NEW YORK – AI-driven cancer genomics firm Freenome shared new data this week for its early cancer detection platform in pancreatic cancer, showing that a combination of epigenetic patterns and CA-19-9 levels could detect pancreatic cancers with between 82 percent and 100 percent sensitivity depending on stage.
Presented at the American Association for Cancer Research Special Conference on Pancreatic Cancer, the results, while preliminary, represent some of the firm's first early detection data in a non-colorectal tumor type.
Although Freenome still plans to commercialize its test initially for colorectal cancer, CEO Mike Nolan — who assumed the position this May after founder and initial CEO Gabe Otte vacated — said that the new data also mark the company's intention to eventually market the assay for the detection of multiple types of cancer.
Predictive content for tumor types other than colorectal cancer will be in a "latent position," available for gradual validation and iteration while the test is used clinically for CRC. "We think that's the really efficient way for us to go about that," Nolan said.
Freenome is one of several firms that ultimately have multi-cancer detection goals but have diverged in their strategy from companies like Grail and Thrive, which are pushing for broad multi-cancer screening from the get-go.
"With colorectal cancer, the clinical pathways and the reimbursement and medical policy frameworks are so well established that we have a really great opportunity to accelerate our time to clinical impact and also, quite frankly, our time to clinical revenue, given the nature of the company and how it's capitalized," Nolan explained.
"We're going to form baskets of cancer types that can be ordered as baskets or individually, and we're going to do that in a very mechanistic stepwise approach," he said. "CRC is first, and then from among those nine or 10 we'll be looking at how screening pathways look today … and see if we can make a difference to make that more accessible, more actionable, and ultimately reduce mortality."
The new data in pancreatic cancer, Nolan added, illustrate how that approach will progress, essentially demonstrating how the firm's central multiomics machine learning platform will be applied to various other cancer types.
Pancreatic cancer, unlike colorectal cancer, doesn't have an established screening paradigm. But because of its aggressiveness and tendency to be diagnosed at late stages, the development of effective screening tools is recognized as a significant unmet need.
At the AACR special conference, investigators reported on a retrospective study of 75 individuals, split between those with confirmed stage II-IV pancreatic cancer and a control set of both healthy individuals and patients with benign pancreatic abnormalities.
They used the Freenome machine learning platform to train an algorithm that incorporated both DNA methylation patterns and levels of CA-19-9, a pancreatic protein biomarker currently only used to track therapy response.
The resulting classifier achieved an overall sensitivity of 93 percent at a nominal specificity of 96 percent. Breaking the cohort down by stages, the multiomics approach achieved a sensitivity of 82 percent, 89 percent, and 100 percent, in stages II, III, and IV, respectively.
Although the study cohort was much smaller than what has been described by some competing firms, the performance that the company saw was competitive.
For example, in a recent breakdown of its test performance across different cancer types, published in Annals of Oncology, Grail noted an approximate 84 percent sensitivity at 99 percent or greater specificity for pancreatic cancer.
As blood-based genomic technologies have begun to vie for a place in cancer screening, clinicians have cautioned that screening tests in rarer cancer types like pancreatic cancer face challenges in translating their performance into actual value.
In a commentary on blood-based cancer screening in Clinical Chemistry and Laboratory Medicine this April, the authors wrote that "the [positive predictive value] of a test and thus its potential clinical utility critically depend on the prevalence of the disease within the population undergoing investigation."
In other words, the less prevalent a disease, the lower the PPV, even at very high sensitivity and specificity.
"In pancreatic cancer, the clinical pathways [for screening] are not nearly as well characterized or established as colorectal cancer, so there will be more work to do there," Nolan said, adding that Freenome believes that it can develop a test that performs at a level that makes it compelling for use at "the forefront of clinical care."
Randall Brand, a professor at the University of Pittsburgh School of Medicine and co-author of the study presented at AACR, said in a statement that his team is now working to further validate its findings, which, in such a small cohort, should be viewed as preliminary.
Nolan added that the company is working more broadly on prospective sample collection for studies covering about 16 cancer types, including breast and prostate cancer in addition to pancreatic cancer.
Colorectal cancer progress
Nolan said the company has been encouraged that it is taking the right path in launching clinically for colorectal cancer only in light of the United States Preventive Services Task Force's recent move to lower the recommended age for average-risk CRC screening to 45.
The firm also heralded the Centers for Medicare and Medicaid Services' move in January to preemptively finalize a national coverage determination for blood-based colorectal cancer screening tests meeting certain criteria.
According to Nolan, those criteria are not expected to be a high bar for Freenome's assay. "We believe we can do much better and we will," he said. Some of the company's most recent data, for example, represented a sensitivity of 94 percent at a specificity of 94 percent for stage I and stage II tumors.
In its ongoing registrational trial to support US Food and Drug Administration approval, Freenome has now completed data collection for over 22,000 subjects and expects to conclude the study by the end of this year or early next.
Nolan said that the company continues to work closely with the agency and is anticipating a commercial launch in late 2023 or early 2024.