What new approaches exist for early cancer detection?
Blood tests for early cancer detection show promising results and are good at ruling out cancer in many cases, but they still miss a substantial proportion of early-stage cases, and that is the main caveat to keep in mind.
Blood tests for early cancer detection are being developed rapidly. Many studies combined show that such tests are good at reassuring healthy people: only rarely does someone without cancer receive a wrongly positive result. The downside is that the tests still miss a notable share of real cases. Of all the traces measured in the blood, cell-free DNA (cfDNA) performs best. These are tiny fragments of genetic material that cancer cells release into the bloodstream.
Several commercial tests try to find many different types of cancer at once from a single blood draw. All of them are good at clearing healthy people, but their ability to actually find cancer varies widely. In the early stages, where treatment offers the best chance of recovery, the tests perform worse than they do for advanced cancers. Research showing whether these tests genuinely lead to fewer deaths or a better quality of life does not yet exist.
Using artificial intelligence to analyse cfDNA makes the results slightly more accurate: somewhat fewer false positives and somewhat fewer missed cases. But the results vary considerably between countries and between groups of people studied, so those numbers cannot simply be applied everywhere. Researchers are also looking at other traces in the blood, such as circulating microRNAs. These look promising, but the evidence for them is still thin and comes largely from older studies.
A completely different approach is the use of artificial intelligence in tissue examination: not via blood, but via tissue that has already been taken. AI can automatically identify tumours in that tissue and predict how someone will respond to treatment. This is not yet widely used, because access to the necessary data and regulatory requirements still present barriers. There is also an experimental method that analyses patterns in dried droplets of body fluid, such as blood or urine. The figures published for this method look impressive on paper, but the studies are unreliable and whether it works in practice is entirely unclear.
This answer draws on several review studies, including many studies combined, covering 17 and 13 studies respectively on liquid biopsy and machine learning applied to cfDNA, a review of 36 studies on multiple commercial tests funded by the UK's health research body, and a review on AI in pathology. The reviewers of the tests that screen for multiple cancers at once judged most of the underlying studies to be at high risk of bias. For the dried-droplet pattern method, variation between studies is exceptionally large. The researchers themselves emphasise that for none of these methods have large studies been completed that measure death rates or quality of life, the outcomes that matter most to patients.