AI flags heart‑failure years early
Oxford researchers developed an AI tool that can predict heart‑failure risk up to five years before onset with about 86% accuracy in a study of 72,000 English patients. Early prediction like this could enable targeted monitoring and earlier intervention, though the result remains a study‑level finding. (theguardian.com)
Heart failure usually shows up late, after the heart muscle has already been damaged enough that it cannot pump blood around the body properly. Oxford researchers are trying to catch it much earlier by reading clues from routine computed tomography scans, which are the detailed X-ray pictures many hospitals already use for chest pain workups. (rdm.ox.ac.uk) The clue is not the heart muscle itself at first. It is the fat wrapped around the heart, which changes texture a bit like grass changing color when the soil underneath is sick. (rdm.ox.ac.uk) Inflammation is the body’s alarm system, and long-running inflammation can alter tissue before a scan looks obviously abnormal to a doctor. The Oxford team says the fat around the heart acts like a sensor, picking up those early disease signals years before heart failure becomes visible in the usual way. (rdm.ox.ac.uk) That is where the artificial intelligence comes in. The model looks for tiny textural patterns in that heart fat on cardiac computed tomography scans, and those patterns are too subtle for the human eye to spot on routine imaging. (rdm.ox.ac.uk) The new result is that the system predicted who would develop heart failure within five years with about 86 percent accuracy. The study used data from more than 70,000 people, reported as about 72,000 patients, across nine National Health Service trusts in England who were followed for a decade after their scans. (rdm.ox.ac.uk) (sciencedirect.com) The gap between high risk and low risk was wide. People the model placed in the highest risk group were 20 times more likely to develop heart failure than people in the lowest group, and about one in four of the high-risk group developed it within five years. (rdm.ox.ac.uk) This was built around scans that already happen in large numbers. Oxford says about 350,000 patients a year are referred for a cardiac computed tomography scan in the United Kingdom, usually to check for fatty plaque in the coronary arteries rather than to forecast heart failure. (rdm.ox.ac.uk) So the pitch is not a brand-new machine in every hospital. It is adding a second read to an existing scan, where software turns an image taken for one reason into a warning about a different problem that may not appear for years. (rdm.ox.ac.uk) The paper was published on April 8, 2026 in the Journal of the American College of Cardiology, and the work was led by Professor Charalambos Antoniades at the University of Oxford. The team says it is now working toward regulatory approval and wants to expand the approach beyond dedicated cardiac scans to other chest computed tomography scans as well. (sciencedirect.com) (rdm.ox.ac.uk) It is still a study result, not a standard hospital test. But if the finding holds up in real clinics, a scan done in 2026 for chest pain could also tell a doctor that the same patient has a serious chance of heart failure by 2031, while there is still time to watch more closely and treat earlier. (theguardian.com) (rdm.ox.ac.uk)
