NEW YORK – Investigators this week shared some of the first formal data on detection rates in undiagnosed individuals for two of the most prominent multi-cancer blood-based cancer screening tests being developed by commercial firms Grail and Thrive Earlier Detection.
In an online presentation for the American Association for Cancer Research annual meeting, researchers working with Grail shared an analysis of detection results in a subgroup of 300 subjects from the larger CCGA (Circulating Cell-free Genome Atlas) study. The investigators reported, among other things, that the test maintained 100 percent specificity: None of the suspicious cases who were cleared by their doctors of having cancer tested positive with the firm's assay.
Grail's test uses targeted bisulfite sequencing to assess methylation status across a panel gleaned from initial genome-wide profiling of the CCGA cohort to define epigenetic patterns that can distinguish cancer from non-cancer controls.
As part of the AACR meeting, David Thiel, chair of the Mayo Clinic Florida department of urology, reported on his colleagues' analysis of the subset of CCGA subjects who were recruited to the study while they were being evaluated for suspicion of cancer. These cases were then either confirmed as a cancer-positive or a cancer-negative via pathologic testing of tissue samples.
Although still retrospective, the analysis of this subgroup gave researchers the opportunity to glimpse the assay's performance in a population reflective of what would actually be used in in clinical practice, as opposed to the contrived case-control matchup of the overall CCGA study.
According to Thiel, the Grail test detected cancer in 40 percent to 47 percent of the individuals who were confirmed as having cancer, depending on whether they were in the trial's training or validation sets. It also correctly predicted no cancers in 64 individuals who were confirmed not to have cancer.
Excluding stage I renal cancers, which comprised 20 percent of participants in the subset, but which seemed to present the most challenge to the method, detection across stages rose to between 50 and 59 percent. For stages II and above, the test correctly detected up to 79 percent of individuals confirmed to have cancer.
Notable, according to Thiel, is the fact the performance numbers across the board in this undiagnosed subgroup hewed closely to what Grail has calculated across its larger CCGA analyses — a consistency that helps raise confidence that sensitivity and specificity won't collapse as the company works to replicate its findings prospectively.
During the same AACR session, investigators from Johns Hopkins University also discussed findings from their prospective, interventional study of a different blood-based screening method — an earlier version of their CancerSEEK method that is now being commercialized by Thrive Earlier Detection. This study examined nearly 10,000 women aged 65 to 75 who were part of the Geisinger Health System.
Although Grail is pursuing various prospective trials it has not yet reported data from these efforts, making the Hopkins team the first to share this level of data on a multi-cancer blood screening test.
The JHU researchers reported that the CancerSEEK assay successfully detected a number of incipient cancers across the 10,000-person DETECT-A study cohort, including high-risk types like ovarian cancer, which is typically diagnosed at an advanced stage with poor prognosis. A simultaneous publication describing the group's results in detail also appeared yesterday in Science.
While Grail has focused on methylation, the Hopkins approach combines mutation detection with measurement of cancer-associated protein signals. Speaking during the AACR session, JHU professor of oncology and pathology Nickolas Papadopoulos reported that, in total, 96 cancers were diagnosed during his team's 12-month analysis. Of these, 26 were detected by the group's genomic/proteomic blood test, including nine lung cancers, six ovarian cancers, and two colorectal cancers.
Another 24 cancers were detected in the cohort by standard screening like mammography and colonoscopy, but not in blood, and 46 others were diagnosed by other means.
"Blood testing identified cancers in 10 different organs, seven of them in [types] that do not have standard-of-care screening," Papadopoulos said in his presentation. "More important, 65 percent of them were localized, or regional, with a higher chance of successful treatment with intent to cure," he added.
Overall, the group calculated that the positive predictive value of the blood test alone was 19 percent. When combined with PET CT, this rose to 41 percent. "Also, both mutations and proteins contributed to the identification of screen-detected cancers," Papadopoulos said.
Introducing the AACR session on liquid biopsy advances, University of Turin professor Alberto Bardelli highlighted the presentations as some of the first hints of the type of data necessary to actually advance liquid biopsy into the cancer screening space.
"The idea that circulating tumor DNA [and] other alterations present in circulation can be a proxy of the development of tumors is not novel," he said. But while the application of these concepts to early cancer detection has been an area of great excitement, "only recently has it become possible to test [this] in longitudinal and prospective clinical trials."
As tests like these are further validated, the main hurdles for their creators will be cementing that the ability to detect otherwise cryptic tumors presents more of a benefit than a harm — both to individuals and to the larger socio-economic healthcare architecture.
"If screening is [at] the top of the heap in terms of the potential impact, it's also the greatest challenge," said the University of British Columbia's David Huntsman, serving as a discussant for the two presentations.
One major fear is that tests that detect genomic and other signals of cancer in the blood will not have enough specificity to avoid returning significant numbers of false positives.
According to the Hopkins authors, an important aspect of their CancerSEEK study was that the results paint an encouraging picture in this regard. They calculated that the combination of a blood test followed by PET-CT imaging was 99.6 percent specific.
Similarly, in Grail's sub-cohort analysis, the company's test was 100 percent specific. According to Thiel the smaller sample size means this result should be taken cautiously on its own, but at least provides initial evidence that specificity in undiagnosed subjects probably won't suffer significantly relative to the 99 percent the company could achieve across the larger case-control CCGA study.
A related issue is what to do with positive results and data from both studies also spoke to this notion of feasibility. In the DETECT-A study, the Hopkins investigators designed their protocol to require a defined set of steps that must be taken prior to initiating a diagnostic workup for cancer after a positive liquid biopsy result.
Physicians could only refer individuals for the final component, a full-body PET CT scan, if they couldn't find a non-cancer-related cause for an abnormal result in the previous designated workup steps. According to the authors, this seemed to work, as women in the study did not receive large numbers of futile, invasive follow-up tests.
Only 1 percent of participants underwent PET-CT imaging based on false positive blood tests, and a fraction of a percent underwent a futile invasive diagnostic procedure, according to the authors. Blood-based screening also did not seem to influence them away from adhering to existing standard-of-care screening practices like mammography.
To allay fears about uncertain follow-up procedures and diagnostic odysseys Grail has worked to enable its methylation assay to report tumor localization, or origin prediction. The AACR presentation noted that tumor-of-origin prediction was possible for about 94 percent of the subjects, with an accuracy of 85 percent in the training set and 97 percent in the validation set.
A third hurdle for blood-based cancer screening is the worry that tests like these might be confounded by, or even preferentially detect cancers that don't need to be found — in other words, that they will lead to overdiagnosis and wasteful overtreatment of tumors that pose no actual risk to patients.
This will largely be a question for future prospective studies to answer though current data provides some hints. In DETECT-A, the authors reported, for example, that several blood-test-identified cancer cases were serious diagnoses including late-stage ovarian cancer.
Grail investigators have also shared separate analyses at earlier meetings indicating that tumors detected by its test tend to skew toward more agressive cases with worse outcomes, while the class of cancers that the assay misses is enriched for tumor types or classes that are more likely to be indolent.
"Ultimately both studies are showing we're moving in the right direction, but we're not there yet," Huntsman argued. "We need to show proof of overall survival and this will take large prospective studies — such as follow up from the DETECT-A and the other studies described by Dr. Thiel — to really get us to the point where these tests could be adopted."
Grail, the JHU researchers and spinout Thrive are in the midst of these efforts. In Grail's case, its first prospective trial data is still yet to be released but the company has several observational studies that have been fully recruited as well as a new trial returning results that began this January.
The DETECT-A team plans to continue to follow its nearly 10,000 study participants, including those with positive and negative test results, for a total of five years.
"Longer follow-up will undoubtedly reveal more undetected cancers, thereby increasing the number of false negatives. It could also reveal patients with cancer with positive blood testing but in whom diagnostic PET-CT failed to identify a cancer, thereby decreasing the number of false positives," the authors wrote.