A new AI model has been developed to predict patients’ risk of heart disease, and even their risk of death, when using electrocardiograms (ECGs), recent research in Lancet Digital Health reveals. The model can predict potential heart issues like arrhythmias, heart attacks, and heart failure, as well as when someone will die from heart-related or non-heart related causes. Researchers at Imperial College London found the AI model correctly identified risk of death within ten years of an ECG in 78% of cases. The model’s inaccuracy in the remaining 22% of cases could be down to “unknowable” issues like the patient receiving treatment, or unexpected death.
Trained on ECG data from millions of patients
ECGs record the heart’s electrical signals to detect abnormalities with the heart rate or rhythm. The AI model was trained to “read” ECG data from millions of international patient records and spot patterns in the electrical signals, which allows it to accurately predict which patients develop worse diseases, new diseases, or die. The researchers also used imaging and genetic information to substantiate that the AI predictions were directly related to the heart’s biology. This proved that the model can detect minute changes in the heart’s structure, which signify risk of disease or death.
“We cardiologists use our experience and standard guidelines when we look at ECGs. We sort them into ‘normal’ and ‘abnormal’ patterns to help us diagnose disease”, said Dr. Arunashis Sau, lecturer at Imperial College London. “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.”
Beyond ECGs: AI algorithms improve implantable loop recorder accuracy
In cases where standard ECGs aren’t enough to detect heart problems, internal loop recorders (ILRs) are often used instead. An accurate diagnosis was made in 73% of patients with ILRs, compared to just 23% of patients with ECGs, one study found. An internal loop recorder is inserted under the skin to monitor the heart and record abnormal electrical signals for up to three years. It saves and sends the data to the cardiologist multiple times a day.
Similar to AI-enhanced ECGs, AI-powered medical algorithms have also been developed to analyze heart rate and rhythm data collected by ILRs to improve arrhythmia detection and minimize the rate of false positives. As the algorithms use precise signal processing and analysis, they can detect irregular heartbeats with greater accuracy. The algorithm also orders the data to prioritize real health issues that require investigation and treatment, which results in improved patient care.
The new AI models ultimately have the potential to improve heart disease diagnosis, so patients receive better, faster treatment. “The important next step is to test whether using these models can actually improve patient outcomes in clinical studies”, said Dr. Fu Siong Ng, Reader in Cardiac Electrophysiology at Imperial College London. The clinical trials, set to begin mid-2025, will assess how beneficial the model is to real patients in outpatient clinics.