At least 12 of my colleagues boarded planes to the American Society of Clinical Oncology conference in Chicago last week to hear Charles Swanton of the Francis Crick Institute in London share results from the world’s first randomized trial on a multi-cancer early detection test. The trial — a partnership between the U.K. National Health Service and Grail, which makes the Galleri MCED test — randomized 143,000 average English adults aged 50-77 to receive either usual care or MCED testing. The primary endpoint was reduction of stage 3 and 4 cancers across all cancer types. After three rounds of annual molecular screening, the trial failed to meet its primary endpoint — a punch line we have known for months. Nevertheless, hearing about the results in detail was invaluable. As an oncologist and active researcher in the field of screening, I found a lot of promising signals in this trial. But technical and methodological challenges got in the way of clear results and definitive conclusions. In 2025, the American Cancer Society published guidance for physicians on MCEDs. This January, a federal bill passed that enables Medicare to cover MCEDs if any are FDA approved, although no MCED is yet FDA approved or guideline-recommended, and the world of screening, primary care, and oncology has waited for evidence of clinical benefit for years. Understanding the statistical and epidemiological complexities that drove the negative results may help us get closer to assessing that benefit. The story of MCEDs is more than a decade old. In 2016, I was an M.D.-Ph.D. student at Harvard University, frustrated with the limitations of drug development for late-stage lung cancers — they develop resistance to everything! I looked into methods of catching cancers earlier, when surgeons or radiation oncologists still could treat some tumors with local therapies rather than relying on drugs alone. That’s when I stumbled into the world of MCEDs. Scientists, mostly in the private sector, were looking for molecules secreted by tumors into the blood of otherwise healthy people. Assays looking for circulating tumor DNA, epigenetic modifications, fragment length patterns, and different molecules, such as RNA, proteins, glycosaminoglycans, and extracellular vesicles, were being trained to catch cancer using machine learning. In 2020, the first single-arm trials were published on two MCEDs: the Pathfinder study by Grail and the DETECT-A study by Exact. Both demonstrated that the tests have high specificity, low-moderate sensitivity for early stage cancer, and extremely high positive predictive value (PPV) compared with single-cancer screens. Spectators of the field had mixed reactions — from baseless skepticism to over-optimism. For academic experts in the space, MCEDs seemed promising but needed to be studied in a randomized trial to understand how they will affect cancer outcomes. Enter the much-anticipated Grail results. On Saturday, at ASCO, evidence of clinical benefit in the U.K. study was discussed through the lens of numerous metrics. Among people with positive Grail tests, 52% actually had cancer based on confirmatory workups. This PPV of 52% is quite good compared with the PPV of other screening tests that are approved and recommended — in many studies, a low dose chest CT in smokers has a single-digit PPV, and screening mammogram has a PPV of about 4%-5%. But Grail placed its bet on a very hairy primary endpoint. Grail saw a substantial reduction in stage 4 cancers. But this was countered by a substantial increase in stage 3 cancers detected. Of course, the true number of cancers people have in both arms of a randomized trial should be equal, but the trial compared the Grail test to standard care, in which most cancers are unscreened. It is highly likely that many people in the standard care arm had undiagnosed stage 3 cancers, cancers that were diagnosed in the Grail arm. Some of those will be curable. However, the paradoxical “decrease in stage 4 but increase in stage 3” canceled one another out, erasing any statistically significant difference in Grail’s conglomerate endpoint. So the MCED diagnosed cancers that were never screened for before, revealing a prevalence of cancers at different stages, quietly existing in the population. As an oncologist, this makes me restless: Many people with cancer have not yet been diagnosed, and a lack of current screening for most cancer types is missing them before they become symptomatic. The ability to plan future screening trials on MCEDs requires us to understand cancer prevalence and stage distributions in the status quo better, and it is difficult. The data generated by Grail is one of the richest sources of learnings for the space. In many other randomized trials of cancer screening, reduction of stage 4 did precede reductions in mortality, with some exceptions. As an oncologist, I can say that reduction of stage 4 cancer is meaningful in many cancers; generally speaking, if a patient has stage 3 cancer, they are typically candidates for some combination of curative-intent surgery, radiation, and systemic therapy. This is true whether the disease is in the breast, head and neck, esophagus, or most other organs. Immunotherapies have improved survival outcomes in numerous types of stage 3 cancers. These definitive treatment programs aren’t possible at stage 4, when we generally rely on drugs to extend survival without expectation of cure. However, for some cancer types, reduction of stage 4 is less meaningful. In cancers of the liver, bile duct, or pancreas, for example, a patient’s shot at cure depends on candidacy for definitive surgery and little else. In these handful of cancers, reduction of stage 4 doesn’t typically represent the difference between “hope” and “no hope.” I wish Grail and NHS had picked a primary endpoint that better captures whether cancers are diagnosed early enough to offer curative-intent treatments. Regardless, the complex stage shifts achieved by this trial are actionable for oncologists in many — not all — cancer types. Beyond misadventures with the primary endpoint, the three-year duration of this trial was short — by this timepoint, other historical cancer screening trials that were ultimately positive were not yet statistically significant. For example, the landmark National Lung Screening Trial (NLST) changed national guidelines and is the reason we have lung cancer screening for smokers. My close collaborators at Brown School of Public Health led the statistical analysis of this study. The NLST involved a median of 6 1/2 years of follow-up, roughly double the Grail study. Had Grail run a longer study with four or five screening rounds and longer follow-up, would the results have been different? My best guess is “very possibly.” Overall, Grail quadrupled the number of screen-detected cancers, which are typically associated with better outcomes than cancers detected by symptoms. In the U.S., only 14% of cancers are screen-detected. Grail also reduced emergency room presentations of cancer due to symptoms. Although this is not an endpoint that epidemiologists or purists in cancer screening typically care about, most patients and oncologists would agree this is meaningful. To me, a few truths are clear. 1. Improved cancer screening must be a major strategy in the national war on cancer — we have over 600 FDA-approved cancer drugs with new approvals nearly every week, but only five guideline-recommended screening tests. In general, we can only cure cancers caught early enough, and we need to start focusing on early detection nationally rather than continue to throw the kitchen sink at patients in late stages. 2. The future of cancer screening should include molecular testing — technologies like mammography and the Pap smear remain the gold standards but are startlingly antiquated compared to the cutting-edge world of cancer therapeutics. Research in this space must continue aggressively — because 35%-40% of us will one day be cancer patients. 3. The Grail assay is not perfect, but it has paved the way for the entire field of MCEDs. Designing the world’s first randomized study of an MCED took courage. Grail took a stab at a novel primary endpoint to expedite evidence generation, and we are all learning from it. They developed numerous publications that help us think about how to work up molecular signals in clinic after a positive test, how to utilize repeat testing and mathematical modeling, and started a booming field of research that will change the face of cancer screening. No matter how you feel about this study, future trials in this space will be standing on the shoulders of the Grail experience and learning from the models, population, endpoints, and timepoints used here. As a young physician-scientist in this space, I am grateful for this trial and its lessons. Grail will measure their results again in six and 12 months. Let’s see what they find. Sana Raoof is an assistant professor in radiation oncology at the Warren Alpert Medical School of Brown University. She is also a member of the National Guideline Development Group for cancer screening recommendations at the American Cancer Society. She previously served as a consultant to several MCED companies, including Grail.
