Journal Club: NHS-Galleri (GRAIL)

A note on sourcing: This journal club is based on a conference presentation rather than a peer-reviewed publication. The data below come from the ASCO 2026 late-breaking abstract LBA100 and the accompanying presentation. A full manuscript has not yet been published at the time of writing, so the detailed methodology and subgroup analyses are not yet available for independent scrutiny. Update, October 2026: the full manuscript was published in the New England Journal of Medicine on September 22, 2026; see the last section of this post.

Study Design

NHS-Galleri is the largest randomized controlled trial ever conducted for an MCED test. It was launched in England in 2021 and randomized 142,250 adults aged 50 to 77 with no symptoms or suspicion of cancer. Participants were randomized to either an intervention arm, in which they underwent testing with Galleri, or to a control arm, in which they did not undergo testing with Galleri and instead underwent “routine” care. In the intervention arm, a “cancer signal detected” result triggered referral into the NHS for diagnostic workup.

The primary objective was a reduction in the combined incidence of stage III and stage IV cancers across 12 prespecified cancers (lung, head & neck, colorectal, pancreas, myeloma/plasma cell neoplasm, liver/bile duct, stomach, esophagus, anus, lymphoma, ovary, and bladder), measured roughly three years after the last participant was randomized. A reduction in stage IV cancer was a secondary objective. So, this trial was built to test whether the test could produce a “stage shift” (catching cancers earlier than they would’ve otherwise been caught) rather than whether the test leads to fewer deaths.

The study measured something called an “incidence rate ratio” (IRR) for each arm. The IRR is the rate of new cancers in the intervention arm divided by the rate in the control arm. Therefore, a value of 1.0 means no difference between the groups, while a value below 1.0 means fewer cancers with screening, and a value above 1.0 means more cancers with screening.

Results

The primary endpoint was not met. Across the 12 prespecified cancers, there were 706 stage III and IV cancers in the intervention arm and 688 in the control arm, for an IRR of 1.03 (95% CI 0.92-1.14, p = 0.63). In the first (prevalence) round, the Stage III and IV IRR was 1.19, which is roughly what would be expected when a relatively sensitive test is first used in a population and pulls forward a backlog of existing cancers. By the second and third rounds, the IRR fell to 0.95 and then 0.88.

The secondary endpoint was positive. The IRR for stage IV cancers was 0.86 in the intervention arm (342 stage IV cancers in the intervention arm vs 397 in the control arm; 95% CI 0.74-1.00), and, like the IRR trend seen for stage III and IV cancers, the effect grew with each round of screening (a 9% reduction after year 0, 22% after year 1, and 26% after year 2).

The number of stage I and II cancers was higher in the interventional arm (647 in the interventional arm vs 559 in the control arm; relative risk 1.16, 95% CI 1.03-1.30), as were the number of stage I-III cancers (1,007 in the interventional arm vs 846 in the control arm; RR 1.19, 95% CI 1.09-1.30).

Across all cancer types, 3,637 cancers were diagnosed in the intervention arm and 3,400 in the control arm over three years of screening. Adding the MCED test quadrupled the number of “screen-detected cancers” (1,173 in the interventional arm vs 290 in the control arm). It is my understanding that “screen-detected cancers” are those that are found in individuals without symptoms or signs of cancer. Additionally, adding the MCED test reduced the number of cancers detected clinically (meaning after they caused symptoms or other problems) by 21% (2,464 in the interventional arm vs 3,110 in the control arm), and reduced emergency presentations of cancer by 21% (225 in the interventional arm vs 286 in the control arm). The test’s aggregate performance (99.55% specificity, 52.0% PPV, and 92.5% cancer signal origin accuracy for the top two predicted cancer origin sites) was consistent with what GRAIL has reported in its other studies described in this newsletter. With regard to safety, there were 381 related adverse events in the intervention arm and 333 in the control arm, and no related serious adverse events. Not much information was provided on these adverse events in the conference abstract.

What I Think

The good news:

1. In my opinion, this is the most important dataset in the field, and parts of it are encouraging. Specifically, the 14% reduction in stage IV cancer diagnoses (IRR 0.86, 95% CI 0.74 to 1.00 in the published paper, a secondary endpoint that was not formally tested) is, in my opinion, clinically meaningful.
2. The fact that testing with Galleri led to fewer emergency presentations and fewer cancers presenting clinically with symptoms suggests the test changed how and when cancers are found in a large, real-world health system (not just in a research cohort).

That said, the headline is the headline. The primary endpoint was not met.

1. The trial was designed and powered to show a reduction in combined stage III and IV cancer across 12 different types of cancers, and it did not. As stated above, the favorable trends in the later rounds and the stage IV reduction are encouraging, but they are, formally, secondary. Reading too much into secondary and exploratory findings from a trial that missed its primary endpoint is exactly the kind of thing I try to be careful about elsewhere in this newsletter.
2. A stage shift is a surrogate endpoint, and it is not the same as showing that screened people live longer than unscreened people, which, to me, is the most important question for cancer screening tests.
3. The sensitivity of the test by stage was not reported. That omission limits interpretation, because across every other study covered here, early-stage sensitivity has been the weakest performance metric, and it is the number I would most want to see.
4. The data were from a conference presentation when this was written. The full manuscript was published in the New England Journal of Medicine on September 22, 2026 (see below).

Published in the New England Journal of Medicine

On September 22, 2026, the primary results were published in the New England Journal of Medicine (Sasieni P, Johnson P, Round T, et al. Effect of screening with multicancer early-detection test on late-stage cancer diagnosis. N Engl J Med. Published online September 22, 2026. doi:10.1056/NEJMoa2505723), with a companion paper on the test’s performance in Nature Medicine (Neal RD, Dolly S, Johnson P, et al. Performance of a multi-cancer early detection test in the randomized controlled NHS-Galleri trial. Nat Med. Published online September 22, 2026. doi:10.1038/s41591-026-04652-8).

A few figures differ from the ASCO 2026 abstract, and the figures above have been updated to the paper. The NEJM paper reports that 142,826 people consented and 142,250 were randomized (71,122 to the intervention group and 71,128 to the control group), aged 50 to 77 at invitation; the abstract reported 142,924 enrolled, aged 50 to 79. The primary endpoint result is unchanged (706 vs 688 stage III or IV cancers among the 12 prespecified types; IRR 1.03, 95% CI 0.92 to 1.14, P = 0.63). The stage IV result is 342 vs 397 (IRR 0.86), and the 95% confidence interval is printed as 0.74 to 1.00 in the paper, where the abstract gave 0.744 to 0.998. Clinically detected cancers were 20.8% fewer in the intervention group (2,464 of 3,637 vs 3,110 of 3,400), and diagnoses after an emergency presentation were 21.3% lower (225 vs 286). The authors state that the trial was not powered for mortality.

The companion paper gives the test’s performance by screening round rather than as a single aggregate: specificity of 99.56%, 99.60%, and 99.50% in rounds 1 to 3; PPV of 58.0%, 50.4%, and 45.8% (52.0% across all rounds); and 12-month episode sensitivity for all cancers of 37.2%, 27.1%, and 26.7%. Sensitivity by stage is still not reported; the authors write that it is not calculated because cancers could progress between testing and diagnosis.