NEW YORK – With the commercial launch of its blood-based multi-cancer screening test Galleri imminent, cancer early detection firm Grail has released updated performance data on the test. At the American Association for Cancer Research's annual meeting this week, the firm reported that Galleri has about 50 percent sensitivity at nearly 100 percent specificity.
The updated numbers come from the third and final sub-study of the company's circulating cancer genome atlas (CCGA) study, a large, prospective case-control effort that has informed both the design and training of Galleri, as well as its subsequent validation.
In addition, the company shared new data on the test's performance in a subset of patients from the second cohort of the study who had incidental cancer diagnoses soon after their blood was drawn for analysis.
According to Grail, the data lay the foundation for population-scale clinical implementation of its test, which will begin with an initial launch, planned for the end of this quarter.
Grail's basic premise for the clinical and economic value of the test is that a disproportionate number of cancer-related deaths are the result of a relatively small percentage of cancer diagnoses, mostly late-stage cases. "This creates a huge opportunity for the development of a screening test that could shift our diagnosis of cancers to earlier stages when they're more curable," said Eric Klein, chairman of the Cleveland Clinic's Glickman Urological & Kidney Institute, the first author of the study.
In its impact modeling, Grail has estimated that if a test like Galleri were implemented, it could reduce late-stage (stage III and IV) cancer diagnoses by more than half in the age-50 and older US population. Such a shift could translate to an overall reduction of five-year cancer deaths by 26 percent.
For the CCGA-3 study, investigators trained the Galleri classifier to target a specificity of 99.4 percent, something Klein said is "critically important for use in a broad population."
"This sort of test cannot be telling people who do not have cancer that they might have cancer," he said.
Across more than 5,000 individuals, overall sensitivity for the test was 51.5 percent in detecting the presence of cancer, with a significant and steady improvement in detection across cancer stages, ranging from just 17 percent at stage I to 40 percent at stage II, 77 percent at stage III, and 90 percent at stage IV.
Defining stage I to III cancers as a single group, the test's sensitivity was a little over 40 percent. This improved to more than 67 percent in a pre-specified set of 12 cancers that Grail estimated account for about 63 percent of annual cancer deaths in the US. "These are cancers that, by and large, do not have an established screening paradigm. And this is where this sort of test could have the biggest clinical impact," Klein said.
Tissue origin, or tumor location prediction accuracy, was 88.7 percent.
The assay picked up tumors representing 50 different cancer types, but its performance varied across this group. For example, Klein said, "the test did not work terribly well for early-stage prostate or breast cancer, but it worked very well for cervical and colorectal cancers."
This issue of tumor type variability was of particular interest to AACR participants, who probed for more information about how test specificity broke down across different cancer types. The question reflects a lingering worry that multi-cancer screening assays, apart from calling false positives, may overdetect cancers that might not need to be treated, creating an unnecessary burden on the healthcare system.
Klein said the group didn't pull data for specificity by tumor type, but previous research has suggested that the cancers that are detected by Galleri are biologically meaningful and more aggressive, on average, than those the assay misses.
Grail is currently in the process of completing and analyzing data from the first interventional studies of Galleri. But in the meantime, there are already hints from the CCGA that prospective data from such trials might shift the ultimate readout on the test's sensitivity a bit higher.
Klein said his research group doesn't have outcome data from the handful of putative false positives in the CCGA-3 cohort, but he cited Grail's previous report that among individuals in the first study subset, there were approximately eight false positives, three of whom shortly went on to be diagnosed with clinically significant stage II or III cancers.
"All the patients who enrolled in CCGA consented to five years of follow up, so in the next year or so, we should have [more] data on what has happened to patients who had both false positives and maybe some false negatives.
Grail has said it will launch Galleri for clinical use, by physician prescription, by the end of this quarter, and recently clarified that this will initially be only through what it called "partner health systems, medical practices, and self-insured employers."
Thus far, only one partner health system has been announced — Renton, Washington-based Providence, which is initially providing the test to its California, Washington, and Oregon points of care but could eventually make it available across its seven-state footprint that includes more than 50 hospitals, nearly 1,100 health clinics, and 5 million patients.
In an email this week, a Grail spokesperson added that Galleri will also be accessible by "individual medical practices."
To help facilitate access and provide support for doctors in seeking follow up for patients with positive results, the company announced last month that it will be working with Quest to provide blood collection, as well as with a company called AccessHope, which delivers an employee health benefit that connects cancer patients and their families with expertise from City of Hope and other NCI-designated comprehensive cancer centers.
After the launch of the core Galleri test, Grail has said, it intends to market additional assays that run on the same core platform for different indications, including as a diagnostic aid for patients in whom a doctor suspects a cancer but can't easily identify it by other means.
The company had shared some CCGA subgroup data on the performance of its technology in these cancer-suspicious cases at last year's AACR annual meeting, reporting similar specificity, sensitivity, and tumor origin prediction as in the full CCGA cohort.
At this year's meeting, investigators updated these data in a poster presentation, analyzing sensitivity and specificity both based on non-cancer controls and in individuals with potentially confounding factors, such as liver cirrhosis or enrollment in a hematology clinic.
The authors reported 66.4 percent sensitivity for this cancer-suspicious group overall, which increased with clinical stage much like the larger CCGA results. Specificity was 93.8 percent for cirrhosis and 99.3 percent for hematology patients.