select search filters
briefings
roundups & rapid reactions
Fiona fox's blog
Scientists comment on a press release from Grail and presented at ASCO on unpublished results from the NHS-Galleri trial.
Prof Nitzan Rosenfeld, Director, Barts Cancer Institute, Queen Mary University of London (QMUL), said:
“It is good to now see details of the results of this important and ground-breaking trial, which was conducted with great care and to the highest scientific and statistical standards. Trials in early detection of cancer are difficult and expensive which is why we have so few of them. Cancer can develop over many years, from cellular and then tissue abnormalities into symptomatic disease, which at that point is often lethal or incurable. We don’t always know if what looks like early cancer will become lethal, or conversely if we find it early whether we can still cure it. This is why the only reliable evaluation for a cancer early detection test, in theory, is a lengthy randomised trial with mortality endpoints. The problem is, that to run such a trial to completion would take close to two decades, which means a long delay and high costs, and it means that by the time the trial is complete the technology will likely be obsolete and so much will change that it might no longer reflect the disease incidence or treatment path. So, this trial is by necessity a compromise. It tells us a lot about how the test performs, and it will continue to collect data which will tell us more about the impact of the test in the coming years. It is an ambitious project which was designed and initiated despite a high degree of uncertainty, which means that some decisions about the trial design were based on optimistic guesstimates of the effect of the intervention: blood-based screening for cancer detection using the Galleri test. To compensate for the lack of a mortality endpoint, and to keep the timelines shorter and costs contained, the primary endpoint goal was ambitiously set at an overall reduction in stage III-IV cancers over a fairly short period of 3 years.
“It is disappointing that this goal was not reached. We should keep in mind that even if the primary endpoint would have been met, there would still be valid criticism and uncertainty about the interpretation of the results, because this is not a mortality endpoint. The partial results disclosed today, and the further results we will see when this is published after peer review, allow us to start to dig into the details of the findings, and pre-planned secondary analysis, and to draw some optimistic inferences. As we do that, we should be mindful that when we interpret results in hindsight there is a risk that we might be “moving the goal posts”; and that results that are shared so far are a subset of observations that highlight (as far as we can surmise) more favourable findings. Also, we should keep in mind that health-economics considerations which were based on a pre-specified statistical performance may no longer hold, and the cost-effectiveness of the test might not be as favourable, even if results suggest possible benefits.
“With those caveats in mind, it is encouraging to see several results. Performance metrics of the test, as were assessed before the trial, continue to hold in the “real world” trial setting. The test is very good at detecting undiagnosed stage III and stage IV disease. Previously reported performance statistics for the test showed that it is not as sensitive for detecting cancers at stage I and II. In the trial, the test detected 30.7% of all cancers, and 54.7% of cancers in the pre-specified group of 12 cancers, that were diagnosed within 12 months after the blood test (“episode sensitivity”), including cancers with no screening options. If used to supplement existing screening options, the results show that the test can detect a subset of cancers earlier. The numbers suggest that it will be important to maintain current screening in parallel – a further breakdown into cancer types will be needed to make a direct comparison.
“The test maintained an impressively high specificity (99.55%) and a high positive predictive value of 52.0% which means few false positives, and only a small number of individuals that may undergo excessive tests with associated costs and anxiety. More than 50% of individuals in the trial population who had a positive finding were found to have cancer. In comparison, statistics published in 2024 by NHS England showed that of individuals who had urgent referrals for suspected cancers, a much lower rate of 6% resulted in a cancer diagnosis*. The accuracy of the Cancer Signal of Origin (CSO) prediction was very high at 92.5%, which can give confidence in deciding what to do next if a test result is positive. Because many of the cancers were diagnosed in stages III and IV, we need to wait for more data to be released to evaluate if the CSO prediction was also accurate for early-stage disease.
“Notably fewer cancers were diagnosed in stage IV which is the most lethal. This does not hold when combining stage III and IV, due to the increase in stage III cancer detection rates, especially in the first (prevalent) screening round, which is why the primary endpoint was not reached. For the subset of 12 pre-specified cancers, there is an increase in the rate of cancers diagnosed in stage I and II. Survival rates for patients with earlier stage cancer are better, which is partly a result of better cure rates, and partly a “lead time bias” which means that even if we’ve done nothing to cure the disease there is a longer time from diagnosis to death. It is difficult to disentangle the two effects. This is a key shortcoming of the stage-shift endpoint.
“Perhaps the most exciting part of the results is the trend that is seen in the subset of results released so far, which shows a stronger reduction in stage IV disease and in combined stage III and stage IV disease diagnoses as the trial progresses from the first (prevalent) screening round into the second and third rounds. It was expected that the first round of screening would create a “bump” in cancer detection, by finding disease that’s prevalent in the population and has not yet been diagnosed. Presumably, there was an (optimistic) expectation that this initial bump would include more early-stage disease. This did not pan out, and the first screening round resulted in detection of many stage III cancer cases. The decrease in the latter years suggests a provocative proposition, that if screening continues for more years the reduction in late-stage disease might continue to accumulate. With the current results, this remains an untested possibility and does not provide the evidence to assess the long-term effects, and cost-effectiveness, of screening with this test.
“What is clear beyond doubt from these results is that signal indicating cancer and cancer subtypes can be detected, sometimes also at early stages, by analysis of epigenetic patterns in circulating cell-free DNA in blood. The research community, both academic and in industry, is continuing to develop this valuable source of information into a variety of tests with increasing resolution and sensitivity, to provide more accurate tools to improve outcomes for individuals with cancer.”
*https://digital.nhs.uk/data-and-information/publications/statistical/mi-urgent-suspected-cancer-referrals-referral-conversion-and-detection-rates/april-2021-to-march-2023
Professor Richard Houlston, Head of the Division of Genetics and Epidemiology, Institute of Cancer Research, London, said:
“This is the largest randomised trial so far to evaluate a multi-cancer early detection blood test. However, the researchers have presented their findings far more positively than the overall results justify.
“The study’s main goal was to show a reduction in late-stage cancers overall, and this primary endpoint was not met. While some secondary findings are encouraging, in so far as a possible reduction in the most advanced cancers after repeated screening rounds, these results remain uncertain and should be interpreted cautiously.
“Some of the statistical analyses could still be explained by chance, and one explanation proposed by the researchers that cancers were simply being detected earlier rather than prevented is speculative. Longer follow-up will be needed to understand whether these tests genuinely improve long-term outcomes for patients.
“There are also important practical questions about how this type of screening could work within the NHS. Because the test has poor sensitivity for early-stage cancers and can also produce false positive results, many people may go on to have additional scans or biopsies that ultimately do not find cancer.
“Expanding this kind of testing at population scale would place massive pressure on already stretched diagnostic services at a time when the NHS is struggling to meet cancer waiting-time targets.”
Does the press release accurately reflect the science?
“Not entirely. The trial’s pre-specified primary endpoint, which was a reduction in combined Stage III and IV cancer diagnoses, was not met, yet the press release leads with one of the secondary endpoint results, thereby misrepresenting the overall picture.
Is this good quality research? Are the conclusions backed up by solid data?
“The trial is important as it’s the largest randomised controlled trial of a MCED test so far conducted. While Stage IV reduction signal, strengthening to 26% by the third screening round, is potentially meaningful, several confidence intervals cross 1.0, and the explanation for the Stage III increase in the first round that Stage IV cancers were being reclassified earlier is entirely speculative.”
What are the implications for policy and practice?
“The failure to meet the primary endpoint is the crux of the issue here. Mortality outcomes will be available in a couple of years, examination of which will be warranted. However, on the basis of results from this and smaller trials, there is no evidence base upon which to justify implementation of Galleri at a population scale. Furthermore, this would risk diverting finite diagnostic resources, such as CT scans and biopsies away from symptomatic patients to investigate asymptomatic individuals, driven by the test’s low sensitivity and high false-positive burden. In addition, accommodating this vast new diagnostic cascade would demand infrastructure and financial investment at a time when the NHS is already failing to meet its 80% Faster Diagnosis Standard, based on the most recent 2026 performance.”
‘NHS-Galleri: Primary Results From a Randomised Controlled Trial to Assess the Clinical Utility of a Multi-Cancer Early Detection (MCED) Test in Population Screening’ by C. Swanton et al. was presented at American Society of Clinical Oncology (ASCO) Annual Meeting 2026; and was under embargo until 13:00 UK time on Saturday 30th May 2026.
Declared interests
Prof Richard Houlston: no COIs to declare.
Prof Nitzan Rosenfeld:
-I was an officer of Inivata and Neogenomics.
-I have had collaborations and joint publications with Inivata, Neogenomics, Illumina (many years ago now), Ultima Genomics, as part of these collaborations data was generated by commercial companies (contribution in-kind) but no direct funding was provided.
-I was a member of a CRUK committee and I received funding from CRUK and EU/UKRI.
-I am an inventor on patents on cancer detection that are managed institutionally.
– I am a QMUL employee, and QMUL has a relationship to the trial via my colleagues. I am not directly part of it or funded by it.