New AI tool that can predict death to be trialled on heart patients

An artificial intelligence (AI) tool to help doctors identify high-risk heart patients will soon be rolled out on a trial basis in England after a study found it can accurately predict the risk that someone will die in the years after a heart scan.

The global research team led by Imperial College London trained their AI model, known as AI-ECG risk estimation or AIRE, on millions of results from electrocardiograms (ECG), a common medical test that records electrical signals within and between the heart’s chambers. It is typically used to diagnose heart attacks and other irregularities.

The goal was to identify nuanced patterns that could mean someone is at high risk of health problems or death.

Put to the test, the model predicted the likelihood of death in the decade following an ECG – and it was correct 78 per cent of the time.

“We believe this could have major benefits for the NHS, and globally,” Dr Fu Siong Ng, a cardiac electrophysiology researcher at Imperial College London who worked on the project, said in a statement.

The system can also predict heart attacks, heart failure, and heart rhythm problems, and the researchers said it could be rolled out across the National Health Service (NHS) within the next five years.

Trials using real patients are already planned for several sites in London and are expected to begin by mid-2025.

They will evaluate the benefits of the model using patients from outpatient clinics and hospital medical wards.

Potential for AI to improve heart health

AI-powered ECGs have already been used to diagnose heart diseases, but they are not part of routine medical care and have not yet been used to identify a specific patient’s risk levels.

“This could take the use of ECGs beyond what has previously been possible, by helping assess risk of future heart and health problems, as well as risk of death,” said Bryan Williams, chief scientific and medical officer at the British Heart Foundation, which funded the study.

The researchers, who published their results in the Lancet Digital Health journal, said the predictions where the AI was wrong could be because of other unknown factors, such as whether the patient got additional treatment or died unexpectedly.

But they stressed that the model could still generally pick up subtle changes in the heart’s structure, which can serve as a warning sign for illness or death but that doctors might miss.

“We cardiologists use our experience and standard guidelines when we look at ECGs, sorting them into ‘normal’ and ‘abnormal’ patterns to help us diagnose disease,” said Dr Arunashis Sau, an academic clinician at Imperial College London who led the new research.

“However, the AI model detects much more subtle detail, so it can ‘spot’ problems in ECGs that would appear normal to us, and potentially long before the disease develops fully,” Sau said.

Sau said more research is needed from hospitals and other healthcare settings to determine the model’s future role in diagnosis and treatment, but that patients with other health issues could also likely benefit because other diseases, such as diabetes, also tend to affect the heart.

ADVERTISEMENT

Ng agreed. “This could have a positive impact on how patients are treated, and ultimately improve patient longevity and quality of life,” he said.