Among 100,000 U.S. adults age 50–79 receiving annual Galleri + usual screening for three years, how many fewer Stage IV cancers will be diagnosed within four years versus usual screening alone?

TL;DR

My forecast is 105 fewer people diagnosed at Stage IV per 100,000 U.S. adults over four years. The central 80% interval is about 15 to 195 fewer, with a 7% chance that the difference is negative. The randomized trial drives the estimate; older simulations get little weight because they projected much larger effects than the trial observed (ASCO 2026; Dai et al.).

Context

This quantity cannot be observed directly in the stipulated U.S. population. The closest experiment is NHS-Galleri, which randomized more than 142,000 English adults, used blood draws at baseline, year 1, and year 2, and counted cancers from the first appointment through up to 18 months after the third appointment. Its combined Stage III/IV primary endpoint failed, while its prespecified Stage-IV-only secondary endpoint favored Galleri (trial abstract, data current May 26 and published June 3, 2026; NHS results page, May 30, 2026).

The target here is narrower and harder. It asks about a current U.S. age-50–79 population, exactly four years, complete adherence, all cancers with a meaningful Stage IV category, and unique people rather than tumor records. I therefore estimate a U.S. usual-care baseline and multiply it by a transported causal relative reduction.

Evidence

The historical backbone is a SEER18 analysis of U.S. adults aged 50–79 diagnosed in 2006–2015. It estimated a crude annual Stage IV incidence of 196.6 per 100,000, from 1,078 invasive cancers of all stages per 100,000; Stage IV was 18% of diagnoses. The analysis used 18 registries covering about 28% of the U.S. population and AJCC sixth-edition stage, and was published in 2020 (Clarke et al.). (aacrjournals.org) A simple four-year multiplication gives 786 Stage IV tumor diagnoses per 100,000.

That 786 figure is too low for the target cohort because it preserves the older 2006–2015 age mix and does not age the cohort during follow-up. A separate SEER17 analysis contains 1,154,515 cancer cases aged 50–84 diagnosed in 2006–2010 and reports Stage IV counts by five-year age band (Chang et al., 2024). The 2024 American Community Survey counts 109,709,577 U.S. residents aged 50–79, with more weight in older bands than the 2010 population (Census S0101, 2024 vintage; 2010 Census age table). Reweighting the SEER Stage IV age gradient to that population adds about 10%; aging the closed cohort across four years adds about 9%; I then subtract 4% for competing mortality and conversion from tumor events to unique people. The resulting usual-care baseline is about 910 unique people per 100,000, with a plausible range of roughly 830–1,000. The person correction matters because public registry records can include multiple primary cancers and cannot identify linked multiple primaries in the public-use file (CDC variable definitions, June 1, 2026).

The randomized anchor is 342 Stage IV diagnoses in the Galleri arm versus 397 in control among the 12 prespecified cancers, with 71,122 and 71,128 randomized participants. The incidence-rate ratio was 0.86, with a 95% interval of 0.744–0.998. The raw absolute gap is 77.3 per 100,000 over the trial window. The same trial found 706 versus 688 combined Stage III/IV cancers, an incidence-rate ratio of 1.03 and p=0.6324, so the primary endpoint failed (ASCO abstract). Round-specific Stage IV reductions were 9%, 22%, and 26%, but only the last round excluded no effect.

The 14% figure should not be applied to every U.S. Stage IV cancer. The 12 selected cancers account for roughly four-fifths of the older U.S. Stage IV burden (site-specific SEER rates), but Galleri's episode sensitivity across 197,146 analyzable screening episodes was 54.7% for those 12 cancers and only 30.7% across all cancers (SEC-filed trial presentation, May 2026). This pulls the all-cancer effect down. Complete adherence pulls it back up, but only modestly: pooled visit retention was about 91.4% at year 1 and 88.6% at year 2, and the trial's retention targets themselves allowed for cancer diagnosis and death rather than treating every missing visit as avoidable nonadherence (EDCC retention poster, October 2025). I use a transported all-cancer relative reduction of 11.5%.

The central calculation is simple: 910 usual-care Stage IV diagnoses per 100,000 multiplied by an 11.5% causal reduction equals about 105 fewer. Pretrial models projected 37%–46% fewer Stage IV diagnoses after three annual rounds and 45% fewer over ten annual rounds (Dai et al., 2024; Chhatwal et al., 2025). Those models use case-control sensitivity and assumed natural histories. I use them only to shape the optimistic tail.

What's non-obvious

Two large corrections nearly cancel. Updating the old U.S. base rate for today's age mix and four years of cohort aging raises the baseline from about 786 to about 910. Moving from the favorable 12-cancer secondary endpoint to all stageable cancers, then shrinking for primary-endpoint failure and transport uncertainty, lowers the relative effect from the headline 14% to about 11.5%. The product stays near 105.

The 26% third-round result is not a four-year cumulative effect. It applies to one incident round, has a wide interval, and is followed by no year-3 Galleri test. Applying 26% to the whole four-year U.S. burden would produce about 235 fewer diagnoses and belongs in the upper tail, not at the center (round-specific trial results). (discovery.ucl.ac.uk)

Uncertainties

The main gaps are clear:

• The public randomized report gives exact Stage IV counts for the 12 cancers, but not a complete numerical all-stageable-cancer Stage IV table with confidence intervals (ASCO abstract). A full peer-reviewed paper would settle the cancer-scope adjustment.
• The randomized report speaks mainly in cancer diagnoses, while this question counts unique people (ASCO abstract). Patient-level deduplication data would settle the event-to-person correction.
• No current national table directly reports four-year AJCC Stage IV risk for unique U.S. adults initially aged 50–79; current USCS public documentation uses registry records and stage variables rather than this linked person-level estimand (CDC technical notes). A custom SEER or USCS extract through 2023 would sharpen the 910 baseline.
• The exact 48-month effect after three tests is unobserved. The NHS team says it will examine an extra 12 months of data; that follow-up and a U.S. randomized utility trial would close the horizon and transport gaps (NHS trial update, May 30, 2026).

I encode the remaining uncertainty as an 82% trial-anchored component centered at 105 with a standard deviation of 55, a 10% near-null/adverse component centered at 10 with a standard deviation of 60, and an 8% optimistic component centered at 250 with a standard deviation of 110. The mixture has mean 107, median 103, a 90% interval of −12 to 238, and a 6.7% probability of a negative effect.