Johns Hopkins has developed a lung cancer blood test
Powered by artificial intelligence, a new lung cancer blood test developed at Johns Hopkins, combined with other metrics, correctly identified 94% of cancer cases in almost 800 patients.
The lung cancer blood test, published in Nature Communications, searches for tiny fragments of DNA released by the tumor cells. The AI looks for patterns in this shattered DNA, rather than looking for specific pieces of cancer DNA like other blood tests in development, New Atlas explained.
Lung cancer kills the most people in the world, the authors note, “largely due to the late stage at diagnosis where treatments are less effective than at earlier stages” — and lung cancer rates are increasing, worldwide.
“We believe that a blood test, or ‘liquid biopsy,’ for lung cancer could be a good way to enhance screening efforts, because it would be easy to do, broadly accessible, and cost-effective,” study first author Dimitrios Mathios said.
The DNA difference: Blood tests for cancer typically focus on finding pieces of mutated tumor DNA.
Hopkins’ lung cancer blood test is based instead on cancer cells being much more chaotic than healthy cells when it comes to their DNA. Healthy cells pack their genetic code like a well organized suitcase, the researchers said. When cancer cells die, it’s like their sloppily thrown together suitcase has flown open, strewing all the pieces about in a way that is different from the organized, healthy cells.
To take advantage of this sloppy signature, the researchers developed a technique called DELFI.
DELFI uses machine learning to spot the patterns of DNA pieces that are associated with tumors, including their size and how many of them are present, and score them based on how likely they are to indicate cancer.
“DNA fragmentation patterns provide a remarkable fingerprint for early detection of cancer that we believe could be the basis of a widely available liquid biopsy test for patients with lung cancer,” Rob Scharpf, associate professor of oncology at Johns Hopkins, said.
Searching for lung cancer’s signature: The researchers put DELFI to the test using blood samples from 796 patients in the U.S., Denmark, and the Netherlands. When combined with a clinical evaluation of risk markers, the use of a protein biomarker, and CT scans, the model correctly flagged 94% of lung cancers. That dipped to 91% for early stage lung cancer, and climbed to 96% for more advanced cancers.
But there’s a tradeoff to be made in making sure you catch as many cancer cases as possible and finding false positives. The test had an 80% specificity rate, meaning that 20% of people without lung cancer would also incorrectly test positive.
The test uses machine learning to spot the sloppy patters of scattered tumor DNA.
With such a high false positive rate, screening everyone would result in false positives vastly outnumbering real cancers — requiring still further exams to sort it out. However, the model could be tweaked to raise the bar for a positive result, which would miss more real cancers but also potentially make the result more useful.
As the researchers note, DELFI will need to be evaluated in a large-scale clinical trial before it is ready for broader use. They plan to test DELFI in over 1,000 patients across the U.S., including healthy patients, patients with lung cancer, and those with other types of cancers as well.
We’d love to hear from you! If you have a comment about this article or if you have a tip for a future Freethink story, please email us at [email protected]