The FDA’s Molecular and Clinical Genetics Panel sits down at 9 a.m. Eastern on Wednesday to tell the agency what it thinks of Galleri, the blood test GRAIL says can pick up a signal from more than 50 cancers in a single draw. Nearly every writeup of the meeting has been assembled from the same company materials, and the numbers they are all repeating come from the study that did not ask the hard question.
Here is the part almost nobody is leading with. The only randomized controlled trial ever run on this test enrolled 142,942 people in the United Kingdom, ran three years, and missed its primary endpoint. It did not produce a statistically significant drop in late-stage cancer diagnoses, which is the entire premise of catching cancer early. The FDA’s own meeting notice describes what the panel is weighing, and the application leans on PATHFINDER 2, a study with no control arm at all.
What Each Study Was Actually Built to Measure
PATHFINDER 2 is the good news, and it is real. OHSU researchers reported a specificity of 99.6 percent, meaning false alarms were rare. Of 287 participants whose blood carried a cancer signal, 173 turned out to have cancer, a positive predictive value of 60.3 percent. Added on top of the screenings the U.S. Preventive Services Task Force already recommends, the test multiplied the number of cancers found several times over.
But PATHFINDER 2 was a single-arm study. Everybody in it got the test. It can tell you how many cancers the test found and how often it was wrong. It cannot tell you whether the people who took it ended up better off than the people who did not, because there were no people who did not.
NHS-Galleri was built to answer exactly that. Adults aged 50 to 79 were randomly assigned to annual Galleri screening plus normal NHS screening, or normal screening alone. The question was whether three years of blood tests would cut the number of cancers caught at stage III or IV. In February the answer came back no, and as the ASCO Post summarized the full results, the primary endpoint was not met.
One study asked whether the test finds cancer. The other asked whether finding it helped. The FDA panel is being handed the first one.
The Secondary Results Are Not Nothing, and They Are Not the Endpoint Either
The NHS trial did produce findings worth taking seriously. Stage IV diagnoses fell in the screened group. So did emergency presentations, the awful route where a person learns they have cancer because they collapsed. There was a favorable trend across a pre-specified group of twelve of the deadliest cancers, most of which have no routine screening at all. Roughly 70 percent of cancers have no USPSTF-recommended test, which is the real hole in American screening and the reason this technology keeps drawing serious researchers.
None of that is the endpoint. Trials name a primary endpoint in advance precisely so that a disappointing result cannot be renegotiated afterward into a collection of encouraging secondary numbers. The Cancer Letter’s account of the panel referral lays out how much weight the single-arm data is being asked to carry. The American Academy of Family Physicians called multi-cancer detection a promise yet to be proven and told primary care physicians to use these tests inside research studies and not in place of screenings that work. That advice was written before the randomized data existed. The randomized data did not make it obsolete.
Where LiveNewsChat Lands on This
Approve it if the panel thinks the balance of evidence justifies it, but do not let the approval pretend the NHS result did not happen. A screening test is not a diagnostic. Its whole claim on a healthy person’s blood and a healthy person’s anxiety is that using it changes what happens to them, and the one experiment designed to test that claim came back negative on the measure it chose for itself.
The failure mode here is not a bad test. It is a good test approved on the wrong question, marketed on the strength of detection counts, and then quietly absorbed into routine care before anyone establishes that detection counts translate into people living longer. We have watched that movie in screening before, and the sequel is always overdiagnosis: treating cancers that were never going to hurt anyone, with surgery and radiation that certainly will.
So the panel’s job today is narrower than approve or reject. It is to say out loud, in the record, what the evidence supports:
- That Galleri reliably detects a cancer signal, with a false-positive rate far lower than most screening most Americans already undergo.
- That a three-year randomized trial in 142,942 people did not show a significant reduction in late-stage diagnosis.
- That nobody has yet shown the test distinguishes cancers that will kill you from cancers that will sit there.
- That the label and the marketing should carry all three of those facts, not the first one.
The Part That Has Nothing to Do With the Science
Galleri currently lists at $949, with a self-pay price around $799, and Medicare and most insurers do not cover it. FDA approval is the gate that changes that, which is the commercial reason this meeting matters more than a typical advisory panel. NPR reported in June that these tests can already be ordered in the United States under a separate pathway, and approval is what moves them from a thing affluent people buy to a thing the health system pays for at scale.
That is not an argument against approving it. It is an argument for being honest in the paperwork, because once a screening test is covered it stops being a choice anyone deliberates over and starts being something a physician orders at a physical. We have written before about how differently the same health number reads depending on who is doing the counting, and this is the same problem wearing a lab coat: the figure that ends up on the marketing page is the one that survives, and right now the figure winning is 60.3 percent, not 142,942.
The panel’s recommendation is not binding. The FDA can take it or leave it. What the panel can do is refuse to let the smaller study speak for the bigger one, and put the awkward result where a doctor, and a patient, will actually see it. Stanford’s clinicians laid out five things to know about blood-based cancer testing this month, and the most useful of them is the least marketable: a negative result does not mean you do not have cancer, and a positive one is the beginning of a workup, not an answer. Whatever comes out of the vote tonight, that is the sentence that should travel.