Sign in Subscribe

The Blood Test That “Failed” — And Why It’s Still the Future of Cancer Detection

The NHS-Galleri trial missed its primary endpoint and erased 50% of GRAIL’s market cap. Here’s why the science underneath is more promising…

The NHS-Galleri trial missed its primary endpoint and erased 50% of GRAIL’s market cap. Here’s why the science underneath is more promising than the headlines suggest.

February 2026 · 14 min read

Here is the friend link

Here’s something worth sitting with before you dismiss GRAIL’s Galleri test.

On February 19, 2026, a stock ticker lost half its value in a single trading session. The cause? A randomized controlled trial of 142,000 people — the largest multi-cancer early detection study ever conducted — failed to hit its prespecified primary endpoint. The financial press called it a catastrophe. TD Cowen analysts flagged risks to FDA approval and Medicare coverage. Investors fled.

But here’s what those headlines buried: in the same trial, a single blood draw found stage IV cancer diagnoses dropping by more than 20% over three sequential screening rounds. For twelve of the deadliest cancers — pancreatic, ovarian, esophageal, liver, lung — the test caught them earlier than standard care alone ever would have.

Failing a statistical threshold and failing patients are two very different things. The NHS-Galleri trial is not the end of multi-cancer early detection. It may, paradoxically, be its most important proof of concept yet. And for those of us working in clinical genomics and oncology, the signal buried inside this “failure” tells a story that deserves a far more careful reading.

Why We Need This Test to Work — The Scale of the Problem

Before dissecting the trial, let’s be precise about what we’re trying to solve.

Of the more than 200 recognized cancer types, only four — breast, cervical, colorectal, and lung — have U.S. Preventive Services Task Force (USPSTF) recommended screening guidelines. That leaves roughly 70% of cancers with no approved population-level screening protocol. For cancers like pancreatic, ovarian, and liver, the default detection pathway is still symptomatic presentation — which typically means stage III or stage IV, where five-year survival rates collapse from above 90% to under 30%.

This is the screening gap that killed the majority of the 609,000 Americans who died from cancer in 2024 alone.

Conventional screening is also anatomically siloed by design. A colonoscopy tells you nothing about your esophagus. A mammogram says nothing about your pancreas. Building out individual tests for every cancer type is neither biologically feasible nor economically viable. What we need is a single, sensitive, reproducible liquid biopsy that can survey the entire oncological landscape from one blood draw.

That’s exactly what the Galleri test is trying to do. And the science underlying it is, frankly, extraordinary.

The Molecular Architecture: Why cfDNA Methylation Is the Right Signal

The Galleri test doesn’t look for cancer mutations. It reads the epigenome.

When cells die — including cancer cells — they shed tiny fragments of DNA into the bloodstream, called circulating cell-free DNA (cfDNA). In healthy individuals, this cfDNA originates mostly from blood and immune cells undergoing normal turnover. In individuals with cancer, tumor cells shed cfDNA that carries a distinctive molecular fingerprint: aberrant methylation patterns.

DNA methylation is a chemical modification — a methyl group attached to cytosine residues — that acts as a master regulator of gene expression. Healthy differentiated tissues have highly specific, stable methylation landscapes. Cancer cells disrupt this landscape in tissue-specific ways that are measurable and reproducible. Critically, these methylation patterns are not just cancer-specific; they’re tissue-of-origin specific, meaning a methylation signature derived from a pancreatic tumor looks different at the molecular level from one shed by a lung adenocarcinoma.

GRAIL’s approach — first validated in the Circulating Cell-Free Genome Atlas (CCGA) study across approximately 15,000 participants — uses whole-genome bisulfite sequencing to interrogate millions of methylation sites across the cfDNA. A deep learning classifier then does two things simultaneously: it distinguishes tumor-derived cfDNA from normal biological noise, and it predicts the cancer signal origin (CSO) — the organ or tissue most likely responsible for the signal — with approximately 90% accuracy.

The result is a test with >99% specificity, meaning fewer than 1 in 100 positive results is a false positive. For a population-screening tool, that’s a critical number. False positives in cancer screening cascade into expensive, anxiety-inducing diagnostic workups that erode patient trust and healthcare system resources.

“MCED tests fill an unmet need. We have all of these cancers that are causing cancer-related mortality every single year, but we don’t know about them until it’s often too late. This test gives us a way to potentially screen for those cancers, find it early and give our patients a much greater chance of being cured.” — Ali Nabavizadeh, MD, Radiation Oncologist, OHSU Knight Cancer Institute

Real-world data backs this up. A 2025 Nature Communications analysis of 111,080 Galleri tests found a 0.91% cancer signal detection rate consistent with clinical projections, with 87% CSO accuracy in confirmed cancer cases — spanning 32 distinct cancer types.

Anatomy of the NHS-Galleri Trial: What Actually Happened

The NHS-Galleri trial (NCT05611632) was the most ambitious prospective randomized controlled trial ever mounted for a multi-cancer screening technology. Understanding its design is essential to understanding the results.

The setup: 142,000 asymptomatic participants aged 50 to 77, enrolled across England’s National Health Service. Participants were randomized to either receive annual Galleri testing alongside standard NHS cancer screening, or standard care alone. The study ran over three years of active screening.

The primary endpoint: A statistically significant reduction in the combined incidence of stage III and stage IV cancer diagnoses in the intervention arm versus the control arm.

Why it missed: The topline results did not show a statistically significant reduction in the combined stage III-IV endpoint. GRAIL noted one contributing factor: a higher-than-anticipated incidence of stage III cancers in the trial overall. This is a meaningful confounder. Stage III cancers are biologically heterogeneous — some shed ctDNA robustly, others minimally — and their detection is inherently harder than stage IV where tumor burden is higher and cfDNA shedding more abundant. A skewed baseline distribution in stage III incidence directly dilutes the statistical power of the primary endpoint.

But this is where the story gets genuinely interesting.

The Signal That Wall Street Missed

Beneath the failed headline endpoint, the NHS-Galleri data contains findings that should be the actual focus of scientific attention.

Finding 1: Stage IV reduction in deadly cancers. For the pre-specified group of 12 deadly cancer types — anus, bladder, colorectal, esophagus, head and neck, liver/bile duct, lung, lymphoma, myeloma/plasma cell neoplasm, ovary, pancreas, and stomach — adding Galleri to standard care produced a clinically meaningful and substantial reduction in stage IV diagnoses. Stage IV diagnoses decreased with each year of sequential screening, reaching greater than 20% reduction in rounds two and three.

This is the finding that matters clinically. Stage IV cancer is where survival rates collapse. Shifting even a fraction of diagnoses from stage IV to stage I or II translates directly into lives saved. A 20%+ reduction in stage IV diagnoses across twelve of the most lethal cancer types is not a footnote — it’s the signal the whole field has been waiting for.

Finding 2: A four-fold increase in detection rate. Adding Galleri to standard-of-care screening for breast, colorectal, cervical, and high-risk lung cancers resulted in a four-fold improvement in overall cancer detection rate compared to standard screening alone. This means the combined approach caught vastly more cancers, and caught them earlier.

Finding 3: Fewer emergency cancer presentations. The trial reported fewer cancers detected via emergency presentation in the Galleri arm. This is a profoundly important clinical and health economics outcome. Emergency-presentation cancers carry the worst prognoses and the highest NHS resource utilization. Every cancer shifted away from emergency detection represents both a survival gain and a cost saving.

Finding 4: Cumulative screening effect. The fact that the stage IV reduction grew with each screening round — from round one through round three — is one of the most biologically coherent signals in the entire dataset. It suggests the test’s population-level impact compounds over time as the algorithm’s learning improves and as the pre-clinical detection window is exploited more fully. This has direct implications for trial design: a three-year study may simply not be long enough to fully capture the mortality benefit of annual screening.

GRAIL has already announced an extension of trial follow-up by 6 to 12 months. Detailed results will be presented at the ASCO 2026 Annual Meeting. Full data from the 35,000-participant PATHFINDER 2 study will follow.

The Trial Design Problem: Are We Asking the Right Question?

This is the conversation that needs to happen in clinical genomics circles.

The NHS-Galleri trial was designed around a stage-shift endpoint — a reduction in late-stage diagnoses within three years. This is a reasonable proxy for mortality benefit. But it may not be the right primary endpoint for a technology that works through cumulative detection over time.

Consider the analogy of colorectal cancer screening. It took more than a decade of colonoscopy screening data to establish a statistically robust mortality benefit at population level. The biology of cancer evolution, the variability in tumor shedding rates, the learning curve for physicians interpreting positive MCED results — all of these factors mean that three years of screening may represent the early plateau, not the full curve.

There is a broader evaluation framework challenge that the entire MCED field now confronts. Unlike single-cancer screening tests (PSA for prostate, CA-125 for ovarian), MCED tests operate across 50+ cancer types simultaneously. The statistical power required to show stage-shift significance across such heterogeneous biology — in a real-world population where some cancers will be stage IV regardless of screening due to aggressive biology — is immense. The trial may have been underpowered not for lack of effect but for lack of time.

A 2025 review in PMC noted that MCED evaluation frameworks explicitly differ from single-cancer screening and that clinical utility — including downstream workup burden, overdiagnosis risk, and patient-centered outcomes — requires longitudinal assessment that three-year trials cannot fully capture.

The time-to-diagnostic-resolution finding is also worth noting. GRAIL reported that the time required to reach a diagnostic conclusion after a positive Galleri result has been improving over time as clinicians gain experience with Galleri-directed workups. This is a physician learning curve, not a test performance failure. As MCED testing becomes embedded in clinical practice, diagnostic pathways will become more efficient, further improving the population-level benefit.

What Remains Standing: The FDA, the Market, and the Science

Despite the trial miss, several structural realities keep Galleri’s trajectory viable.

The FDA pathway is active. GRAIL completed its final modular Premarket Approval (PMA) submission to the FDA in February 2026 — earlier in the same month the trial results were announced. Critically, the PMA application incorporates data on test performance, clinical validation, and specifically the benefit of detection at stages I through III, including reduction in stage IV diagnoses — exactly what the NHS-Galleri secondary endpoint data supports. The PMA does not depend solely on the failed primary endpoint.

Commercial momentum is real. GRAIL sold more than 185,000 Galleri tests during fiscal year 2025. U.S. Galleri revenue grew 26% year-over-year to $136.8 million. New distribution partnerships with digital health platforms like Hims & Hers have expanded access. The test is already reaching patients.

The balance sheet provides a buffer. Post-crash, GRAIL held approximately $904 million in cash and equivalents against $54.9 million in lease liabilities. With approximately $299 million used in operating cash per year, the company has a multi-year runway to see the ASCO data, extend the NHS trial follow-up, and pursue further regulatory dialogue.

The 35,000-participant PATHFINDER 2 trial results are forthcoming. This U.S.-based study, with its own distinct design and population, will provide an independent evidence dataset for the FDA and CMS to evaluate. The REACH study — a 50,000-participant Medicare-backed trial at OHSU — is actively enrolling and could provide the Medicare coverage evidence that insurance payers require.

The competitive landscape is maturing, not retreating. Grail is not alone in this field. Competitors including Exact Sciences, Guardant Health, Foundation Medicine, and international players like Singlera Genomics with PanSeer are all advancing MCED approaches. The NHS-Galleri result sets a higher bar for trial design and endpoint selection — but it does not invalidate the underlying biology. If anything, it accelerates the field’s sophistication around what constitutes a valid proof of clinical utility for population MCED screening.

The Deeper Lessons for the Field

The NHS-Galleri result is a lesson not in the failure of the technology, but in the ambition of the question we asked of it.

We designed a three-year trial and asked it to produce mortality-level evidence for a technology that requires a decade of patient follow-up and physician learning to fully express its benefit. We defined success as a statistically significant reduction in a combined stage III-IV endpoint, when the most clinically meaningful data — stage IV reduction in the twelve deadliest cancers — sits just outside that composite definition. We launched a trial in a healthcare system where standard-of-care screening itself is catching a non-trivial proportion of the very cancers we hoped to intercept earlier.

None of this diminishes the scientific achievement. The NHS-Galleri trial generated the world’s richest dataset in MCED research: 142,000 participants, three years of annual screening, complete linkage to NHS health records, and secondary endpoints showing a biologically coherent, compounding benefit in exactly the cancers where early detection matters most.

“The number and distribution of cancer stages across screening rounds suggests the potential for a stronger effect with longer follow up as data matures.” — GRAIL Press Release, February 19, 2026

The oncology field spent decades learning how to design trials for immunotherapy, targeted therapy, and precision medicine. MCED is a paradigm shift of comparable magnitude — and it deserves trial designs that match the biological timescale of the benefit it delivers.

The signal in this data is not noise. It’s the early measure of a technology that, given time, rigorous follow-up, and clinician education, has the potential to fundamentally change the trajectory of cancer mortality at population scale.

Your Action Items — Start Here

Immediate (< 5 minutes)

  • [ ] Read the full GRAIL NHS-Galleri press release for the primary data
  • [ ] Register for ASCO 2026 session alerts — the full NHS-Galleri data will be presented mid-2026

This week

  • [ ] Review the Nature Communications real-world performance analysis (111,080 tests) for context on operational performance
  • [ ] Read the PMC MCED review for a framework-level understanding of MCED evaluation challenges
  • [ ] If you practice in oncology or primary care, review the PATHFINDER 2 enrollment criteria — understanding how to counsel patients about MCED testing is now a clinical competency

Ongoing

  • [ ] Monitor GRAIL’s FDA PMA timeline — a decision here will have broader implications for CMS coverage
  • [ ] Track the REACH study (50,000 Medicare beneficiaries, OHSU-led) as the next major evidence readout
  • [ ] Watch the competitive landscape: Guardant, Exact Sciences, and Foundation Medicine will all respond to this result with their own trial designs

Key Resources

Official Sources

Essential Reading

Research Papers

Before You Go

The NHS-Galleri result is a setback for a stock, not for a science. The biology of methylation-based cancer detection is sound. The signal is real. The question is whether we are designing trials ambitious enough to measure an effect that takes more than three years to fully manifest.

A few questions worth discussing:

  • Should randomized controlled trials of MCED screening adopt mortality as a co-primary endpoint from the start, even if it requires 8–10 year follow-up? What would the ethical implications be of withholding screening from a control arm that long?
  • As cfDNA shedding rates vary profoundly by cancer type and stage, should future MCED trials be stratified by cancer type for primary endpoint analysis rather than using a composite stage III-IV metric?
  • What does a physician education and clinical decision support infrastructure for MCED-positive workups need to look like — and who builds it?

If this was useful, please:

👏 Clap — it directly helps this reach oncologists, clinical genomicists, and healthcare leaders who need to think carefully about these results (up to 50 claps)
 💬 Comment with your own clinical or research perspective on MCED trial design — the most nuanced takes deserve a wider audience
 🔔 Follow for the next piece — I’m writing about the computational biology of cancer signal origin prediction and where cfDNA methylation classifiers go next

Early cancer detection is moving faster than most clinicians can track. Understanding the difference between statistical endpoints and clinical signal is the difference between dismissing a breakthrough and recognizing one.

Data and topline results cited from GRAIL’s February 19, 2026 press release. Secondary analyses pending full ASCO 2026 presentation. The full NHS-Galleri dataset will be the subject of peer-reviewed publication — watch for it.

Read next

Comments ()