Every screening programme carries two promises — one explicit, one implied. The explicit promise is that the test will look for something specific. The implied promise, the one the patient carries home whether or not it was ever stated, is that a negative result means safety.
In established screening, this implied promise is managed through decades of public health communication. Most women understand that a normal mammogram does not guarantee the absence of breast cancer. Most people who receive a negative bowel screening result understand, at least in principle, that the test has limitations. The language of screening — “no abnormality detected,” “recall for further assessment” — has been carefully shaped to avoid categorical reassurance. It does not always succeed, but the intent is embedded in the system.
Multi-cancer early detection exists outside that infrastructure. There is no established public health framing. There is no decades-long programme of patient education. There is a blood test, commercially available, that screens for signals associated with more than fifty cancer types — and a patient who receives a negative result is left to interpret what that means largely on their own.
The technology is elegant. Tumours shed fragments of DNA into the bloodstream — cell-free DNA carrying methylation patterns that differ between cancerous and non-cancerous tissue. Machine learning reads those patterns, detects a signal when present, and predicts the tissue of origin with near-90% accuracy. When it works, the diagnostic workup can be directed to the right organ rather than leaving the physician searching without a map. GRAIL’s Galleri test is the most advanced commercial implementation. It has been available in the United States since 2021, at roughly $950 per draw.
The performance characteristics are what require careful thought. Specificity of 99.5% means that in every thousand people without cancer, five will be told a signal was found when none exists. In a screening population of millions, that half-percent generates a large absolute number of false positives — people sent for PET scans, MRIs, biopsies, and months of investigation for a cancer that is not there. In the PATHFINDER study, among 92 participants with a detected signal, 57 were ultimately confirmed as false positives. The median time to diagnostic resolution was 162 days. Five months of clinical investigation and psychological distress for something that was never present.
This is a real cost. But it is not an unfamiliar one. Mammography produces false positives routinely — approximately one in ten women screened over a decade will experience at least one. PSA testing generates false positives at a scale that led the US Preventive Services Task Force to recommend shared decision-making rather than routine screening. The question with any screening modality is not whether false positives occur, but whether the system around the test is prepared to manage them — and whether the patient is educated before the blood is drawn, not after the result arrives.
What concerns me more is the other side of the error.
Overall sensitivity across all cancers and stages is 51.5%. For Stage I cancers, sensitivity is approximately 17–20%. This means that roughly four out of five Stage I cancers present at the time of testing will return a result of “no signal detected.” The patient receives a clean report. And the clinical danger begins not with what the test found, but with what the patient now believes.
A person who has never taken a multi-cancer screening test carries an ambient, undifferentiated awareness that cancer is possible. That awareness, uncomfortable as it is, serves a purpose: it keeps them responsive to symptoms. The unexplained weight loss, the persistent change in bowel habit, the new abdominal discomfort — these prompt investigation precisely because there is no reassuring test result sitting in the patient’s recent memory telling them everything is fine.
A person who has taken the test and received a negative result is at risk of replacing that productive uncertainty with false confidence. They have been screened. The test found nothing. The diffuse worry they carried into the consultation has been addressed. Except it has not been addressed — it has been deferred, and the deferral now carries the authority of a medical investigation. The patient may dismiss the very symptoms that would otherwise have brought them to a doctor, because they believe those symptoms have already been accounted for.
This is the most underappreciated risk of any screening test with moderate sensitivity: not that it finds things that are not there, but that it misses things that are there and, in doing so, converts appropriate vigilance into inappropriate reassurance. The gap between “no signal detected at the moment of this blood draw” and “you do not have cancer” is a gap in language — but it is a gap in which real clinical harm can accumulate.
The NHS-Galleri trial — 142,000 participants, three annual screening rounds, randomised, controlled, embedded within the NHS — was designed to test whether this technology could shift cancer diagnoses toward earlier stages at population scale. The primary endpoint was a statistically significant reduction in combined Stage III and Stage IV diagnoses. The trial did not meet that endpoint.
But the endpoint design contains a structural tension that deserves scrutiny. Every time the test succeeds in catching a cancer early enough to prevent it presenting at Stage IV, that cancer may instead be diagnosed at Stage III. In a composite endpoint that bundles III and IV together, the cancer has moved from one side of the threshold to the other, but the total has not changed. The test’s success, in a specific and measurable sense, becomes invisible to the metric intended to quantify it. This does not mean the trial was designed poorly — bundled endpoints are standard practice and were prespecified. But it means the primary result alone does not tell the full story.
Within a prespecified group of twelve high-mortality cancers — the cancers this technology most needs to detect, because no population screening exists for them — there was a greater than 20% reduction in Stage IV diagnoses in the second and third screening rounds. Stage I and II detections increased substantially. The overall cancer detection rate was four times higher than standard care. These are secondary findings, and once a primary endpoint fails, secondary results become hypothesis-supporting rather than practice-changing. But they are prespecified secondary findings, not post-hoc selections, and they point in a direction that warrants the full dataset.
That full dataset has not yet been published. Everything in public discussion — including everything I have said here — is based on a corporate press release, not a peer-reviewed paper. The data are expected at the ASCO annual meeting in late May or early June. Until then, strong conclusions in either direction are premature.
The honest position is this. For cancers where established population screening exists — breast, cervical, bowel — multi-cancer blood testing does not compete and should not be presented as an alternative. For the cancers that account for the majority of cancer deaths — pancreatic, ovarian, liver, oesophageal, stomach, kidney — the current screening infrastructure is nothing. No test, no programme, no protocol. The question is not whether a first-generation technology with moderate sensitivity is perfect. The question is whether it is additive — and for whom, under what clinical conditions, with what framing, that addition represents a net benefit.
That is a question for a physician and an informed patient in a consulting room, not for a headline. And it is a question that cannot be fully answered until the complete data are in the public domain and open to the scrutiny they require.