That liquid biopsies – blood tests that can detect the presence and track the treatment of solid tumours via the DNA or cells they shed into the bloodstream – hold potential is not in doubt. But the developers of these tests tend to dripfeed small subanalyses of large trials rather than presenting more sweeping datasets, which can make it hard to evaluate how useful these tests might truly be.
Yesterday at the virtual AACR meeting one developer achieved something more concrete. A prospective, interventional study used the blood test from Thrive Earlier Detection to screen for cancer in apparently healthy people – the first time such a thing has been achieved.
The Detect-A trial, conducted by researchers from Johns Hopkins University and health services group Geisinger, enrolled more than 10,000 women with no prior history of cancer. It evaluated a version of CancerSeek, Thrive’s liquid biopsy, that was developed in 2016 – the test has been tweaked since, though no version is yet on sale.
Among 10,006 seemingly healthy women, 96 subjects developed cancers: 26 of these were identified by Thrive’s test, 24 were identified by standard-of-care screening, such as mammography for breast cancer, and 46 were first identified by symptoms or other means. The test allowed treatment of some of the patients: 12 of the tumours detected by CancerSeek were surgically removed.
The sensitivity of CancerSeek was 27.1% across all cancers. This rose to 52.1% when it was used in conjunction with standard-of-care testing, allowing the group to claim a doubling in the number of cancers that were first detected by screening.
Specificity was better: Thrive’s test alone had a 98.9% specificity, and when combined with imaging this hit 99.6%. In an asymptomatic population with a cancer incidence of approximately 1%, CancerSeek plus imaging had a positive predictive value of 40.6%; this may not sound like much, but according to Thrive this is “considerably higher than the PPV of existing single-cancer screening tests available today”.
This is only a start, and higher sensitivity would be necessary for a viable commercial product – presumably Thrive is working on making newer versions of CancerSeek more sensitive. But Thrive, which emerged from stealth less than a year ago, has at least shown some real, prospective data.
This is more than can be said for Grail, whose own AACR data look rather like small potatoes in comparison. The Illumina spinout presented a pre-specified cut of the vast CCGA study, in which the test was used to identify patients with cancer. The test was trained on 164 patients known to have cancer and 47 without, and then validated in 75 patients with cancer and 15 without.
All samples from cancer-free patients were correctly predicted as non-malignant, giving a 100% specificity rate. Sensitivity was less impressive: only 35 of the validation set samples from patients with confirmed cancer were correctly picked, giving a sensitivity of 46.7%.
Grail offered a defence of a sort in the shape of a subset of a subset. Claiming that renal cancers were overrepresented and subject to poor detection at early stages owing to low tumour DNA fraction, the group excluded these patients, allowing the detection rate to reach 59.3% (35 out of 59 patients). For cancers of stage II and above, sensitivity was 78.9%.
The test was then used to identify the tissue of origin of those tumours it did detect. It was able to come up with an answer for 93.9% of samples in the training set and 100% in validation. Of those, 85.5% (53/62) in the training set, and 97.1% (34/35) in the validation set were right.
Grail’s data comes from a retrospective study – the patients’ cancer status was already known. It is interesting enough as an indication of Grail’s test’s accuracy, but clearly has less real-world relevance than Thrive’s data.
It might also serve as a step towards FDA approval of CancerSeek. No liquid biopsy is currently approved, though some, including Guardant Health’s Guardant360, are on sale in the US as lab-developed tests. Thrive has said it intends to pursue FDA approval; a 3,000-strong trial of the new version of CancerSeek is soon to conclude and the company aims to launch a larger, potentially registrational trial next year.
Grail has made no such declaration, which perhaps explains why it is moving more slowly.