Galleri's randomized endpoint came back at 1.03 — and an FDA advisory committee is reported for September

Randomize roughly 140,000 consenting adults 1:1 to NHS care with or without a methylation-based blood test, screen them annually for three years, and the combined stage III/IV cancer incidence across the 12 pre-specified deadly cancers comes back three percent higher in the screened arm than in control: incidence rate ratio 1.03, 95% CI 0.92–1.14, p=0.6234. That number sits in GRAIL's own analyst deck, filed as EX-99.1 to an 8-K on 1 June 2026 ^[2]. Endpoints News reported on 7 August 2026 that the FDA will convene an advisory committee in September 2026 to review Galleri ^[3]. That single trade report is the only source for the meeting — treat it as trade-reported until FDA posts a notice.

What the trial measured, and what it did not

NHS-Galleri is registered as a randomized device study with one primary outcome: the incidence of stage III and IV cancers in the intervention arm versus control. GRAIL is the sponsor, registered enrollment is 142,318, the study began on 31 August 2021 and the primary completion date was 12 June 2026 ^[1]. GRAIL's deck states that endpoint more narrowly than the registry field does — a reduction in combined stage III and stage IV cancers across 12 pre-specified deadly cancers (lung, head and neck, colorectal, pancreas, myeloma/plasma cell neoplasm, liver/bile duct, stomach, esophagus, anus, lymphoma, ovary and bladder), where the registry simply reads stage III and IV cancers, intervention versus control ^[1][2]. The registry describes approximately 140,000 participants actively followed for about three years, testing whether Galleri used alongside standard NHS testing finds cancers at an early stage ^[1]. GRAIL's deck reports 142,826 consented and arms of 71,122 versus 71,128 — which sum to 142,250, not 142,826, and neither figure matches the registry's 142,318. That is reason enough to wait for the peer-reviewed report before quoting a single n ^[1][2].

The endpoint is a stage shift, not a death. A screening test ultimately earns its place by moving mortality; this trial was built on a surrogate one step short of that, and the surrogate came back flat. Prognyx found no mortality result — cancer-specific or all-cause — in the registry record or in the filed deck ^[1][2]. GRAIL describes additional follow-up as ongoing and years 4+ as planned ^[2].

GRAIL is leading with the 20%. Read the intervals instead

The company's framing is that the miss is an artifact of the first screening round. Its deck states that an increase in stage III cancers impacted the primary endpoint, and that the relative incidence of stage III/IV cancers decreased after the first round ^[2]. The round-level table, as Prognyx reads the column mapping off a flattened slide, runs 1.19 (0.98–1.43) in the first round, 0.95 (0.77–1.17) in the second and 0.88 (0.73–1.07) in the third round plus follow-up — an assignment that is an inference and should be checked against the primary publication before anyone builds on it ^[2].

Now look at the intervals rather than the point estimates. Every one of the three includes 1.0. The directional story — a prevalence round drags stage III disease forward, then incident rounds start catching cancer earlier — is biologically coherent and statistically unproven in the data as filed.

The secondary results carry GRAIL's weight: a 20% reduction in stage IV cancers across the 12 pre-specified deadly cancers in incident rounds, and roughly 16% more stage I/II cancers detected ^[2]. Two cautions travel with that. A slide header in the same deck states a 22.0% reduction in incident rounds — the deck disagrees with itself ^[2]. And the filed deck states no confidence interval and no p-value for the stage IV secondary ^[2]. A secondary with an unstated interval, in a trial whose primary missed, is a hypothesis with good manners.

Test performance is the least disputed part: PPV 52%, false positive rate 0.45%, episode sensitivity 31% across all cancers and 55% across the 12 deadly cancers, cancer-signal-of-origin accuracy 92.5%, and no serious related adverse events ^[2]. Sit with the 31%. Two of every three cancers captured by that episode-sensitivity denominator were never signalled — and neither the registry record nor the filed deck puts a figure on what those misses cost ^[1][2].

Who is exposed — and only one listing in the set states the multi-cancer claim

Registry listings state what a company is testing, never how well it works — and in the set Prognyx reviewed, most of them do not state a multi-cancer claim or an asymptomatic population at all. One listing says multi-cancer early detection on its face: Shanghai Weihe Medical Laboratory's PROFOUND (NCT06217900, recruiting), a prospective case-control study to develop and validate a blood test for multi-cancer early detection ^[16]. The rest of the blood-based early-detection group is registered in narrower or unspecified terms — Harbinger Health (NCT07046260, recruiting, a prospective study to assess a diagnostic aid for cancer) ^[11], Adela (NCT05366881, recruiting, prospective observational validation of a cfDNA assay for early cancer detection and minimal residual disease) ^[12], Universal Diagnostics (NCT06059963, recruiting, a prospective multi-centre observational evaluation of the Signal-C test) ^[13], and Helio Genomics (NCT05199259 recruiting, NCT03694600 active but no longer recruiting, a multi-analyte blood test) ^[14][15]. Multi-analyte is not multi-cancer, and none of those titles names the population screened; that silence is the point, not a gap in the search. Three of the five listings state a design — Adela and Universal Diagnostics register observational studies, Shanghai Weihe a case-control study ^[12][13][16] — while Harbinger's and Helio's state none, so their design is not established here ^[11][14][15]. What a September panel would test, if it happens, is not their dossiers but GRAIL's: the registrational label in GRAIL's own filing sits on PATHFINDER 2, a study evaluating a single MCED test in roughly 35,000 asymptomatic individuals aged 50 and over, rather than on the randomized trial that missed ^[2].

A second group sits in cell-free DNA but not in this question. Guardant Health's SIBYL (NCT05935384, recruiting) is registered as observation of therapy response with liquid biopsy — monitoring, not screening ^[17]. Sequenom (NCT02586389, recruiting) is collecting specimens for liquid-biopsy assay development ^[18]. Suzhou Huhu Health & Technology runs two gastric-cancer studies, on early detection with post-operative monitoring and on post-operative monitoring alone (NCT06232395, NCT07026240, both recruiting) ^[19][20], and JaxBio is testing epigenetic mapping in cfDNA for lung cancer alone (NCT06963307, recruiting) ^[21]. Single-cancer and monitoring claims rest on a different evidence base; a bad day in September costs them less — on these listings alone, and a sponsor's registry footprint is not its pipeline. Two more sit further out on modality: DL Analytics (NCT06815939, recruiting, a lab-free low-cost cervical screening test) ^[23] and an investigator-sponsored platelet- and immune-cell transcriptomic program (NCT06717295, recruiting) ^[24]. The set behind this section is blood- and cell-free-DNA cancer detection trials on the public registry, industry-sponsored except where noted; nothing in it is drawn from company disclosures.

Behind GRAIL, the academic pipeline is thin and slow. Dana-Farber runs Sentinel in a high-risk military and veteran population (NCT06523868, recruiting, n=1,500, primary completion 30 September 2027, endpoints PPV, NPV, specificity, yield and number needed to screen at 12 months) ^[6], INFORM in a cancer-predisposition population (NCT06450171, recruiting, n=1,000, primary completion 31 January 2027, cancer detection rate) ^[7], and DISCOVER (NCT07390643, n=100, listed to start in July 2026 but still shown as not open in the registry record, primary completion 31 January 2027) ^[8]. Every endpoint there is accuracy, yield or detection characterisation — none is clinical utility. None of the studies Prognyx reviewed is a second randomized clinical-utility trial.

What Prognyx could not verify

This briefing rests on the ClinicalTrials.gov record for NCT05611632 dated 10 July 2026, GRAIL's SEC filings of 1 June 2026, 12 November 2025 and 16 October 2025, the registry listings named above, and one trade report of 7 August 2026 ^{[1][2][3][4][5]}. The September meeting rests on that trade report ^[3]: Prognyx found no FDA notice, committee name, meeting date or briefing-document date among those records, and has not independently searched FDA's advisory-committee calendar or the Federal Register. That gives the claim a dated test: a meeting held in September 2026 would leave a public paper trail, so watch for a notice, a named committee and a briefing package through the rest of August and September, and if none has surfaced by 30 September 2026, read the report as unconfirmed rather than early. Whether the Galleri PMA was actually submitted and accepted for filing is unestablished — the only sourced statement is GRAIL's 12 November 2025 guidance that submission was anticipated in the first quarter of 2026 ^[4] — and whether that PMA rests on PATHFINDER 2, the study in roughly 35,000 asymptomatic individuals aged 50 and over that GRAIL labels registrational ^[2], with NHS-Galleri as supportive, is not confirmed. That question is the crux of the whole evidentiary argument. No published target action date or review clock exists for a PMA whose submission is itself unconfirmed, so no decision date can be derived: the reported September 2026 panel is the only datable regulatory event here, and it is trade-reported. No FDA statement, no CMS coverage decision, no USPSTF position and no NHS statement on a national rollout appears anywhere in these records. The registry entry for NCT05611632, dated 10 July 2026, still shows the trial active but no longer recruiting, with no reason for stopping recorded and no results posted, even though GRAIL's full NHS-Galleri results were before the SEC on 1 June 2026 ^[1][2] — do not read that empty record as a pending readout. Finally, a commentary titled "Outcomes from the NHS-Galleri trial: the good, the bad, and the uncertain" appeared on 3 August 2026 ^[9] and a JAMA analysis of diagnostic-delay spillover from the screening trial on 1 July 2026 ^[10]; Prognyx has titles and objectives only and does not characterise either conclusion.

What to watch, with dates

September 2026 — the reported advisory committee ^[3]. If the meeting is noticed, the document to read is FDA's briefing package published ahead of it: that is where the agency would set out its own read of the evidence, and any objection, and it would do more to set the field's bar than the vote. The peer-reviewed NHS-Galleri publication, for which GRAIL's filings give no date, is the document that would resolve the 20% versus 22.0% stage IV discrepancy, supply the missing interval and settle the round-level mapping ^[2]. Results posting on NCT05611632 — the 10 July 2026 registry record still carries no results, six weeks after GRAIL put the full NHS-Galleri numbers before the SEC on 1 June 2026 and a month after the 12 June 2026 primary completion date ^[1][2]. Prognyx has not established what results-posting obligation, if any, applies to this trial; the observation that stands is that the registry sits behind the filing. GRAIL's next quarterly filing — PMA submission and acceptance, plus whether the Samsung transaction closed: the 16 October 2025 binding letter of intent put $110M of equity at $70.05 per share against exclusive Korean commercialization, with definitive agreements and closing intended in early 2026, and Prognyx found no confirming filing after that date ^[5]. 31 January 2027 — INFORM and DISCOVER primary completion ^[7][8]. 30 September 2027 — Sentinel ^[6].

The now-what

Re-cut your endpoint before a panel does it for you. NHS-Galleri is the field's warning that a combined stage III+IV endpoint may be dominated by what a prevalence round pulls forward — a hypothesis the filed intervals cannot yet settle. If your registrational design still uses it, the weeks before September are the last cheap moment to move to a late-stage-specific or presentation-based measure — GRAIL's own exploratory readouts point that way, with a reported 21% reduction in clinical presentation and 25% reduction in emergency presentation ^[2]. Assume a panel would be unmoved by round-level splitting, whose pre-specification is not stated in the filed deck.

Instrument the harm side now. What a panel and a payer weigh is not sensitivity. It is PPV 52% and a 0.45% false positive rate in NHS-Galleri ^[2], and in PATHFINDER 2 — the study in roughly 35,000 asymptomatic individuals aged 50 and over that GRAIL labels registrational ^[2] — a pre-specified analysis of the first roughly 25,000 participants presented at ESMO 2025 ^[2][4]: PPV 61.6%, specificity 99.6%, cancer-signal-of-origin accuracy 92%, median diagnostic resolution of 46 days and 0.6% of all participants undergoing an invasive procedure ^[4]. If you cannot produce those prospectively, you will be arguing against someone who can.

Hedge the claim, not just the trial. Population screening is the longest evidence path in diagnostics. The referred symptomatic population is shorter and higher-prevalence: SYMPLIFY reported a Galleri PPV of 75.5% in its primary Lancet Oncology analysis, and an extended registry follow-up presented in October 2025 found roughly a third of apparent false positives were later diagnosed with cancer, lifting PPV in that population to 84.2% ^[4]; a further exploratory retrospective analysis published on 28 May 2026 examines how an MCED test could contribute to faster diagnosis there if acted upon ^[22]. A narrower first claim buys a label while screening evidence matures. And note the money: Q3 2025 US Galleri revenue was $32.6M, up 28%, while total Galleri tests sold exceeded 45,000, up 39% ^[4]. Against that sat an $89.0M net loss and a $13.7M gross loss, with $547.1M in cash, equivalents, restricted cash and short-term securities at 30 September 2025 and a headline position above $850M once the roughly $325M October private placement is counted ^[4]. As of that last reported quarter, ended 30 September 2025, the test was selling without FDA approval ^[4]. September decides the claim, not the cash flow.

Verdict — Prognyx's read, with no FDA, CMS or USPSTF position on MCED in the record: field-defining for the evidentiary bar. The window to shape how that bar is framed closes the day FDA posts a briefing document — if the reported September 2026 meeting is confirmed and held — because after that you are arguing against a written position.