new publications supply the first of those.

Specificity ran at 99.5 to 99.6 per cent across the three rounds, or fewer than one in two hundred tests producing a false positive, as Queen Mary puts it. About one test in a hundred came back positive in each round. Of those positives, between 46 and 58 per cent turned out to be cancer, and 937 participants with a positive test were diagnosed across the three rounds.

Where the test did find a signal, it predicted the cancer's location with better than 90 per cent accuracy within its first two predicted sites.

A false-positive rate is not the whole harms ledger. Overdiagnosis figures, which would say how many detected cancers would never have caused harm, are still not published, and neither is mortality.

What the company release leaves out

GRAIL's announcement of the vote does not mention the NHS-Galleri trial at all. It gives no specificity, sensitivity or predictive-value figure. Its one quantified claim is that Galleri has detected approximately four to seven times more cancers than standard of care alone.

The release describes the test as detecting cancer-specific methylation patterns before symptoms appear, in adults aged 50 and over, and is explicit that it is intended as an addition to guideline-recommended screening rather than a replacement for it.

The omission is not a factual error. The randomised trial that missed its primary endpoint and the favourable panel vote are simply being communicated separately, a day apart, by parties with different interests.