Prospective and comparative data needed
In a comment, Eric Klein, MD, emeritus chair of the Glickman Urological and Kidney Institute at the Cleveland Clinic explained that the “only accurate way to know how a test will perform in an intended-use population is to actually test it in that population. It’s not possible to extrapolate results directly from a case-control study.”
Cancers shed many different biologic markers into body fluids, but which of these signals will be best to serve as the basis of an MCED (multi-cancer detection test) that has clinical utility in a screening population has yet to be determined, he noted. “And it’s possible that no single test will be optimum for every clinical situation.”
“The results of this study appear promising, but it is not possible to claim superiority of one test over another based on individual case-control studies because of uncontrolled differences in the selected populations,” Dr. Klein continued. “The only scientifically accurate way to do this is to perform different tests on the same patient samples in a head-to-head comparison.”
There is only one study that he is aware of that has done this recently, in which multiple different assays looking at various signals in cell-free DNA were directly compared on the same samples (Cancer cell. 2022;40:1537-49.e12). “A targeted methylation assay that is the basis for Galleri was best for the lowest limit of detection and for predicting cancer site of origin,” said Dr. Klein.
Another expert agreed that a direct head-to-head study is needed to compare assays. “Based on this data, you cannot say that this method is better than the other one because that requires a comparative study,” said Fred Hirsch, MD, PhD, executive director of the Center for Thoracic Oncology, Tisch Cancer Institute at Mount Sinai, New York.
Metabolomics is interesting, and the data are encouraging, he continued. “But this is a multicancer early detection test and metabolism changes may vary from cancer type to cancer type. I’m not sure that the metabolism of lung cancer is the same as that of a gynecologic cancer.”
Dr. Hirsch also pointed out that there could also be confounding factors. “They have excluded inflammatory disease, but there can be other variables such as smoking,” he said. “Overall it gives some interesting perspectives but I would like to see more prospective validation and studies in specific disease groups, and eventually comparative studies with other methodologies.”
Study details
The authors evaluated if plasma and urine free GAGomes (free glycosaminoglycan profiles) deviated from baseline physiological levels in 14 cancer types and could serve as metabolic cancer biomarkers. They also then validated using free GAGomes for MCED in an external population with 2,064 samples obtained from 1,260 patients with cancer and healthy individuals.
In an in vivo cancer progression model, they observed widespread cancer-specific changes in biofluidic free GAGomes and then developed three machine-learning models based on urine (nurine = 220 cancer vs. 360 healthy) and plasma (nplasma = 517 cancer vs. 425 healthy) free GAGomes that were able to detect any cancer with an area under the receiver operating characteristic curve of 0.83-0.93 (with up to 62% sensitivity to stage I disease at 95% specificity).
To assess if altered GAGome features associated with cancer suggested more aggressive tumor biology, they correlated each score with overall survival. The median follow-up time was 17 months in the plasma cohort (n = 370 across 13 cancer types), 15 months in the urine cohort (n = 162 across 4 cancer types), and 15 months in the combined cohort (n = 152 across 4 cancer types).
They found that all three scores independently predicted overall survival in a multivariable analysis (hazard ratio, 1.29; P = .0009 for plasma; HR, 1.79; P = .0009 for urine; HR, 1.91; P = .0004 for combined) after adjusting for cancer type, age, sex, and stage IV or high-grade disease.
These findings showed an association of free GAGome alterations with aggressive cancer phenotypes and suggested that scores below the 95% specificity cutoff might have a better prognosis, the authors comment.
In addition, other analyses showed that free GAGomes predicted the putative cancer location with 89% accuracy. And finally, to confirm whether the free GAGome MCED scores could be used for screening, a validation analysis was conducted using a typical “screening population,” which requires at least 99% specificity. The combined free GAGomes were able to predict a poor prognosis of any cancer type within 18 months and with 43% sensitivity (21% in stage I; n = 121 and 49 cases).
Dr. Gatto believes that these results, as well as those from other studies looking at glycosaminoglycans as cancer biomarkers, will lead to the next steps of development. “But I speculate that this test could be most useful to assess in a cheap, practical, and noninvasive manner if a person at increased risk of cancer should be selected for cancer screening as part of established or emerging screening programs.”
The study was sponsored by Elypta. Dr. Gatto is listed as an inventor in patent applications related to the biomarkers described in this study and later assigned to Elypta, and is a shareholder and employed at Elypta. Dr. Hirsch reports no relevant financial relationships. Dr. Klein is a consultant for GRAIL and an investigator for CCGA and Pathfinder.
A version of this article first appeared on Medscape.com.