Identifying tumours using blood samples is possible, but researchers are still working out how to validate and deploy multi-cancer tests

Putting early cancer detection to the test

Identifying tumours using blood samples is possible, but researchers are still working out how to validate and deploy multi-cancer tests

In the United States, healthy people are routinely screened for signs of a handful of cancers — women entering middle age should schedule a routine mammogram to check for breast cancer, for example.

Unfortunately, this kind of early-detection programme is not the norm for all cancers; most malignancies are spotted only at more advanced stages, when people already have symptoms and treatments are much less effective. “All those cancers that we don’t screen for amount to somewhere around 70% of all cancer-related deaths,” says Nima Nabavizadeh, a radiation oncologist at Oregon Health Sciences University in Portland.

Devising ever-more swabs and scans to detect each and every cancer out there “is just not scalable”, Nabavizadeh says — achieving broad access and adherence to the four cancer screening tests currently recommended in the United States has already proved to be challenging. But what if you could get all the answers from a vial of blood? Therein lies the appeal of multi-cancer early detection (MCED) tests, which scan body fluids for snippets of DNA, proteins and other molecular traces that might reveal a hidden tumour. Several such tests are in development, and a few have even been assessed clinically.

The results from some of these early studies are promising. “I know cancer patients who now have an early-stage cancer diagnosis that they wouldn’t have known about otherwise,” says Nabavizadeh, who has been involved with studies of an MCED test called Galleri, developed by a company called Grail in Menlo Park, California. A few of these tests — including Galleri — are making their way to the market. However, no MCED has yet received formal regulatory approval, and none is covered by US health insurance.

Many uncertainties still surround MCED tests. These include determining which cancers they can detect reliably, how clinicians should follow up on positive tests, and whether the use of these tests can meaningfully reduce the number of cancer deaths. “Screening interventions that are done on very broad swaths of the population — most of whom will not benefit — should be very well characterized,” says Hilary Robbins, an epidemiologist at the World Health Organization’s International Agency for Research on Cancer in Lyon, France. “And their benefits should be proven before they’re rolled out.”

Tracking a tumour’s traces

The goal of cancer screening is to detect molecular or physiological indicators associated with early-stage malignancy while generating as few false positives as possible. Familiar methods such as mammography for breast tumours are imperfect, but have a track record of reducing the number of cancer deaths.

For MCEDs, the challenge is steeper. A test must identify a set of molecular features, or biomarkers, that are associated with a broad spectrum of tumours and that can be measured in an easily obtained specimen such as blood. The most widely employed biomarkers for MCEDs are based on cell-free DNA (cfDNA) that is circulating in the bloodstream. cfDNA is a natural by-product of cell death, and is already used to look for fetal congenital abnormalities.

Genomic fragments are also a known by-product of tumour-cell proliferation. But turning these fragments into a viable indicator for early cancer detection has taken considerable effort, because the small fraction of DNA shed by tumour cells is drowned out by the much larger quantity from normal cells. Finding cancer in this way is a “needle in a haystack” problem, says Nickolas Papadopoulos, an oncologist at Johns Hopkins Medicine in Baltimore, Maryland.

Rapid improvements in the speed, accuracy and throughput of DNA sequencing, accompanied by falling costs, have made it possible to uncover the signal buried in this noise. In 2014, Papadopoulos was part of a team that demonstrated that they could detect cancer-related cfDNA signatures in people already known to have one of a variety of different cancers¹. The team subsequently expanded on its work with one of the earliest MCED tests, called CancerSEEK. That system is now being developed commercially under the name Cancerguard by Exact Sciences, based in Madison, Wisconsin.

Other companies are also clinically testing cfDNA-based MCED tests, including Grail, Singlera Genomics in La Jolla, California, and Burning Rock Biotech in Guangzhou, China. Some tests search for mutations or chromosomal structural abnormalities linked to cancer. Others, including Galleri, examine DNA methylation, which exerts a potent regulatory effect on gene expression. Grail oncologist Eric Klein says the company has developed machine-learning algorithms that can use methylation profiles to reveal not only the presence of cancer but also its tissue of origin, with an accuracy exceeding 90%. Cancerguard also assesses cfDNA methylation, but alongside other protein biomarkers that are known indicators of specific cancers, such as prostate-specific antigen for prostate cancer.

Most MCED tests have been assessed only using samples from people with a known cancer diagnosis — which often means that the tumour is sufficiently developed enough as to cause symptoms. But there have been two prospective studies in which apparently healthy participants were tested and then evaluated more closely to determine whether positive results were accurate. The first, the DETECT-A study, evaluated the CancerSEEK test in more than 10,000 women aged 65—75². Grail’s Pathfinder study of Galleri, by contrast, looked at both men and women over the age of 50, and included more than 6,600 people³.

DETECT-A combined its blood test with whole-body imaging, and achieved a positive predictive value (PPV) of 28%. This means that around 1 in 4 positive tests represented a real cancer diagnosis. Pathfinder relied entirely on blood-based cfDNA testing to make its preliminary diagnostic call, and achieved a PPV of 38% — more than 1 in 3 positive tests were accurate. These PPVs might seem modest, but given that testing generated a relatively small number of hits overall, this means that few test recipients will experience a false-positive test (see ‘Two major trials of multi-cancer early detection systems have yielded encouraging results’).

Two major trials of multi-cancer early detection systems have yielded encouraging results.

*Positron emission tomography–computed tomography

Exact Sciences’ chief medical officer Tomasz Beer says this modest true-positive rate is a feature, not a bug, of current MCED studies. “They’re all aiming for high specificity to limit false positives,” he says. And indeed, both Pathfinder and DETECT-A achieved an overall false-positive rate of roughly 1% across the entire cohort that underwent testing — notably better than what many existing single-cancer screening methods achieve. “Currently recommended screening tests pick up 5–10% of people who don’t have cancer,” says Allan Hackshaw, a cancer epidemiologist at University College London. Those people “go through biopsies, all the imaging scans, and it can take one or two months before they find out that they don’t have cancer after all”.

It is also important to note that a majority of the tumours picked up by MCEDs would have otherwise gone undetected. Indeed, says Klein, Pathfinder caught twice as many cancers as would have been observed with standard screening procedures.

Early enough to matter?

These encouraging numbers conceal hidden complexity, however. “It’s not just about the test — it’s about the test combined with the cancer’s amenability to early detection,” explains Ruth Etzioni, a biostatistician at the Fred Hutchinson Cancer Center in Seattle, Washington.

Although the aim is to catch cancer early, some of the correct diagnoses that the tests yielded were of late-stage disease that had passed unnoticed. In DETECT-A, only 65% of diagnoses represented early-stage disease; in Pathfinder, it was only 48%. Performance also varied across tissues. For example, most of the early-stage cancers detected in Pathfinder were blood malignancies. “That’s something to consider because detecting haematological cancers like lymphoma early is not as important as detecting solid tumours early,” says Robbins.

This inconsistency is partly attributable to the characteristics of cfDNA itself as a biomarker, which shows considerable variability across cancers. Lung, ovarian, liver and gastric tumours shed more cfDNA than do thyroid, breast and prostate cancers, says Beer, and the heavier shedders will be easier to spot. Very early-stage and premalignant growths also release less cfDNA and are likely to be missed by MCED testing. Hackshaw thinks this could be advantageous, because many such growths never turn into cancer, and a ‘hit’ here could drive more people towards overtreatment of cancers that would not have caused problems if they had gone undetected, which remains a problem for prostate and breast cancer. “If you see a cancer, then you’ve probably got to act on it,” says Hackshaw. “It might be a really good thing if MCED tests preferentially pick up cancers destined to be fatal.”

Some cancers become near impossible to stop as they spread, but not all do — and therapy options continue to emerge for advanced disease. Nabavizadeh says that in his experience, “there’s a lot of stage 3 cancers that we can actually have pretty favourable outcomes for, and identifying that stage 3 cancer is not a loss”. Thus, even an MCED that falls short on the timeliness dimension could still yield substantial clinical benefit.

Despite the promising diagnostic performance of current MCEDs, oncologists are awaiting proof that these systems actually improve patient outcomes. Beer, Papadopoulos and their colleagues have followed up on people four to five years after the conclusion of DETECT-A, with encouraging results⁴. “The ones that had stage 1 and 2 cancer, and half of the ones that had stage 3, they are still alive,” says Papadopoulos. But this is not proof that MCEDs save lives that would otherwise be lost.

The gold standard for assessing efficacy is the randomized controlled clinical trial. Only one such MCED trial is currently under way — a Grail-sponsored trial of Galleri in the UK National Health Service (NHS), which began in 2021 and is expected to report results in 2026.

But there is considerable debate around how best to measure the benefits delivered by MCED screening. “We have a lot of confidence in cancer-specific mortality,” says Lori Minasian, deputy director of the Division of Cancer Prevention at the US National Cancer Institute (NCI). “If you die of cancer, it’s a pretty objective end point.”

Still, such end points can make for long, costly studies. Klein cites an NCI trial to study the value of screening for prostate, lung, colon and ovarian cancer, which tracked nearly 155,000 participants between 1993 and 2006. “That cost upwards of US$450 million and it took 20 years for the results to be reported,” Klein says. “I’m not willing to wait, as a professional, for another 10–15 years to have a mortality end point to figure out if this new technology will have a beneficial effect.”

And so the hunt is on for alternative end points that capture mortality benefit. For the NHS-Galleri trial, the main measure of outcomes is how many advanced cancers are diagnosed among individuals who receive the MCED versus those who do not. This approach is predicated on the idea that early detection will help to prevent the emergence of late-stage cancers, and thus will result in fewer cancers that ultimately prove fatal.

A 2024 study by Robbins and her colleagues reveals a more complicated story, however. For some cancers, such as of the lung and ovary, early detection seemed to reduce the number of deaths, but this relationship was weak for colorectal and prostate cancer⁵. “It’s just a big mix of stuff that is all different,” she says, and the extent of this heterogeneity will only grow as the number of cancers being screened for increases.

Putting early detection to the test

Efforts are under way to bring further clarity to the field. The MCED Consortium has brought together leading figures from academia, industry, government and health care to develop best practices for development, testing and deployment of these systems. And in 2024, the NCI launched the Cancer Screening Research Network, which is preparing to launch its first clinical trial of MCEDs.

This trial, known as Vanguard, will evaluate at least two commercially developed MCEDs, and aims to identify sound strategies for recruiting participants, evaluating test performance and developing guidelines for clinical care to accompany screening. “The first question on the docket is, is it feasible?” says Minasian. But other challenges loom.

For example, there is the issue of how to proceed after a positive result. Some MCED tests, such as Cancerguard, cannot reveal the tissue source of a tumour signal in the blood, and require follow-up with whole-body imaging. By contrast, Galleri’s algorithms can reveal the location of a suspected tumour with striking accuracy, entirely on the basis of cfDNA profiling. But these results are presented as probabilities — and the highest-ranked tissue source might not always reflect a tumour’s true location.

Different tissues require different follow-up procedures, and Etzioni is concerned that confirmation could require substantial effort. “You need a statistician and a radiologist sitting together to figure out how to strategize it,” she says. This could become a substantial obstacle to widespread deployment of MCED screening, and might require reorganization in the health-care system. Nabavizadeh advocates creating regional centres of MCED excellence, which would be staffed with specialists who can assist those facilities that lack the expertise to properly validate MCED results.

Nevertheless, the clinical community as a whole will need to be educated about how to use MCEDs effectively and what their obligations are in following up. “There are no guidelines on when to stop, how much to do, and if I stop too soon, will I be sued?” says Minasian.

Patients will also need to understand the realities of these tests, and that a seemingly clean bill of health from a well-validated MCED does not mean they can skip the colonoscopies or mammograms. “MCED testing is intended to be used in addition to ‘standard of care’ screening, not to replace it,” says Klein.

Both Exact Sciences and Grail are conducting multiyear prospective studies to evaluate the benefits and impact of MCED screening, and how people respond in the aftermath of negative or positive test results. Exact Sciences’ Falcon study, launched last year, will track 25,000 participants for 5 years. “We’re going to be looking at how patients feel about their experience,” says Beer. A prime concern, he adds, will be “making sure that people don’t use a negative blood test as a reason to skip standard-of-care screening”.

In the United States, there are pathways that allow the public to access MCED testing even without formal regulatory approval from the US Food and Drug Administration (FDA). Grail’s Galleri test has already been prescribed to numerous US patients, and Klein says that the resulting data — from 290,000 tests so far — offer valuable insight into Galleri’s real-world performance. “It’s naysayers who say, ‘well, we can’t do anything until the FDA approves it’,” says Klein. The FDA is looking into tightening rules around its oversight of such ‘laboratory-developed tests’ (LDTs), but with a new presidential administration now in place, the future of LDT regulation is an open question.

Minasian is among those who are concerned about moving too quickly to market with a new category of tests — especially if other test developers follow Grail’s lead, presenting clinicians and patients with a confusing abundance of options. “What we really don’t know is how best to use it — I’m not sure that these are optimal for everybody who’s at low risk,” she says. Still, she adds, “they’re good enough that we want to study them so that we can figure out how to use them the right way, so that everybody benefits.”

doi: https://doi.org/10.1038/d41586-025-00530-4

This article is part of Nature Outlook: Medical diagnostics, a supplement produced with financial support from Seegene. Nature maintains full independence in all editorial decisions related to the content. About this content.

This story originally appeared on: Nature - Author:Michael Eisenstein