31 AUG 2026 — GRAIL's Galleri blood test goes before a US Food and Drug Administration advisory committee on 23 September, the first multi-cancer early detection test to get that far. The trial that supports it missed its primary endpoint.
What the trial set out to prove, and did not
NHS-Galleri was a randomised controlled trial of annual screening in England, running three rounds across 142,000 demographically representative participants aged 50 to 77. That is a large, well-designed study, and randomised controlled evidence is what this field has been short of.
Its primary endpoint was a significant reduction in the incidence of cancers diagnosed at stages III and IV after three rounds of screening. The February 2026 readout did not achieve it.
Why a missed primary endpoint is not the same as a failed test
The two easy readings are both wrong. The distinction matters because it recurs across medicine.
A primary endpoint is nominated before the data exists. It is the single question the trial is powered to answer, and it is protected from the statistical problem that arises when you ask many questions of one dataset: ask enough, and some will return a positive result by chance alone. That protection is the entire reason the endpoint is declared in advance.
A missed primary endpoint means the trial failed to demonstrate what it was designed for. That does not invalidate the test or its secondary findings, and it does mean those other results have not met the same evidentiary standard and cannot be treated as if they had.
Read the 26 per cent carefully
The headline result being promoted is a 26 per cent reduction in stage IV diagnoses, among 12 particularly deadly cancers, by the third screening round.
Every clause in that sentence narrows it. Not all cancers, but a subset of 12. Not across the trial, but by the third round. Not the primary measure of stage III and IV combined, but stage IV alone. Each restriction may be clinically justified — late-stage diagnosis in aggressive cancers is exactly where a screening test should help, and a screening effect building over successive rounds is biologically plausible.
The problem is that the justifications arrive after the result. A subgroup defined in advance carries far more weight than one that emerges from the analysis, and from the outside it is usually impossible to tell which you are looking at. This is not an accusation against GRAIL, which reported the full results correctly at a major oncology meeting. It is a limit on what a reader can conclude from them.
The harms side of the ledger is missing from the coverage
Screening debates go wrong when they count only detections.
Any screening programme has costs that a detection rate misses. False positives send healthy people for investigation — for a multi-cancer test, that can mean whole-body imaging and invasive biopsy of an unspecified site. Overdiagnosis finds cancers that would never have caused harm, and treats them anyway. And a false negative can be worse than no test, because a person reassured by a clean result may dismiss a symptom that would otherwise have sent them to a doctor.
None of these harms appear in the four-fold higher detection rate from the promotional material, but they are exactly what the advisory committee will spend its time on. A screening test is approved or refused on whether its benefit outweighs those harms for a whole population, not on its ability to find cancer.
Why this hearing matters beyond one test
This is the first multi-cancer early detection test to reach an FDA advisory committee, and several others are behind it.
The panel's reasoning will set precedents for this class of test. It will have to decide whether shifting diagnosis to an earlier stage is an acceptable goal or whether only mortality data will do, how to weigh subgroup findings when a primary endpoint is missed, and what standards apply to a single test that screens for many cancers. Those questions have no settled answer, because no product has previously forced them.
Mortality is the crux. A test can shift diagnoses to earlier stages without anyone living longer, if the cancers caught early were the slow ones that would have been survivable anyway. Distinguishing those two worlds needs longer follow-up than three rounds provides, which is the honest reason this is difficult rather than a criticism of the trial.
What it means for this region
Health systems across ASEAN will face the adoption question with less data than the United Kingdom generated, and probably sooner than they would like.
Singapore, Malaysia and Thailand all have private healthcare markets in which a test like this can be sold directly to individuals well before any public system evaluates it, and a blood test marketed as detecting many cancers is close to an ideal consumer product. The trial's population was drawn from England, and cancer incidence patterns differ across this region, with markedly higher rates of nasopharyngeal and liver cancer than the trial's cohort would reflect.
The practical point for anyone offered this privately is that a test which missed its primary endpoint in a 142,000-person randomised trial has not been shown to save lives. It may yet be. The evidence for that is what the September hearing exists to weigh, and it is not in hand now.
View author profile → · Editorial policy
