Blood test shows promise for predicting cancer, tissue of origin ahead of histologic diagnosis

A cell-free DNA multicancer early detection test demonstrated an ability to predict cancer and the tissue of origin in individuals with clinical suspicion of cancer, according to results of the Circulating Cell-free Genome Atlas study presented at the virtual American Association for Cancer Research Annual Meeting.

The high specificity of this test suggests the false-positive rate may be comparable in populations with average vs. higher risk for cancer, researchers noted.

“Many cancers are detected too late,” David D. Thiel, MD, chair of the department of urology at Mayo Clinic in Jacksonville, Florida, said during a presentation. “A large percentage of breast, lung and colorectal cancers are detected simultaneously with metastatic disease. Detection of cancers prior to the development of metastatic disease can significantly increase 5-year survival.

“In the case of lung cancer, the detection of cancer prior to the development of metastatic disease can lead to a 10-fold increase in survival,” he added. “Currently, screening paradigms are available for five cancer types: colon, lung, cervical, breast and prostate. Adherence to these paradigms can often be suboptimal, and the absence of screening programs for other lethal cancers has contributed to more than half of all cancer-related deaths.”

The multicenter, case-control, observational Circulating Cell-free Genome Atlas study collected deidentified biospecimens from 15,254 people in North America with cancer (56%; n = 8,584) or without (44%; n = 6,670). The study was structured in three parts: discovery, further analysis through training and validation sets, and further validation.

Researchers reported previously that the targeting methylation assay, made by GRAIL Inc., detected and localized more than 20 cancers types with greater than 99% specificity among individuals with cancer in the study.

Thiel and colleagues analyzed the performance of the test in predicting the presence or absence of cancer and tissue of origin among a subgroup of participants with clinical suspicion of cancer but no pathologic diagnosis or treatment at enrollment.

Researchers collected plasma cell-free DNA from blood samples and subjected it to targeted methylation sequencing. They then divided the samples into a training set and an independent validation set to train and validate a machine learning classifier to assess cancer and predict tissue of origin.

Presence of cancer was confirmed by evaluating a pathologic specimen.

Participants evaluated for suspicion of cancer were classified as confirmed cancer (more than 20 cancer types; n=164 in training, n=75 in validation) or confirmed noncancer (n=49 in training, n=15 in validation).

Researchers excluded stage I renal cancers, which represented 20% of cancers in the subset and are difficult to detect because of low tumor cell-free DNA fraction at that early stage..

Results showed that, among the confirmed noncancer group, all training and validation samples had been accurately predicted as noncancer, reflecting 100% specificity.

Among those in the confirmed cancer group, cancer detection across all stages was 40.2% (95% CI, 32.7-48.2) in training and 46.7% (95% CI, 35.1-58.6) in validation. Additionally, detection across stages was 50.4% (95% CI, 41.5-59.2) in training and 59.3% (95% CI, 45.7-71.9) in validation.

Among cancers of stage II and above, detection was 70.7% (95% CI, 59.6-80.3) in training and 78.9% (95% CI, 62.7-90.4) in validation.

For detected cancers, the test predicted tissue of origin at rates of 93.9% in training and 100% in validation. Among those with a tumor of origin call, the rate of accuracy was 85.5% (95% CI, 74.2-93.1) in training and 97.1% (95% CI, 85.1-99.9) in validation.

“This multicancer, early-detection test is being validated in population-scale clinical studies to ensure robust clinical validation,” Thiel said. “This is intended to be used alongside guideline-recommended screening, not replace it.” – by John DeRosier

Reference:

Thiel DD, et al. Abstract CT03-01; Presented at: AACR Annual Meeting; April 27-28, 2020 (virtual meeting).

Disclosures:

Grail Therapeutics funded this study. Thiel reports to relevant financial disclosures. Please see the abstract for all other researchers’ relevant financial disclosures.