Oxford develops AI tool to forecast heart failure five years ahead

Researchers at Oxford have unveiled an AI tool that can estimate the risk of heart failure at least five years before it develops. The system works on data from a routine cardiac CT scan and, according to a study of patients in England, predicts that risk with 86% accuracy.

How it works

The tool does not analyze the heart muscle itself, but changes in the fat tissue around the heart. The Oxford researchers start from the premise that this tissue acts as an early "sensor" of trouble: when the heart is inflamed or gradually losing normal function, the structure and composition of the surrounding fat also change. A doctor cannot see these microscopic signals on ordinary images, but the algorithm can pick up texture patterns and turn them into a numerical risk score.

An important detail is that the system needs no rare examinations or separate complex test. It uses data from a standard cardiac CT, which British hospitals already perform on patients with chest pain and suspected coronary artery problems. The algorithm calculates absolute risk automatically, without manual input, and then gives the doctor a guide: who needs closer monitoring, whose treatment approach should be changed earlier, and who does not need to be moved to stricter control.

What the study showed

The team trained the model on anonymized CT results from more than 59,000 people in England, then tested it separately on another 13,424 patients. In total, the study included over 72,000 people from nine NHS hospital trusts, who were followed for ten years after their scans. The results were published in the Journal of the American College of Cardiology.

This design made it possible to test not only agreement with a patient's current condition but also the algorithm's real ability to predict the development of heart failure several years ahead.

• More than 72,000 patients in the overall sample
• 9 NHS hospital trusts in England
• 86% accuracy for five-year risk prediction
• Patients in the top risk group were 20 times more likely to develop the condition
• In the highest-risk group, the probability of developing heart failure within five years was about 25%

According to the researchers, patients in the top risk group may benefit the most: in them, heart failure is often detected too late, after serious damage to the heart muscle has already occurred. Now a doctor gets not a vague suspicion but a concrete risk estimate that can be used when planning monitoring and treatment. This matters especially for a condition that more than 60 million people live with worldwide, and over a million in the UK.

"This tool will allow doctors to decide more precisely who needs the most intensive monitoring," says study lead

Charalambos Antoniades.

The next stage of rollout

The Oxford team is already seeking regulatory approval to build the tool into the routine CT analysis workflow in NHS radiology departments. The idea is pragmatic: rather than creating a new patient pathway, add another layer of interpretation to examinations that already exist. According to the researchers, about 350,000 cardiac CT scans are performed in the UK every year, so even partial adoption could give the health system a large stream of early signals.

Separately, the developers are working to make the algorithm applicable not only to specialized cardiac CT but to any chest CT done for another reason. If the adaptation for lung CT is indeed ready in the coming months, it will widen coverage: heart failure risk could be noticed incidentally, even when a person is scanned because of breathing problems or as part of lung screening.

What this means

Systems like this show where medical AI is really heading: not toward abstract promises, but toward integration into existing clinical processes. If the Oxford tool gains approval and enters routine practice, an ordinary CT scan could serve not only as an image for the current complaint but also as an early warning of a serious disease that was often noticed too late, at the stage of hospitalization or pronounced symptoms.