Opinion: Dear NHS: You need a control arm for Grail’s cancer test

Editor’s note: On January 28, after this article had been submitted to STAT, the president of Grail Europe published a letter to the BMJ that responded to an editorial on the Grail-NHS partnership. The letter indicated that Grail and the NHS planned to conduct a “pragmatic, controlled study” with randomization.

A simple blood test able to detect 50 different cancers is the grail of cancer screening. And Grail — the aptly named startup that makes the test — has persuaded the United Kingdom’s National Health Service to pilot test it in 140,000 symptom-free Britons.

But the pilot is missing a control group: 140,000 similar individuals who don’t get the Grail test.


What? This from the country that pioneered the randomized trial — and rapidly executed one early in the pandemic demonstrating that steroids can be lifesaving for critically ill Covid patients? That achievement was rightly hailed as a “world beating” British success story. If the NHS was willing to randomize people critically ill with Covid-19 to treatment vs. no treatment, it should certainly be willing to randomize seemingly healthy people to test vs. no test.

A randomized trial is the only way to learn whether a new screening test helps more than it hurts. There is little doubt the Grail test will engender more testing: more scans, more endoscopies, more biopsies. But what we really want to know is whether getting the test helps people live longer or live better.


The Grail test works by looking for tumor DNA circulating in a person’s blood. All of us have DNA in our blood released from cells in our body; in people with cancer, some of that DNA also comes from the cancer. The challenge is finding the tiny amount of tumor DNA mixed in with a much larger amount of DNA from healthy cells. When tested in patients known to have cancer, the Grail test was proficient at identifying people with advanced cancer, but not at finding early cancers: It detected less than 20% of all stage 1 cancers, including less than 10% of stage 1 breast and prostate cancers.

That’s a terrible screening test. But it’s not that surprising: small, early-stage tumors don’t provide much DNA to detect. Researchers estimate that tumors must be more than 2 centimeters — about the size of a grape — before generating enough DNA to be reliably detected in the blood.

What we don’t know is how much harm the Grail test will cause. Some patients will be made to worry about cancers they do not have, which won’t be clear until they are subjected to more extensive and invasive testing. Others will be overdiagnosed and treated for cancers not destined to cause symptoms or death. Still others will be told they have cancer but no one knows where that cancer is (in 7% of Grail-detected cancers, the anatomic site was wrong).

We worry that the NHS’s pilot of the Grail test is being driven by misleading metrics, specifically the desire to increase the amount of early cancer detection and to increase cancer survival rates. But those are the wrong metrics. Survival rates always go up with early cancer detection, but that doesn’t necessarily mean fewer people die from cancer. It may just mean people are being diagnosed earlier and live a longer time knowing they have cancer but don’t actually live longer lives (lead time bias). Or it may mean that people are being diagnosed with “cancer” that was never destined to shorten their lives (overdiagnosis bias).

The right metric to judge the benefit of testing for 50 different cancers is the rate of death. To do that you need to randomize individuals to getting the test or not getting it and, at some point down the road, compare the number of deaths in the two groups. The hope is that being tested will lead to a lower death rate. But it could lead to a higher one since those in the test group will likely undergo more invasive interventions, which come with their own risks for complications.

The truth is this: It is hard for a test to make symptom-free people live longer.

But it’s not hard to make money off the process. While the market for diagnostic tests is limited to people who are sick, the potential market for screening tests is huge. Illumina acquired Grail for $8 billion in September 2020, a steep price for a company without an approved test that has yet to turn a profit. In a call to investors, Illumina’s CEO, Francis deSouza, was clear about why he made the investment, “We want to pick the largest market opportunities … no other space meets that criteria as well as early cancer detection.” DeSouza predicts a $60 billion market for a test with no randomized controlled evidence that it helps people live better or live longer.

This highlights the worst of the medical-industrial complex: promising profit before ever proving benefits to patients.

Cancer screening has taught medicine a difficult lesson: Early detection is a double-edge sword. The only way to determine whether the benefits exceed the harms — and whether it saves lives — is to do a randomized trial. Even Grail scientists agree on this point: “a randomized design is likely required to achieve a practice-changing level of evidence.”

American physicians look to their colleagues across the pond to do this kind of rigorous clinical research. We trust the NHS can make that happen. But it should wait until the pandemic subsides. There is plenty of health and economic-related anxiety out there already; looking harder for cancer in people who seem to be healthy will only add to it.

David Carr is a molecular genetic pathologist, medical director of the Division of Laboratory Genetics and Molecular Pathology at the Detroit Medical Center, and an assistant professor at Wayne State University School of Medicine in Detroit. H. Gilbert Welch is a general internist, a senior researcher in the Center for Surgery and Public Health at Brigham and Women’s Hospital in Boston, and the author of “Less Medicine, More Health — 7 Assumptions that Drive Too Much Medical Care” (Beacon Press, 2016).

Source: STAT