We have discussed “liquid biopsy” for cancer screening several times in past news items. All these liquid biopsy techniques look for cancer cells or more commonly cell-free DNA in blood to try to discriminate people with cancer from those without. Current methods look for some combination of mutation patterns or DNA methylation patterns to make this discrimination but so far have proved susceptible to unacceptably high false negative and false positive test results.
A new approach reported in the journal Nature
is showing early promise of improved performance. Dr. Victor Velculescu, MD, PhD, professor of oncology at Johns Hopkins Kimmel Cancer Center, Baltimore, Maryland, explained that; "Our approach is very global — it looks broadly across the genome at what we call fragmentation profiles of cell-free DNA (cfDNA), and by analyzing these fragmentation profiles in the bloodstream, we can identify abnormalities genome-wide and avoid many of the pitfalls that have to do with specific mutations or other factors."
The researchers used this method to analyse the fragmentation profiles of 236 patients with breast, colorectal, lung, ovarian, pancreatic, gastric or bile duct cancer and 245 healthy individuals. A machine learning model that incorporated genome-wide fragmentation features had sensitivities of detection ranging from 57% to more than 99% among seven cancer types at 98% specificity, with an overall area under the curve value of 0.94. Fragmentation profiles could be used to identify the tissue of origin of the cancers to a limited number of sites in 75% of cases. The cancers studied were early stage breast, colorectal, lung, ovarian, pancreatic, gastric, or bile duct cancers. The specificity of the test in this limited study was 98% meaning that only four healthy controls received false-positive test results.
The researchers have called their test DELFI (DNA Evaluation of Fragments for Early Interception). While these results are promising, they need to be confirmed in a much larger clinical trial where the clinical status of the subjects is unknow prior to testing. In the longer term, studies will have to determine whether test results and subsequent treatment does result in clinical benefit to patients.
On the positive side, the researchers say that the test uses existing DNA sequencing technology and so should not be too expensive. Since the technique uses machine learning, it is also likely that the performance of the test will improve as more patients are tested and the learning database expands.
Screening tests like this will always perform better in subjects whose pre-test probability of having cancer is higher than the general population. Thus, it is likely to be used first in people with genetic syndromes that put them at higher risk of having cancer or perhaps in smokers who are at higher risk of developing lung and some other cancers.