NEW YORK – Although researchers don't expect to report final results from a UK-based randomized prospective study of Grail's blood-based multi-cancer screening test for about a year, debate over the results has already begun.
Stakeholders gathered at Cambridge Healthtech Institute's Precision Medicine Tri-Conference (Tri-Con) last week to discuss the hurdles this new class of assays face in proving their worth, as well as the challenges to clinicians and the larger public in understanding their limitations.
Central to the debate was the acknowledgement that the expected trial results for Grail's Galleri — a blood-based assay that measures genome-wide methylation to identify cancer-associated patterns in circulating DNA molecules — will provide at most about half of what policymakers may need or expect to render judgement.
The study, being conducted within the UK's National Health Service, is designed to evaluate a specific endpoint — a reduction in the number of late-stage (stage III and IV) cancers for those randomized to multi-cancer early detection (MCED) screening compared to those in the control arm. However, panelists at the Tri-Con said the historical benchmark for cancer screening tests, a reduction in mortality, isn't something that can be directly or easily calculated from this stage-shift.
During a meeting session, University of Washington professor Larry Kessler, former chair of the Multi-Cancer Early Detection Consortium, raised the question of whether MCED tests should be treated differently in this regard, due to the practical hurdles in pursuing trials designed to directly measure an impact on mortality and the potential benefit to patients that could be lost.
"Over 600,000 Americans die each year of cancer, and two-thirds of those die from cancers that are unlikely to ever have dedicated recommended screening tests," Kessler estimated.
The earliest that a US-based randomized trial of MCED tests with a mortality endpoint might see results would be close to a decade, he added.
The NCI recently announced the launch of a three-year study called Vanguard, which is assessing tests developed by Guardant Health and ClearNote. But that study is only intended to investigate the feasibility of a larger randomized trial.
"I will tell you right now, the National Cancer Institute is not funded for [that larger] study, which … couldn't start earlier than 2028 and wouldn't have outcomes until 2034 at the earliest," Kessler argued. "We'll also have to see whether that happens at all, especially in this [political] climate. We'll have to see if the NCI exists in this climate by the time the Vanguard study is over."
Meanwhile, the possibility of making even a small dent in cancer mortality during the next 10 or 15 years is hard to turn away from, Kessler argued. "I'm not saying we must [adopt tests with limited data]. I'm just talking about why we might want to consider alternatives to waiting for the wheels of a large US-based randomized trial."
Exact Sciences Chief Medical Officer Tomasz Beer also highlighted the potential harm in moving too slowly. "If you invent something new every 10 or 20 years, taking 10 or 20 years to prove it out is not so bad," he said.
"But if you can invent something new every six months you suddenly have this gigantic bottleneck where, ultimately, we deliver to patients technologies that are not just obsolete, but many generations obsolete by the time we get to that point," he added. "I'm cautiously optimistic that we can figure it out."
Other alternatives include additional fast-tracked prospective studies with non-mortality endpoints, modeling, and post-market review.
"You could look at individual endpoints such as the proportion of screen-detected cancers that are amenable to treatment with curative intent," Kessler proposed. "You could look at health system endpoints like the economic burden over time. And you could look at cancer survival rates."
Ruth Etzioni, a biostatistician at the Fred Hutchinson Cancer Center and an expert in cancer screening modeling, responded to Kessler's proposition with a sobering projection of possible outcomes of the NHS Galleri study.
The trial randomized 140,000 individuals, with half getting Galleri once per year over three years and the other half receiving standard care under the NHS. Although there is a cancer-specific mortality endpoint planned, it's a secondary analysis for which the trial was not intentionally powered.
"I think it's very important to think about why we look at mortality," Etzioni said. "There were very thoughtful people at the beginning of the age of cancer screening who thought about endpoints and came to the conclusion that cancer mortality was the right endpoint, so I don't think we can just toss it out. We have to think about what they were trying to do and whether we can do without it."
One major issue with looking at stage of diagnosis, she added, is that it's more subjective than mortality, although even that requires adjudication, she admitted.
"If you're going to have stage as an endpoint, then both the screen and control groups have to have access to the exact same technology and the exact same procedures to stage them," she said. "That has to be part of the documentation that happens, and that is not happening at this point."
Modeling the possible outcomes of the NHS Galleri trial, Etzioni and colleagues predicted that the final results could show a reduction in late-stage diagnoses of between 6 percent and 23 percent after seven years, depending on the details of the population recruited.
How this would then translate into an effect on mortality is still an open question, but it infers that it may be significantly below the 20 percent reduction traditionally viewed as the goal for a screening test.
"I will be the first person to say that every screening trial is a half a loaf," but the NHS Galleri trial is unprecedented in its short evaluation time and surrogate endpoint, Etzioni said. The single readout next year may be more like a slice, she argued, considering how much the impact of screening can shift depending on when you assess it.
When researchers try to project the outcome of a screening program, they look at a combination of what Etzioni called "opportunity for early detection" and accuracy of early detection.
Modeling the opportunity for early detection, often called the sojourn time, involves weighing the length of time individuals remain at an early stage of their cancer, how rapidly they progress through further stages, and how long they might remain asymptomatic with advanced disease.
The modeling is complex, Etzioni said, but researchers have methods to infer probable scenarios using data from registry studies, and this is what fueled her group's projections of 6 percent, 20 percent, and 23 percent reduction in late-stage disease, based on three possible combinations of early detection opportunity and test sensitivity assumptions.
"I don't know how things are going, but if everything goes according to plan and there is good compliance with screening follow-up, then I don't think it's unrealistic that we might see something between a 10 to 20 percent reduction," said Etzioni.
"But what are we going to do then?" she added. "When we see the headline saying that the Galleri trial lowers late-stage disease by 15 percent, I expect a lot of journalists will be really excited about this and the public will be excited about it. [A] 15 percent reduction may seem substantial, but we know that a given late-stage reduction has a variable mortality impact across cancers."
Another challenge, according to Etzioni, is that mortality impact usually grows over time before settling, so whatever is calculated for Galleri earlier on may fall short of its ultimate impact, echoing Kessler's call for stakeholders to consider the potential loss to patients, and to the healthcare system, in setting too high a bar.
"I am not averse to using, with all the caveats and all the documentation and the care that you have to have, a stage-based endpoint," Etzioni said. "But we have to bring the … consideration and the thoughtfulness to that that our predecessors did to mortality."
"That is work that really has to be done, and it's not about publishing more papers," she argued.
An early report from the UK Galleri trial last year was met with some anxiety among investors, when the NHS reported that, at the sensitivity seen, it was not going to exercise an option to roll out commercial Galleri screening then and there.
Kessler, who was part of the committee that looked at the interim data and advised the NHS, said that he couldn't share details, but noted that labeling the decision "not to immediately buy a million tests" as a bad omen was not warranted.
"We set an extremely high threshold because the NHS and Grail did not want to … jump the shark. They did not want to have a middling result, buy a bunch of tests, and later discover that the trial was not so successful." The mark would have been "very hard if not impossible" to achieve, Kessler said.
Megan Hall, Grail's VP of medical and corporate affairs, added that the early readout was not an official interim analysis of the study and reiterated that the company is "cautiously optimistic about the three years of screening having an impact."
Kessler raised the question of whether a 20 percent reduction in mortality should be the cutoff point for assessing the value of an MCED test, noting that the public may have a different risk tolerance that policymakers should consider.
"You could determine the degree to which the public might be more comfortable accepting some degree of evidentiary uncertainty in exchange for earlier detection of cancer and the possibility of clinical benefit," he said. "There's room for discussion and debate."
The same point was raised by a meeting attendee, leading to debate between panelists. Etzioni highlighted the fact that although the lay public may have a different risk tolerance, they also have a different understanding of the harms and benefits of early detection that can make them vulnerable.
For example, "if you take those 70 percent of cancer deaths that come from cancers for which we don't have a screening test, and we're able to save 20 percent of them, [that might sound really exciting] but 62 percent of cancer deaths will still come from those cancers," Etzioni said.
That's not because they don't have a screening test but because of their aggressiveness, their prevalence, and a lack of treatments. "Screening isn't going to make up for that," she said.
Guardant Health Chief Medical Officer Craig Eagle argued that 20 percent mortality reduction is not necessarily an ideal threshold, and there should be a broader effort at shared decision-making to determine what that should be. "If you think about it, there are many diseases and interventions we spend a lot of effort on, even advanced cancer, where you don't get a significant mortality reduction," he said. "20 percent is a really high bar."
Exact Sciences' Beer said that he came to the panel prepared for this question and had looked up meta-analyses of both cholesterol treatment and blood pressure medications, which showed a 7 percent and 9 percent reduction in mortality, respectively.
"None of this, by the way, was demonstrated in a prospective randomized control trial," he added. "Only when the meta-analyses were done could we speak to the mortality benefits, so 20 percent is a wonderful goal, but we might want to zoom out and take a look at other areas of health where decisions were made differently."
Kessler reiterated that some of the uncertainty faced by tests like Galleri could be addressed with post-market surveillance.
"A large number of medical devices are classically put on the market with only half the evidence," he said, citing hip implants as an example.
"Hip implants are expected to last in your body anywhere from 10 to 15 years, sometimes more, but no trial for hip implants is made to last more than three years," Kessler said. At that point, the US Food and Drug Administration will clear or approve an implant with a post-market requirement to look for failure rates.
"Sometimes you get high failure rates, and those products can be pulled off the market," he added. "It's different than a cancer screening test, but it's an example of the way in which we use post-market data to facilitate putting products in the market before all the information is in."
According to Hall, Grail is also optimistic about the real-world evidence it is sitting on. "I think that's going to be a big indicator outside of a clinical trial setting as to what we could expect," she said.
She also stressed that the company is trying to be responsible in how it has implemented Galleri in the clinic as a laboratory-developed test, including an effort to follow up on "every single positive test result" that it gets and to track whether those results transition to a cancer diagnosis or not, as well as any issues with follow-up diagnostic workups.
"Knock on wood, so far, based on voluntary data that's reported to us … we're seeing that it's doing what we think it should be doing and what we've predicted that it would," Hall said.