Screening that detects cancers earlier can improve outcomes by increasing opportunities for curative treatment and extending survival. However, guideline-recommended population screening currently covers only a limited set of cancers (for example, breast, cervical, colorectal and lung in high‑risk groups), leaving most cancer types unscreened. Implementing multiple separate single‑cancer screening programs at population scale is constrained by low prevalence of many cancers, cumulative false positives, cost and logistics.
Blood‑based multi‑cancer early detection (MCED) tests aim to address this gap by screening for signals shared across multiple tumor types in a single assay while maintaining a low false‑positive rate. One such test, Galleri (GRAIL Inc.), detects tumor‑derived methylation patterns in cell‑free DNA and, when positive, returns a predicted cancer signal origin (CSO) to guide downstream diagnostic evaluation.
The NHS‑Galleri trial was a randomized controlled study that enrolled adults aged 50–77 years (N = 142,250) who were randomized 1:1 to either an intervention arm offered the MCED test or a control arm receiving usual care. The trial’s primary endpoint — a reduction in the incidence of stage III/IV cancer diagnoses in the intervention versus control arm — was reported elsewhere and was not met. The analyses described here are prespecified secondary performance endpoints among evaluable participants in the intervention arm across three annual screening rounds. These analyses are descriptive; no hypothesis testing was performed.
Intervention‑arm participants who had a positive Galleri result were referred into National Health Service (NHS) standard‑of‑care diagnostic pathways. Those referrals were informed by the test’s predicted CSO(s) to guide clinical evaluation and imaging. The diagnostic workups and any subsequent cancer diagnoses therefore occurred within established NHS processes rather than per a bespoke trial algorithm.
Across the three annual screening rounds, the proportions of participants with positive test results were: round 1, 722 of 70,325 (1.03%); round 2, 518 of 64,498 (0.80%); and round 3, 561 of 62,323 (0.90%).
In aggregate, 937 participants had MCED‑detected primary cancers across the rounds. The by‑round cancer detection rates were 0.60% in round 1, 0.40% in round 2 and 0.41% in round 3. These per‑round detection rates reflect the fraction of screened participants in whom a primary cancer was identified following a positive test and subsequent NHS diagnostic evaluation.
Reported test performance metrics across rounds included:
Positive predictive value (PPV): 58.0% (419/722) in round 1, 50.4% (261/518) in round 2 and 45.8% (257/561) in round 3.
Negative predictive value (NPV): approximately 98.9% across rounds — specifically 98.98% (68,895/69,603) in round 1, 98.90% (63,278/63,980) in round 2 and 98.86% (61,058/61,762) in round 3.
Specificity across rounds ranged from 99.50% to 99.60%.
Episode sensitivity (detection of cancers across the screening episode) for all cancers ranged from 26.7% to 37.2% across rounds. For 12 prespecified cancer types, episode sensitivity ranged from 47.6% to 63.4%.
Predicted CSO accuracy, which reflects how often the top predicted tissue or organ of origin matched the diagnosed cancer site, ranged from 91.1% to 93.6% across rounds.
These metrics were calculated for evaluable participants within the intervention arm and are described without formal hypothesis testing.
The NHS‑Galleri prespecified secondary analyses provide detailed, real‑world performance measures of the Galleri MCED test when used within NHS diagnostic pathways. The test demonstrated high specificity (~99.5–99.6%), high CSO prediction accuracy (>91%), and per‑round PPVs ranging from about 46% to 58%. Episode sensitivity for all cancers was lower (approximately 27–37%) but higher for a set of 12 prespecified cancer types (approximately 48–63%).
These results highlight a trade‑off typical of screening: the test detected a subset of cancers with relatively high PPV and very low false‑positive rates, yet it did not identify the majority of cancers (moderate sensitivity across all cancers). The study’s analyses were descriptive and limited to intervention‑arm evaluable participants; no hypothesis testing or primary‑endpoint inference is reported here. Diagnostic assessments followed routine NHS pathways and were informed by CSO predictions, which may influence downstream diagnostic yields and timelines.
Readers should note that the full trial report, including the primary endpoint result and further contextual analyses, has been reported elsewhere. The data presented here are intended to inform understanding of MCED test performance characteristics in a population screening context within the NHS.
The NHS‑Galleri trial is registered on ClinicalTrials.gov as NCT05611632 and on ISRCTN as ISRCTN91431511. The analyses summarized here were prespecified secondary endpoints focused on descriptive test performance across three annual screening rounds among evaluable intervention participants.