NEW YORK – A slew of researchers shared updates on efforts to develop and show value for multi-cancer early detection (MCED) blood tests at this week's annual meeting of the American Association for Cancer Research (AACR), as others continued to highlight crucial questions about clinical implementation.
The new data included a detailed analysis by Grail addressing concerns about the potential for its clinically available Galleri assay to heighten the overdiagnosis issues currently faced in cancer screening.
Other presentations featured some of the first comprehensive data from two Chinese companies, Geneseeq, and SeekIn, on their respective MCED assays, as well as an introduction to a novel approach being advanced by British firm ProteoType Diagnostics.
Weaved into these webcast presentations were additional analyses probing some of the pressing utility questions about population-based multi-cancer screening that might only be eventually answered by long-term observation and/or randomized studies.
Emerging data
Geneseeq presented two studies during the meeting session, one describing its case-control assay development, which mirrored those taken by other companies in the field. The firm also shared some early results from its ongoing prospective validation of its "CanScan" test in an intended-use cohort.
CanScan, which analyzes patterns of DNA fragmentation, achieved approximately 87 percent sensitivity and 98 percent specificity in an independent training set, researchers reported.
The CanScan test was also able to determine the origin of a cancer signal among 10 pre-specified tumor types with an accuracy of nearly 84 percent. If the top two predicted origins were considered that accuracy rose to about 92 percent.
A second Geneseeq presentation highlighted early results from the company's follow-up prospective study in the asymptomatic population these tests are intended to serve. The company began enrolling patients aged 45 years to 75 years without any cancer-related symptoms starting in June 2022. By June 2023, a total of 3,724 participants with analyzable samples were included in the analysis, 29 of whom were clinically diagnosed with cancer at the study cutoff point.
Notably, eight of these were not in the 13 targeted cancer types that CanScan had been optimized for. The majority, 26 of 29, were stage I or stage II. Investigators calculated an overall sensitivity of 55 percent.
For stage I, sensitivity was approximately 48 percent, for stage II it was 80 percent, and for stage III it was 80 percent.
In contrast to the company's case-control study, the power of tumor origin prediction in its intended-use cohort was significantly lower, at about 62 percent overall, but this again rose when allowing for a top-two predicted origins, to 85 percent on average across stages.
Addressing clinical utility
Grail's contribution to the session was an analysis of data from both its case-control CCGA study, and its prospective intended-use trial PATHFINDER, focused on prostate cancer as a subset of the 50 cancers for which the test reports results.
Prostate cancer is both overdiagnosed with available protein tests and highly prevalent, so it offers a unique window to explore potential clinical impacts of population-based multi-cancer screening, Grail distinguished scientist Eric Klein said.
The company has presented data before suggesting that it's Galleri test appears to preferentially detect more deadly cancers compared to more indolent ones, but not in much detail.
Klein and his colleagues' prostate cancer analysis included results and outcomes data from the final section of the company's large-scale case-control study, plus the limited findings from PATHFINDER, its first prospective intended-use trial that returned results to participants and their doctors. The investigators measured Galleri's detection statistics and compared them against clinical features of the cancers that were eventually diagnosed in the cohort. These clinical features included the cancer's stage and a prostate cancer-specific measure known as Gleason grade.
In the CCGA3 study there were 420 recently diagnosed prostate cancer cases with a median age of 65 years. Galleri detected 47 of 420, about 11 percent, and its cancer origin accuracy was 91.5 percent.
The MCED test detected no low-grade tumors, and only three favorable intermediate grade tumors in the CCGA group. It detected about 5 percent of unfavorable intermediate grade cancers and 31.9 percent of high-grade tumors.
Based on clinical staging, the test detected about 3 percent of stage I tumors, 5 percent of stage II tumors, 15 percent of stage III tumors, and 82 percent of stage IV tumors.
"It seems to be a fundamental property … that it's the more aggressive cancers that tend to secrete higher levels of cell-free DNA," Klein said in an interview. "So … there's a preferential detection for cancers that need to be treated. This is a potential issue for all screening technology because of the overdiagnosis of indolent cancers."
No Galleri skeptics have asked him about prostate cancer specifically, Klein said, but there has been an open question about overdiagnosis." Prostate cancer offers a great opportunity to try to address that question.
"It would be nice to be able to do secondary analysis on all the cancers," he added. "The challenge is that in PATHFINDER in particular, we don't have a lot of any single cancer." Even though prostate cancer is the most common cancer in men, there were only 18 prostate cancers diagnosed in the PATHFINDER cohort.
Overall, Klein stressed that the study should boost confidence that population screening with Galleri is unlikely to lead to overdiagnosis of low-grade tumors. "That takeaway is obvious," he said, "but what I think is more subtle, that's also important, is our ability in individuals who have a cancer signal detected to call cancer origin with very high accuracy" — over 90 percent in the CCGA cohort.
"Coupled with the observation that we preferentially detect high-grade cancers strongly suggests to me that if you do a Galleri test and you get a cancer signal detected with a predicted prostate origin, please don't ignore it ... Please do a diagnostic workup," Klein said.
Impact on outcomes
Amidst rising enthusiasm for the potential positive impact of MCED tests, other investigators continued to inject caution regarding unanswered questions about implied versus actual clinical value.
During the AACR session, Hilary Robbins, an investigator with the International Agency for Research on Cancer discussed her team's investigation of the possibility of using late-stage cancer incidence as an endpoint in randomized trials of MCED tests, rather than the gold standard of cancer-specific mortality, something that has been proposed by companies like Grail.
The fact that the Grail assay may avoid problematic detection of relatively innocuous disease is one thing. An unanswered question is whether it truly shifts detection of cancers to earlier stages, and in turn, improves patient survival or other outcomes.
There's ongoing debate in the screening community about whether fatality or overall survival is a reasonable endpoint for such exciting new technology, said Grail's Klein.
"Here's the issue," he explained. "It will take 12 to 15 years to do a clinical trial to get a mortality readout. Currently in the United States, even though we screen for five cancers, including prostate, we still lose 600,000 people a year due to cancer and mostly because we don't screen for the cancers that people die from. So, the downside of waiting 12 to 15 years, is 600,000 more deaths per year, every year, cumulatively. If you go for 12 years, that's 7.2 million deaths."
Based on the CCGA results, Klein said, Grail researchers have modeled that if Galleri were used on a population level on a repeated basis that it could reduce mortality by about 25 percent.
A second issue, he said, is that technology is evolving so rapidly that it's likely that some new technology, "something that's built on Galleri or something similar," could be available in a few years.
In that light, investing hundreds of billions of dollars in a trial with a potentially obsolete test technology "is a real issue," he said.
Given these challenges, Klein and his team did try to offer a picture of clinical impact in their analysis, comparing outcomes in the CCGA cohort to historical records from an ongoing real-world outcomes database.
On average, Galleri's non-detected cases had better overall survival, while detected cases had similar survival compared to what has been recorded in the SEER registry for a population matched to their age, stage, and Gleason score.
"We've been treating prostate cancer since the start of my career in the late 80s, with surgery and radiation," Klein said. "So, we know, for specific tumor characteristics, grade and stage, what the cure rates are for that cancer. And we don't have any reason to think currently that Galleri detects prostate cancers that are different than what we see by detection by PSA," Klein said.
"Again, we haven't proven that formally, but as an oncologist that makes logical sense," he added.
In other AACR sessions, Grail is featuring extended follow-up data from the CCGA study across cancer types, illustrating the prognostic differences between individuals in which Galleri yields a positive cancer detection result versus those who test negative. The company has previously shared similar analyses based on a three-year longitudinal follow-up, but updated those results at the AACR meeting to approximately four years, adding statistics to account for variation in cancer type and stage.
Company investigators reported on 782 participants with known cancer in the CCGA study who died during follow-up. In this group, the MCED test detected a cancer signal in 668 individuals (85 percent). Accounting for the effect of cancer type and stage, researchers reported that the non-detected group had better survival than the detected group, with a hazard ratio of 0.585.
Meanwhile, both the non-detected group and the detected group had better survival than their matched database comparators.
Robbins and her colleagues hoped to address the question from a more objective position, conducting a systematic review to assess whether stage-based endpoints could have provided a valid alternative for mortality in previous studies.
The team identified a set of 41 randomized controlled trials of cancer screening interventions published before 2023 ranging in size from 2,500 to 409,000 participants. Inclusion required, among other features, that the trials report numbers of cancer-specific deaths and cancer cases that could be split and compared between stages I-II versus III-IV. For each trial, the researchers calculated the percentage reduction in the cumulative late-stage cancer incidence and cancer-specific mortality between the intervention and control arms.
The correlation between a drop in cancer-specific mortality and a drop in stage III-IV cancer detection was strong for ovarian, lung, and breast cancer, but weak for colorectal cancer and virtually nonexistent for prostate cancers.
Of 12 trials demonstrating a statistically significant reduction in stage III-IV cancer incidence, five also showed reduced cancer-specific mortality.
Another study, presented by Delfi Diagnostics Chief Medical Officer Peter Bach compared projected cancer mortality reduction and public health impact for focused versus broad cancer screening models. Unlike MCED companies marketing or planning to market tests for a wide swath of tumor types, Delfi is focused on advancing blood-based screening assays for one indication at a time.
Bach's modeling suggested that focused screening, with tests tuned more toward sensitivity versus specificity, and employed in high-risk or high-mortality cohorts, would require testing only one-fifth of the people compared to broad screening with a test tuned to maximize specificity.
"It's something we have to keep in mind when we think about the tradeoffs. We may have to live with a tradeoff if we actually want the benefits," he said.
Beyond DNA
Meanwhile, another emerging Chinese firm, SeekIn, also presented case-control results at the meeting session, a precursor to the type of intended-use data that competitors are racing to produce.
SeekIn founder Mao Mao reported on the company's study of nearly 10,000 participants, including 2,003 cancer cases and 7,888 non-cancer controls, divided into a training set and five independent validation cohorts.
SeekIn's MCED test, unlike others featured at the meeting, is protein-based, utilizing a pre-defined panel of seven protein tumor markers — AFP, CA125, CA15-3, CA19-9, CA72-4, CEA, and CYFRA 21-1 — assayed using standard immunoassay technology. An algorithm, which the company calls OncoSeek, is designed to distinguish cancer cases from non-cancer cases based on a quantification of these seven markers, plus clinical information like sex and age.
Investigators reported that the overall sensitivity of OncoSeek was 51.7 percent with a specificity of 93 percent, with performance remaining consistent from the training set through the five validation cohorts that were gleaned from three countries (Brazil, China and the US) and using assays run on three different platforms (Roche, Luminex, and ELISA).
Although tumor origin prediction was lower than reported for other tests at about 65 percent, Mao proposed that this feature could still assist clinical diagnostic workup. Highlighting the starkest contrast between OncoSeek and its potential competitors, Mao said that the test costs only about $20, which could be transformative in ensuring equitable and global implementation of MCED tests.
"I come to all these cancer early detection meetings," he said. "I think it's always struck me that now we have these advanced tests in the US [and] we also have one in China. But if we look at other places… when I was in Kenya giving a talk, their cancer screening is limited to PSA. That's their reality."
To drive that point home, session moderator Wendy Rubinstein, program officer for breast and gynecologic cancer at the National Cancer Institute, noted that about 11 percent of Americans are living in poverty. "So, in some sense, at least many of our counties could be considered low- and middle- income countries," she said.
Finally, the AACR session also featured a relative newcomer to the space, Cambridge, UK-based ProteoType, which is advancing a method that profiles the immune response to disease development. The platform is designed to detect alterations in the ratio of immunoglobulins and albumin, as well as class-switching among immunoglobulins, by measuring the overall concentration of amino acid residues incorporated into proteins in patient plasma.
Cong Tang, the company's head of product development, reported at the meeting that ProteoType's assay detected 84 percent of cancers in a 92-patient cohort with a less than 1 percent false positive rate.
It was also able to differentiate cancer from potential confounders like autoimmune disease, infection, and neurodegenerative disease.