Galleri results from ASCO 2026: The Good, The Bad, and What's Next

By SpotitEarly Team · June 9, 2026 · 4 min read

I attended ASCO 2026 and dedicated a big chunk of my time to the multi-cancer early detection sessions, where the Galleri data drew most of the attention. More specifically, I listened to three presentation: Charles Swanton presented the NHS-Galleri randomized trial, Karthik Giridhar presented PATHFINDER 2, the registrational study in a U.S. intended-use population, and Mark Robson gave the discussant talk, titled "Potential versus Proof".

For anyone who has not followed this closely, Galleri is a blood test from Grail that reads cell-free DNA methylation to flag a cancer signal and predict where it came from. In cancer screening, the most important question is whether multi-cancer testing changes anything at the population level, and these large-scale studies were designed to answer this very question. Below are my thoughts and insights from these talks.

The Good

High Specificity

Specificity held near 99.55% in NHS-Galleri and 99.6% in PATHFINDER 2, per the data presented at the meeting. Positive predictive value improved across the program, from roughly 38% in the original PATHFINDER, to 52% in NHS-Galleri, to 60.3% in PATHFINDER 2 with a reported updated classifier.

Reaching the cancers recommended screening misses

The test also reaches cancers that current screening does not cover. Per PATHFINDER 2, about 67.6% of the cancers it found were types with no recommended screening at all, and adding the test to standard screening raised the number of screen-detected cancers roughly 6.5-fold. Sensitivity was highest for the twelve cancers responsible for two-thirds of U.S. cancer deaths, close to 69.8%, and cancer-signal-origin prediction was accurate about 91.3% of the time.

Good safety profile

The safety profile, which was the early worry with these tests, held up. The invasive procedure rate after a positive result was 0.6%, with no serious study-related adverse events. Participant anxiety rose after a positive result and returned to baseline by twelve months. NHS-Galleri also showed 21% fewer clinically detected cancers and 25% fewer emergency presentations in the intervention arm.

The Bad

The endpoint that mattered most did not land

So.. the headline result went the other way. NHS-Galleri, the only randomized trial of the three, missed its primary endpoint. Namely, there was no significant reduction in late-stage (stage III and IV) cancer in the intervention arm, with an incidence rate ratio of 1.03 (95% CI 0.92 to 1.14, p=0.6324), according to Swanton's presentation. The test found more cancer, and it shifted some of it earlier, with a 16% increase in stage I to II detection. But the hard endpoint was not met in the current follow-up window, and that endpoint was the key to make Galleri a recommended screening program.

Why a stage shift may not save lives

Robson's discussant talk made the point the field needs to grapple with, following these results. Catching cancer earlier does not automatically mean patients live longer. He walked through the biases that can make a screening test look effective without changing how long people live: lead-time bias, length bias, the Will Rogers effect, and overdiagnosis. Citing Feng and colleagues in JAMA, he showed that the link between reducing late-stage disease and reducing mortality varies a great deal by cancer type, strong for lung and ovary and weak or even inverse for others. Measured against the WHO screening criteria, several boxes remain unchecked, including an agreed policy on whom to treat, a defined rescreening interval, and an established cost-effectiveness case, with the test listed near $949 in the U.S. and higher price points elsewhere (Canada and Israel for example).

Low sensitivity at the earliest stages

The clearest weakness is low sensitivity at the earliest stages. Robson made this explicit in his appraisal, noting that sensitivity is lowest at stage I, which is exactly the stage where treatment is most often curative. The underlying reason is of course biological. A methylation test depends on how much tumor DNA reaches the blood, and early tumors shed very little, or at least too little to be reliably captured by current methods. The aggregate sensitivity of 30.7% in NHS-Galleri and 39.3% in PATHFINDER 2 hides this, because the average blends high-shedding cancers with low-shedding ones. Or in other words, the test is not great at detecting cancer at stages 0-2. The test does well on the high-mortality, high-shedding cancers and much worse elsewhere, which raises a real question about whether one test trying to cover every cancer is the right design. It may also question whether DNA methylation is a good fit for early detection altogether.

What's Next

Stop treating yield as the finish line

In my humble opinion, there are two directions worth pursuing. The first is to stop treating diagnostic yield as the standard. Robson is right that yield should not be the standard by which a screening test is judged, and the bar is moving toward proven mortality benefit or a properly validated alternative which for late stage diagnosis aren't clear yet. NHS-Galleri continues, and longer follow-up may yet read out differently, so for now, I would hold judgment rather than write the approach off.

The early-stage sensitivity problem

The second is to get serious about early-stage sensitivity, which is where any mortality benefit has to come from. A test that catches most cancers at stage I and II, where surgery and treatment can be curative, is the precondition for that benefit, and it is precisely where the DNA-based approach struggles. This is also where I think other modalities may have something to offer. At SpotitEarly we work on a Bio-AI Hybrid that reads volatile compounds in breath rather than DNA in blood, and the appeal of a breath-based signal is that it does not depend on tumor shedding, so strong early-stage sensitivity is at least biologically plausible. I want to be careful here, because we owe the same burden of proof to ourselves that we are asking of Galleri, and intended-use, population-scale evidence will ultimately decide which of the novel MCED approaches will prevail.

Where this leaves us

Grail has done the field a service, both by running the largest and hardest studies anyone has attempted and by being open about a primary endpoint that did not read out the way they hoped. Multi-cancer blood testing can find more cancer, including types we currently have no way to screen for, and it can do so safely. What it has not shown yet is that finding those cancers helps people live longer. That is the question the next few years have to answer as more MCED approaches will mature.