Newsletter Archive
Listen to Our Podcast
Multi-Cancer Blood Tests Have Yet To Beat Standard Screenings
Melissa Wells was feeling perfectly healthy when an experimental blood test, called Galleri, pointed doctors to a cancer hiding in her bile ducts. Scans and a biopsy confirmed a small, localized tumor — the kind that often grows silently for years before spreading beyond the reach of treatment.
Yet, because it was caught early, surgeons could cut the tumor out. Wells, now one year cancer-free, credits the blood test with giving her a chance at a long life. “If I hadn’t taken this,” said Wells, 54, “who knows what would have happened?”
In September, Wells and three other patients appeared before a panel of FDA advisers who were considering whether Galleri should be approved by the agency. Each patient had an undiagnosed cancer in a different organ — bile ducts, esophagus, gallbladder and pancreas — that were identified by the Galleri blood test. All the cancers were found before they caused symptoms and all were treatable.
“I want to tell you: there are real people here,” Roger Royse, whose early stage pancreatic cancer was detected by Galleri, told the advisory group. “We’re not just statistics.”
Galleri, developed by the California biotech company Grail, is one of the first of a new class of cancer screening tools called multi-cancer detection tests (MCDs) that scour blood for indications of cancer anywhere in the body. There are rival blood tests from established diagnostics firms such as Abbott and Guardant Health that use a similar method. Cancer cells shed DNA into the bloodstream so MCDs look for markers of that chemical effluent. For that reason, the tests are sometimes called liquid biopsies.
Currently, US guidelines advise routine screening for only a handful of cancers like breast, cervical, colorectal and lung. MCDs would vastly expand the testing available and the potential cancers detected. If the FDA authorizes Galleri — as recommended by the advisory panel — the decision could pave the way for broader insurance coverage. Eventually, that could mean millions of Americans taking the test each year.
But, as many in the medical community point out, there are significant risks in making such tests widely available. At this point, the stories of Wells and the other people who testified before the FDA panel are exceptions, not the norm, as many cancers identified by the test are found too late to treat meaningfully. The test can also miss a significant number of cancers altogether; and expanded testing could lead to a surge in unnecessary and expensive lab work.
Nor is there evidence across the population as a whole that finding these cancers earlier helps people live longer, which, as gastroenterologist David Weinberg of the Fox Chase Cancer Center in Philadelphia points out, is “the entire reason we do screening tests.”
“I hope this works,” Weinberg adds, “but I would like some evidence that it does.”
For researchers, the science behind MCDs like Galleri is genuinely compelling and may represent a step on the journey to the holy grail of cancer screening: a single blood test offering definitive diagnoses. But the journey isn’t over.
A new kind of test
Galleri’s story began not in an oncology clinic but in prenatal care.
In the late 1990s, researchers found fragments of fetal DNA floating in the blood of pregnant women. That discovery became the foundation for noninvasive prenatal tests, which now screen for Down syndrome and other chromosomal conditions (and can reveal a baby's sex) with nothing more than a blood draw from the expectant mother.
Then came a twist. As the DNA sequencing company, Illumina, ran its prenatal test on more than 100,000 women, a handful of results turned up chromosomal irregularities that could have indicated cancer, but that follow-up testing failed to find in the babies. Researchers eventually realized that the cancer signals weren’t from the fetal DNA. They were from the DNA of the mothers.
That chance finding led Illumina to spin off Grail in 2016 with the goal of developing cancer tests. Prenatal tests essentially count DNA fragments to spot a missing or extra copy of a chromosome. Grail’s scientists tried that approach for cancer, but found that reading methylation, the chemical tags that dot DNA and help switch genes on and off, worked better. These methylation patterns differ between cancerous and healthy cells and from one tissue type to another.
So Galleri can not only flag a cancer signal but also predict with an accuracy of about 90 percent where in the body it came from, allowing a single blood draw to screen for more than 50 types of cancer at once. In a large North American study of roughly 32,000 people over 50 with no suspicion of cancer, Galleri flagged a cancer signal in 287 individuals, around 60 percent of whom indeed turned out to have a hidden tumor.
Fortunately, the test tends not to pick up the kinds of indolent tumors — a slow-growing prostate cancer, say, or a precancerous breast lesion — that would never have threatened a person’s life, but that other screening programs routinely find, sometimes leading to unnecessary treatments that can harm patients.
The evidence cuts the other way too, though. The largest trial run to date involved more than 142,000 older adults in England. The trial evaluated whether adding the blood test to standard care would lead to fewer diagnoses of advanced cancers, which are far deadlier and harder to treat, over the course of three years.
Galleri did identify some cancers earlier. But more often than not, it turned up tumors that had already grown or spread beyond the point that curative treatment was likely to succeed. About 40 percent of the cancers Galleri detected were caught at stage 1 or 2.
And while a blood test that finds any early-stage cancers before symptoms appear might sound impressive, the real test is how Galleri stacked up against the status quo — and on that measure, the test barely moved the needle. Across all cancers that could be measured, about the same share were caught early whether or not people got the blood test on top of standard screening exams and doctor’s visits. Galleri did turn up more cancers overall, but it boosted the number of early-stage diagnoses by only about 9 percent.
That is why some researchers take issue with the industry claim of early detection. “The term ‘early’ suggests that the test will detect cancer at an early stage, when treatment is more likely to be curative,” said Ariadna Tibau Martorell, an oncologist at the Cancer Innovation and Regulation Initiative at Brigham and Women’s Hospital in Boston.
And for all the unsuspected cancers that Galleri finds, it misses many more. In clinical trials, among the thousands of participants diagnosed with cancer within a year of their blood draw, the test had flagged only about one-third. By comparison, mammograms catch nearly nine in 10 breast cancers, and colonoscopies spot the vast majority of sizable precancerous polyps and colorectal tumors.
A clean result on those other tests is significant. A negative Galleri result? Much less so.
Nonetheless, people may mistake a “no-cancer-signal detected” finding on the blood test as confirmation that they are in the clear, and then forgo recommended screening tests — colonoscopies, for instance, that are designed not to find established tumors but to catch growths while they are still premalignant and removable. A modeling study published in July found that relying on an annual blood test like Galleri alone would save far fewer years of life than if the same people got a colonoscopy reliably every 10 years, starting at age 45.
Costs versus benefits
The mixed results have created a debate among clinicians grappling with how much weight to give a screening tool that can uncover otherwise elusive tumors but has yet to demonstrate a clear population-level benefit.
“We all want this to work so badly because our current system for screening for cancer is so bad and so broken and so expensive — and this technology feels tantalizingly close,” said Deborah Schrag, an oncologist and health services researcher at Memorial Sloan Kettering Cancer Center in New York, who was involved in early trials of the Galleri test. “But it’s tricky business.”
Peter Sasieni, a cancer epidemiologist at Queen Mary University of London, acknowledges the open questions but thinks the test can be deployed safely if doctors, public health authorities and Grail are all careful about what they tell people, noting the limitations of a negative result and the need for people to keep up with their other regular screenings. “The important thing is communication,” said Sasieni, one of the research leads on the large English trial.
He does not oversell it: “Scientifically it’s looking really good, but not good enough for us to say full steam ahead.” The test, Sasieni concedes, “is not finding as many early cancers as we’d like.”
Then there is the cost. Even if Medicare allows people to get the test for well below Galleri’s list price, testing millions of healthy people could produce an annual bill running into the tens of billions of dollars.
“We’re intervening in the whole population to end up saving very few,” said Ruth Etzioni, a biostatistician at the Fred Hutchinson Cancer Center and one of the leaders of the Cancer Screening Research Network, a government-backed consortium evaluating emerging cancer-detection technologies. “The equation does not compute.”
The technology is likely to improve, with innovations in DNA analysis and the integration of signals from blood proteins and other biomarkers — advances that could help catch more early-stage tumors but might also generate more false alarms.
Still, with rapid improvements in cancer treatments — targeted drugs, immune modulators and cellular therapies that can beat back advanced disease — the calculus around multi-cancer testing could change. In the not-too-distant future, even a modest head start in diagnosis could prove more valuable than it is today.
In the interim, millions of Americans could start taking a test whose value is still unproven. For Weinberg, the problem comes down to what’s known and what isn’t. “Ironically, we understand the risks pretty well,” he said. “We don’t understand the benefits.”
Advances That Matter
Cattle take shade at a solar farm in Christiana, Tennessee AP Photo/Joshua A. Bickel
Solar farms are getting new residents: beef cattle. Around two-thirds of agricultural land in the US is used for grazing livestock. Meanwhile, local communities rail against solar farms taking over vast swaths of their landscapes, with a small utility-scale installation of 20 megawatts typically requiring at least 100 acres of land. A company called Silicon Ranch is betting that land can perform double duty by combining cattle grazing with photovoltaic installations. Canary Media reports that the company, building on existing sheep-grazing operations, has set up a 3.25-megawatt trial solar project in Christiana, Tennessee, southeast of Nashville, where a herd of 15 cattle grazes on 40 acres of land beneath solar panels installed at a height above their heads. Unlike sheep, cows are large and heavy and can easily damage the solar hardware, so the company has raised the panels higher than the usual 24 inches, and limits how far panels tilt to ensure the cows can’t reach them. The company has also used a breed of cattle known for its gentle nature to reduce the potential for damage. So far, the project, which provides enough energy for 500 homes, seems to be working. The panels shade the cows and keep the grass lush even during hot spells, and the herd has taken over most of the mowing. It’s cost-effective too: Making the solar arrays cattle-proof added just 3 percent to the cost of the project, and it offers power to the local community at a price far below that of wholesale electricity. The long-term hope is that the approach can ease the concerns of local communities by making the combination of clean power and local cattle ranching an attractive agribusiness.
Putting trackers on AI chips could deter smuggling. Nvidia CEO Jensen Huang’s resignation and reported affair with an alleged Chinese intelligence asset have pushed national security issues around AI chips into the spotlight, raising questions about how US lawmakers will respond. The sale of high-performance AI chips to China has been restricted by the US since 2022, but evidence suggests China has access to the hardware. One course of action lawmakers have proposed would enable chip makers to track where AI chips end up, requiring them to notify the government if they suspect the chips have entered China. But how would that work? There are several approaches, reports IEEE Spectrum. One option: so-called “ping-based location verification,” which would narrow down the chip’s location by measuring how long it takes for a small message, or ping, to travel from a chip to several trusted servers around the world. Another takes the same idea a step further, using cryptographically signed messages to achieve a similar result with increased trust. It has previously been unclear how plausible it might be to enact such approaches with real-world hardware, but MIT Technology Review reports that, according to Representative Bill Foster, Nvidia privately told him that the location tracking required by his proposed legislation could be achieved. Still, such approaches aren’t foolproof, as someone with physical access to a chip could tamper with it to block or spoof its signals. Researchers have proposed further measures that could help address such interference, including tamper-proof designs for chips. But none of this addresses a separate loophole: Chinese firms can legally buy computing power remotely from cloud providers in nearby countries such as Malaysia, a practice that would also need to be regulated in order to prevent China’s access to the hardware.
AI is helping prove software is secure. Artificial intelligence is an undeniable security threat, but it could also help secure software better than any other tools we currently have. One way that could happen is through a process called formal verification, which uses mathematical methods to prove — or disprove — whether a software system can do what it claims to do, including maintaining promised security and privacy. The approach has been around since at least the 1960s, but AI, with its prizeworthy math skills, has made it more accessible. Now, reports Wired, researchers at Google DeepMind have used AI tools to formally verify the core cryptographic code within the open-source Signal protocol, the encryption system that protects billions of private messages in the app of the same name as well as other software, including WhatsApp and Google Messages. Before advanced AI, formal verification was too labor-intensive to employ at a useful scale. Properties of the software system to be analyzed needed to be precisely described in ways that could be turned into the mathematical formulas needed to prove that the software worked as promised. The Google DeepMind analysis of the Signal protocol took four weeks, a job that the company says would historically have taken a team of human experts years. The researchers found no problems in the code it examined, and when — for testing purposes — the researchers reintroduced two bugs that had previously been fixed, the system showed that the buggy code was insecure. Google DeepMind conducted the Signal analysis as a free research project, but has not shared what it cost. Doing this across all software is the dream of cybersecurity experts, but it’s unclear how easily the approach can scale, given the vast quantities of software in existence and the so far unknown cost of having AI inspect.
Magazine and Journal Articles Worth Your Time
Trying to Cure My Cystic Fibrosis, from Asimov Press
4,000 words, or about 16 minutes
Cystic fibrosis — a genetic disorder that causes the body to produce thick, sticky mucus that clogs the lungs and other organs — has become highly treatable in recent decades, and the expected lifespan of patients with the disease has more than doubled in the last 40 years. One of the main solutions, called Trikafta, is a mixture of three drugs that help boost the function of a protein created by a gene called CFTR, which is mutated in sufferers of cystic fibrosis. But as many as 10 percent of cystic fibrosis patients — including Jacob Witten, a professor in genetic sciences at Case Western Reserve University and the author of this essay — have a mutation in the CFTR gene that means they don’t create any of the protein, rendering Trikafta useless. For these people, the best hope is gene therapy, which could theoretically supply a working copy of the gene. But the key word here is theoretically, as getting gene therapy into the lungs is difficult. The natural defenses in this part of the body are significant; the viruses used to deliver most gene therapy provoke immune responses that make repeat doses hard to administer; and the main alternative to a virus as delivery system are tiny fatty particles called lipid nanoparticles, which have so far proved either too inefficient or prone to inflaming the lungs. Yet several companies are working on techniques that overcome some of the issues, and Witten, who has dedicated his life’s work to curing cystic fibrosis, is optimistic that an effective treatment for his form of the disease may arrive within the next decade.
What Constant Surveillance Is Doing to Children, from The Atlantic
6,400 words, or about 26 minutes
Surveilling our children has become normalized, whether that’s using smart baby monitors to track the breathing of infants or using GPS to keep tabs on college-age kids. These days, about three-quarters of parents of 14- to 17-year-olds can, through phone apps, see where their children are. This story, written by a new mother deluged and partially seduced by tools that promise nonstop updates on her daughter’s every breath, poses the questions: “If hypervigilance is the only way to keep your kid safe, how can you ever stop watching?” And, perhaps more important, “What is all of this tracking doing to us?” The story is anecdotally rich — a mom, observing that her daughter was in her dorm room 10 minutes before an exam, exhorted other parents to send their kids to check on her; a dad raced home because his GPS tracker showed his daughter out of school at the wrong time of day, when in fact the app was glitching — and argues that monitoring not only robs children of the chance to build independence and resilience, but feeds parental anxiety. This, if you have children, is the kind of finding that certainly strikes a chord, and the story cites research showing corresponding anxiety in both parents and kids. The next leap the article takes, however, is more questionable — that surveillance can damage the relationship between parent and child. As of now, the handful of studies on teen tracking reveal mixed or very small effects on parent–teen relationships, so it is left up to readers to interrogate their own use of tracking technology and whether it’s a boon or detriment to their relationships.
A New Era for the Dawn of Life, from Quanta
10,800 words, or about 43 minutes, across four articles
How did life get started on this planet more than four billion years ago? The question has occupied scientists for the best part of a century, and also divided them into camps that fiercely defend different models of how basic chemistry could have serendipitously created the building blocks of life. In this package of stories, Quanta reports that the study of early life has been reinvigorated by the discovery that — actually — life’s building blocks form surprisingly easily, and a new generation of researchers is now combining chemistry and biology to try to make sense of the historically divisive problem. To be clear, there are still multiple competing theories about how life began. One suggests that small, fatty bubbles may have contained molecules that sped up reactions that allowed them to make more of each other, creating a rudimentary form of evolution. Another proposes that energy and minerals located at places like hydrothermal vents on the seafloor kick-started self-sustaining cycles of chemical reactions resembling metabolism. But these sorts of ideas alone don’t fully explain the origin of life, and the longest story in the package grapples with the different ways in which researchers are now piecing together ideas to better inform speculation about how it all began.
More stories worth knowing about: New evidence shows that in some cases the viruses used to deliver gene therapy can cause cancer. The economic benefits of the Covid vaccines may be double what we thought. The FDA warned that GLP-1 drugs could increase the likelihood of a rare eye condition. Why it’s almost impossible for doctors to “do no harm.” How AI doomers won the midterms. A new metric based on waist circumference and excess body fat is better than BMI for predicting health. Scientists are finally starting to understand endometriosis. At a closed-door conference in Iceland, influential researchers, advocates and funders gathered to discuss how we should start geoengineering the planet. Scientists are starting to learn about the origins of dementia from people who dodge it. How China became an automobile superpower. And 10 experts weighed in on what the biggest AI nightmares really are. (Spoiler: Don’t sweat extinction, but absolutely fear a nuclear catastrophe.)