A drop of blood from a fingertip is like a photograph. One is just a snapshot, but collect them over years and you get a movie of your health. That's exactly the kind of movie the authors analyzed: they took results from routine complete blood counts that people had taken again and again, and tracked how the numbers shifted in those who later developed cancer, cardiovascular disease, or caught an infection.
It turns out each disease leaves its own characteristic 'trace' in the blood's dynamics. For example, before a tumor appears, some cell types start rising faster, others fall, and this pattern repeats across different patients. Machine learning can spot such patterns even when a person feels perfectly healthy. Moreover, the main signal comes from complete blood count indicators — the ones measured at any clinic. Expensive biochemical tests add little: their contribution is noticeable only for the specific diseases they were designed for.
This approach requires no new devices or complex procedures. Just don't throw away old test forms, and the algorithm can warn of trouble months or even years before symptoms. It's a step toward medicine that doesn't wait for disease but sees it coming in ordinary numbers.
In the same person, platelet levels can fluctuate by 20% within a week without any illness, so a single test often misleads, while a series of snapshots reveals the true trend.