NEW YORK – Multi-cancer early detection (MCED) firm Grail published a new study on Thursday that expands on earlier modeling work to analyze the impact of different testing intervals on the ability of its Galleri circulating tumor DNA (ctDNA) assay to shift cancer diagnoses across a population to earlier, more treatable stages.
The projections also sought to translate that stage shift into an impact on mortality using database records that link stage-specific incidence and cancer-specific survival.
The authors concluded, from their findings, that adding MCED test screening to usual care at either an annual or biennial (once every two years) interval could improve patient outcomes. However, annual screening provided more overall benefit than biennial screening in the team's models.
In the report, published online in BMJ Open, investigators used two models of cancer stage transitions, a "fast" model in which cancer progresses from stage I only after two to four years, and a "fast aggressive" model in which it progresses after one to two years.
They then looked at what the diagnostic yield would be for the two different progression models with testing intervals between six months and three years but focused mainly on comparing one-year and two-year intervals.
In the best case scenario — fast cancer progression and annual screening — the authors projected that there would be 370 more cancer signals detected per year per 100,000 people screened, leading to a 49 percent reduction in stage III and stage IV diagnoses, and 21 percent fewer deaths within five years.
Ruth Etzioni, a leading epidemiologist at Fred Hutchinson Cancer Center, said that to understand this type of work, it is crucial to recognize that all modeling falls in the realm of speculation.
Meanwhile, every choice made — from assumptions about the rate of cancer progression, to presumptions of test performance, to the simplification of the complex real-world posttest landscape of medical follow-up — can skew the results of these models significantly in one direction or another.
Etzioni and her colleagues have also been working on modeling MCED impact on cancer diagnosis stage-shift. At Cambridge Healthtech Institute's Precision Medicine Tri-Conference in March, she shared some of their findings, looking specifically at Grail's ongoing trial of Galleri with the UK National Health System. The choices her team made yielded a much more sober picture of stage shift, projecting that the final results of that trial 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.
In an email, a Grail representative said that while both their and Etzioni's approaches share a lot of technical infrastructure, there are large differences in the cancer case mix. The Grail study modeled all cancer, while Etzioni and colleague's analysis of the NHS study is limited to a smaller number of cancer types.
The Grail method also breaks effects into all four stages instead of breaking stages into two groups. All of this could play a role in the differing conclusions.
In Grail's new paper, the company used performance data for different cancer stages from its earlier case control work to calculate the rate of ctDNA detection of various tumor types in its different progression models, something that again differs from Etzioni and her team's approach.
She raised the question of whether Grail's use of a straight measure of Galleri's performance in a case control setting might affect the modeling results. The test's sensitivity had been calculated for tumors that had already been clinically diagnosed. That doesn't mean the same detection rates would be seen weeks, months, or years before, when there could be less tumor-shed DNA present in the blood.
Of note, alongside data from other sources, Grail's own findings, presented in 2023 in an abstract at the American Society of Clinical Oncology annual meeting, have demonstrated that detection rates degrade the further back in time you go from clinical diagnosis.
In Etzioni and her team's work, they have specified a no-more-than-two-year early detection window, with test sensitivity dropping 50 percent after one year, which could also account for the very different result in terms of late-stage diagnosis reduction.
In Grail's comparison of one- versus two-year screening intervals, the worst-case scenario was biennial screening in a fast-aggressive timeline, which would reduce late-stage (III and IV) diagnoses by 31 percent and five-year deaths by 14 percent. This was still significantly higher than what Etzioni and colleagues have seen in their work.
Regarding Grail's choices on how to model the timing of cancer stage advancement, she called the reported distributions "optimistic."
"We were a little more conservative in the window of opportunity, and a little more conservative in the window of opportunity relative to published sensitivities," she said.
The Grail authors were candid about the caveats to some of their methodology including projecting that individuals would adhere 100 percent to MCED screening at the specified frequency, and assuming 100 percent accuracy of confirmatory tests leading to effective treatment. But they said that these practices are necessary and standard in such modeling.
"Real-world rates of adherence to recommended screening schedules and diagnostic follow-up will vary and result in a lower population benefit. Individuals may also elect against recommendations and warnings otherwise to substitute MCED screening for recommended single-cancer screening, thereby constraining potential mortality benefits," they wrote.
"We further assume that a reduction of late-stage cancer incidence would have an impact on mortality due to detection at an earlier stage, which is contested in the literature, [and] due to these necessary modelling assumptions, real-world benefits are likely to be less than those estimated in the model," the authors added.
The team also acknowledged that cancers may have complex properties that they were not able to model. Some may not progress sequentially through four stages but rather skip straight from stage I to metastatic disease. However, taking these possibilities into account was out of the scope of their study, they said.
As real-world evidence emerges, the Grail authors wrote that it will hopefully offer an opportunity to calibrate the true dwell times and patterns of stage evolution for different tumor types, which will allow for a more accurate assessment of optimal test intervals for MCED.