On Tuesday the New England Journal of Medicine published the randomized trial that was supposed to show GRAIL's Galleri blood test works. The trial missed its primary endpoint. On Wednesday an FDA advisory panel voted that the test is safe and effective. Both things are true. Only one of them is evidence.

Galleri looks for methylation patterns in fragments of tumor DNA circulating in the blood. One tube, dozens of cancer types, a predicted organ of origin. It has been sold as a laboratory developed test since 2021 and holds FDA Breakthrough Device status. GRAIL now wants full premarket approval for adults 50 and older.

The panel voted 10 to 0 on safety, 6 to 4 on effectiveness, and 7 to 2 (one abstention) that benefits outweigh risks. The FDA does not have to follow the panel. It usually does.

NHS-Galleri randomized 142,250 people in England, aged 50 to 77, to three annual blood draws. Half had their blood tested. Half had it stored. The primary endpoint was the incidence of stage III or IV cancer across 12 prespecified deadly cancers. The logic is sound. If screening finds cancer earlier, fewer people should be diagnosed late.

They were not. The incidence rate ratio was 1.03 (95% CI 0.92 to 1.14, P = 0.63). The screened group had slightly more late stage cancer than the control group, a difference best described as zero. The key secondary endpoint, stage IV alone, came in at 0.86 (95% CI 0.74 to 1.00). That is a signal worth following. It is also a secondary endpoint whose confidence interval touches the line of no effect, in a trial funded by the company.

There are no mortality data yet. The authors call for further follow up, which is the correct and honest thing to say.

The NEJM paper lists thirteen authors. The lead is Peter Sasieni of Queen Mary University of London, a cancer screening statistician with a long record in cervical and bowel screening trials. Peter Johnson was NHS England's national clinical director for cancer, and the NHS ran the trial through mobile clinics. Charles Swanton of the Francis Crick Institute is one of Britain's best known cancer biologists. Four authors work for GRAIL Bio UK, including Sir Harpal Kumar, who ran Cancer Research UK before joining the company. GRAIL funded the trial.

That is a lot of institutional weight behind a hoped for result. NHS England had a provisional agreement to roll the test out widely if early data looked strong, and in 2024 it chose to wait for these final results instead. That decision now looks wise.

Give the authors credit. The abstract states the null primary result in its first finding and calls for more follow up rather than spinning the stage IV number. The spin happened downstream, in the press release. The accompanying NEJM editorial, by Oxford epidemiologist David Hunter, concludes the test is not yet ready for population screening.

Screening tests are judged on two numbers that pull against each other. Sensitivity is how many real cancers the test catches. Specificity is how many healthy people it correctly leaves alone. Galleri was engineered for specificity, and it delivers. Across three rounds, specificity was 99.5 to 99.6 percent. Very few healthy people get a false alarm.

Sensitivity is where it struggles. Across all cancers diagnosed within a year of a test, Galleri caught 27 to 37 percent. For the 12 deadly cancers it was built to find, 48 to 63 percent. In the US PATHFINDER 2 study, overall sensitivity was 39 percent. Roughly, the test finds one cancer in three.

Here is round one of the trial, scaled to 1,000 people.

The positive predictive value was 58 percent in round one and fell to 46 percent by round three. A positive result is close to a coin flip. That is actually good for a cancer screen, and the organ of origin prediction was right more than 90 percent of the time, which keeps the diagnostic chase short.

The negative result is the quiet problem. Before testing, 98.4 percent of the round one group was cancer free. After a negative Galleri, 99.0 percent was. A negative test moves the needle about half a point. It feels like a clean bill of health. It is a slightly better guess.

Nobody here is anti-science. The trialists ran a clean, huge, randomized study and reported the null result in the first sentence. GRAIL did the trial that most diagnostics companies never do. The panel asked sharp questions and wrote them down.

The order of events is backwards. A panel vote and a press release arrived the day after a missed primary endpoint, and the headlines followed the vote. Several panelists worried that approving on detection data alone sets the bar for every multicancer test that follows. They are right to worry. Once "detects cancer" becomes the standard, "helps patients" becomes optional.

Measuring whether a test finds disease is easy. Measuring whether finding it changes anything takes years, and the answer is allowed to be no.