The Galleri Blood Test: What It Measures and What the Numbers Support

Oak
mced liquid-biopsy cfdna screening methylation
A tall stilt-legged creature filters glowing pale flecks from a dark river at night, its translucent throat sac lit from within.

Galleri is a $949 blood test that sequences cell-free DNA methylation patterns from a single ~10 mL blood draw and returns a binary call: cancer signal detected or not detected. If detected, it also returns one or two predicted tissues of origin. It is a laboratory-developed test, not FDA-approved, and it requires a clinician order. It is not a substitute for colonoscopy, mammography, cervical screening, or low-dose CT, and its own large randomized trial in the NHS failed to hit its primary endpoint. Whether it is worth $949 to you depends almost entirely on your pre-test probability of harboring an undiagnosed cancer, and for most people under 50 that number makes the positive predictive value uncomfortably low.

How the assay works

Galleri targets methylation, not mutations. Roughly 5–10 ng of cell-free DNA is extracted from plasma, bisulfite-converted (or enzymatically converted), and sequenced across a targeted panel covering on the order of a million CpG sites in about 100,000 genomic regions selected for differential methylation between tumor and leukocyte DNA. A classifier trained on the Circulating Cell-free Genome Atlas cohort scores the fragment-level methylation patterns and emits both a detection call and a cancer signal origin prediction 1.

The choice of methylation over mutation calling is deliberate and, we think, correct. Somatic mutation panels on cfDNA run into clonal hematopoiesis: age-related mutant clones in white blood cells shed DNA carrying DNMT3A, TET2, and ASXL1 variants at allele fractions indistinguishable from a small tumor. Methylation signatures are tissue-specific in a way point mutations are not, so a hepatocyte-derived fragment looks like a hepatocyte-derived fragment regardless of who mutated what. That is also what makes tissue-of-origin prediction possible at all 1.

The tradeoff is that the classifier is a black box trained on case-control data, and its behavior on the screening population (asymptomatic, mostly cancer-free, enriched for benign inflammation and other tissue turnover) is not fully characterized by the training set.

What the performance numbers say

The headline figures from the CCGA substudy: specificity 99.5%, overall sensitivity 51.5% across more than 50 cancer types, and tissue-of-origin accuracy near 89% among true positives. Stage-stratified sensitivity is the number that matters, and it is steep: roughly 17% at stage I, 40% at stage II, 77% at stage III, and 90% at stage IV 2.

Read that gradient carefully. The screening value of a test is concentrated in stage I and II detection, because stage III and IV cancers usually announce themselves. A test that catches 17% of stage I disease is not catching cancers early in the sense that changes outcomes for most people who get one. This is the central critique of the whole MCED category, and it is a statistical argument, not a mechanistic one 3.

PATHFINDER, the prospective study in about 6,600 adults over 50, gave the field its first look at real-world behavior. A cancer signal was detected in roughly 1.4% of participants. Fewer than half of those turned out to have cancer. Median time to diagnostic resolution was around 79 days overall, and substantially longer for the people who did not have cancer — several months of imaging, endoscopy, and biopsy that ended in nothing 2.

Run the PPV math for yourself

This is the calculation we would want any technically literate reader to do before ordering.

Let p be your annual probability of having a currently-undiagnosed, detectable cancer. Sensitivity s = 0.51, specificity = 0.995.

PPV = p·s / (p·s + (1-p)·0.005)

At age 60–70, p is roughly 0.01. PPV ≈ 0.005 / (0.005 + 0.00495) ≈ 50%. A coin flip.

At age 45, p is closer to 0.002. PPV ≈ 0.00102 / (0.00102 + 0.00499) ≈ 17%. Five out of six positives are false.

At age 35, p is maybe 0.0005. PPV ≈ 5%. Nineteen out of twenty positives are false.

Now flip it. The negative predictive value is above 99% at every one of those ages, which sounds reassuring and is nearly meaningless: it was already above 99% before the blood draw. A “not detected” result moves your posterior by a fraction of a percent, because the test misses half of prevalent cancers and the great majority of stage I disease. Treating a negative Galleri as a reason to skip a colonoscopy is the single worst way to use this test 4.

The NHS-Galleri result

The NHS-Galleri trial randomized roughly 140,000 people in England with a primary endpoint of reducing the absolute incidence of stage III and IV cancers at diagnosis. That endpoint is the right one: it asks whether the test moves cancers from late stage to early stage in a population, which is the mechanism by which screening saves lives. The trial did not meet it, and the NHS declined to proceed to broader rollout 5.

This does not mean the assay detects nothing. It means the stage shift it produces in an unselected screening population was too small to clear the threshold that would justify national deployment. Critics had predicted roughly this outcome from the stage-specific sensitivity numbers years earlier 36. Screening tests generate three failure modes that a diagnostic test does not: overdiagnosis of indolent disease, length-time bias making detected cancers look more survivable than they are, and a workup cascade with its own morbidity and cost. Population trials with hard endpoints exist to separate real benefit from those artifacts 4.

What you get back, and what you do not

You get a PDF. Cancer signal detected or not detected, and if detected, up to two predicted signal origins ranked by probability.

You do not get FASTQs, BAMs, per-region beta values, the classifier score, or a calibrated probability. There is no threshold you can adjust, no way to see whether you landed at 0.51 or 0.99 on the decision boundary, and no way to compare this year’s methylation profile to last year’s at the feature level. For a reader who wants to model their own biology over time, this is the structural limitation: Galleri is an oracle, not a data source. Longitudinal cfDNA is where the interesting signal probably lives, since a rising trajectory in a tumor-fraction estimate is far more informative than one thresholded call, but the product does not expose the trajectory 7.

What it misses

Galleri is not organ-indexed, so “which cancers does it not test for” is the wrong shape of question. It detects a shared methylation signal, and detection depends on how much tumor DNA reaches plasma. Cancers with low cfDNA shed perform poorly: early prostate, thyroid, renal cell, many early breast cancers, and central nervous system tumors sequestered behind the blood-brain barrier. Cancers with high shed and poor existing screening (pancreatic, ovarian, esophageal, liver, hematologic malignancies) are where it performs best, and that is a real argument in its favor since no standard screening exists for most of them 89.

It also does not replace stool-based or endoscopic colorectal screening. Blood-based colorectal detection at stage I remains substantially worse than FIT or colonoscopy, and the literature on colorectal biomarkers is explicit about this gap 10.

How we would think about ordering it

If you are over 50, up to date on every guideline-recommended screen, and you can absorb the cost and the possibility of a months-long negative workup without distress, Galleri is a defensible add-on. Talk to a physician first, not after, because a positive result puts you immediately into a diagnostic pathway that requires clinical management and you want that relationship established.

If you are under 45 and healthy, we would skip it. The PPV math is unfavorable, the workup risk is real, and the money buys more information deployed elsewhere.

Questions people also ask

Will insurance pay for it? Generally no. Galleri is a laboratory-developed test without FDA approval, and Medicare does not cover MCED tests absent specific legislation. Some employers and concierge practices cover it, and HSA/FSA funds are often eligible. List price is $949.

Which is better, Cancerguard or Galleri? There is no head-to-head trial, so the comparison is between published operating characteristics on different cohorts. Exact Sciences’ Cancerguard combines protein biomarkers with methylation and reports lower specificity than Galleri. At screening prevalence, a specificity drop from 99.5% to 98.5% triples the false-positive rate, which dominates PPV far more than a few points of sensitivity. Neither has a completed randomized trial showing mortality or stage-shift benefit.

How accurate is it? Specificity around 99.5% and overall sensitivity around 51%, but sensitivity is roughly 17% at stage I 2. Accuracy as a single number hides the part that matters.

What are the risks? False positives leading to imaging, biopsy, and incidental findings, with median resolution taking months. False reassurance from a negative result causing someone to skip established screening. Overdiagnosis of cancers that would never have caused harm 4.

What is the “Holy Grail” of cancer detection? A single cheap test with high stage I sensitivity, high specificity, accurate tissue localization, and demonstrated mortality reduction in a randomized trial. No test currently satisfies the last two conditions simultaneously 6.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Megan P. Hall, Alexander M. Aravanis. The Galleri Assay. Current Cancer Research, 2023. https://doi.org/10.1007/978-3-031-22903-9_25 ↩ ↩2

  2. Cody E. Cotner, Elizabeth O’Donnell. Understanding the Landscape of Multi-Cancer Detection Tests: The Current Data and Clinical Considerations. Life, 2024. https://doi.org/10.3390/life14070896 ↩ ↩2 ↩3

  3. Robert Old, Paul Pharoah, Nicholas Wald. NHS announces a pilot of a blood test for early detection of many cancers. Journal of Medical Screening, 2021. https://doi.org/10.1177/0969141320986823 ↩ ↩2

  4. Caroline M Snead, Dominick Zheng, Quyen Ngo-Metzger, et al. Multicancer Detection Assays: Promise and Potential Harms of a Novel Cancer Screening Tool. The Permanente Journal, 2026. https://doi.org/10.7812/tpp/25.075 ↩ ↩2 ↩3

  5. Eleftherios P. Diamandis. Failure of the Galleri multi-cancer detection trial to meet its primary endpoint. Diagnosis, 2026. https://doi.org/10.1515/dx-2026-0055 ↩

  6. Miyo K. Chatanaka, George M. Yousef, Eleftherios P. Diamandis. The Unholy Grail of cancer screening: or is it just about the Benjamins?. Clinical Chemistry and Laboratory Medicine (CCLM), 2024. https://doi.org/10.1515/cclm-2024-1013 ↩ ↩2

  7. Seung Il Kim, Young Kim. Clinical Applications of Liquid Biopsy. Liquid Biopsy for Cancer, 2026. https://doi.org/10.1007/978-981-99-4418-7_5 ↩

  8. Mitsuho Imai, Yoshiaki Nakamura, Takayuki Yoshino. Transforming cancer screening: the potential of multi-cancer early detection (MCED) technologies. International Journal of Clinical Oncology, 2025. https://doi.org/10.1007/s10147-025-02694-5 ↩

  9. Emma Di Carlo. Early Cancer Detection: What’s Going on and What’s Next. MedComm, 2026. https://doi.org/10.1002/mco2.70653 ↩

  10. Alessandro Mannucci, Ajay Goel. Stool and blood biomarkers for colorectal cancer management: an update on screening and disease monitoring. Molecular Cancer, 2024. https://doi.org/10.1186/s12943-024-02174-w ↩