AI-ECG risk estimation (AIRE), a model that predicts heart disease risks and complications

A multinational research group including participants from Imperial College London and Imperial College Healthcare NHS Trust have published a paper in Lancet Digital Health, where they describe a deep learning model that enhances current AI-assisted electrocardiogram (ECG) practices to deliver actionable predictions on individual patients' risks of developing complications due to heart disease, including early death. AIRE was able to correctly identify the risk of death in the ten years following the ECG (from high to low) in 78% of cases.

Although the idea of AI-enhanced ECG is not new, it has only been used to support more accurate diagnoses, not to predict risks and complications due to known heart disease. By estimating the possibility of risks of complications, in addition to helping professionals detect heart disease earlier, a system like AIRE could provide healthcare professionals with tools and information to prioritize treatment for patients at greater risk. The researchers first trained the AIRE model for mortality predictions and then trained seven sub-models for specific tasks, including prediction for certain types of heart disease, and death from non-cardiovascular causes.

The researchers used millions of ECG scans paired with five-year life statuses for the scans' patients to build a dataset that leveraged 50% of the available data for training, 10% for validation, and 40% for testing. Using this dataset, the researchers essentially taught the model to look for the same patterns and fluctuations in the flow of electrical signals within, and between, the different chambers of the heart that healthcare professionals look for when making a diagnosis.

The researchers believe that one of the conditions for the system's success is that the AI model is capable of detecting a greater level of detail and nuance, leading it to detect abnormalities that healthcare professionals would likely overlook. Moreover, chronic diseases like diabetes, while not directly related to the heart, will affect it and other vital organs in some way. Having a system that is also capable of detecting damage attributable to non-cardiovascular causes, not only helps raise the accuracy of predictions related to cardiovascular disease but also makes the AIRE system a useful tool for assessing the severity of the potential complications that a wider population, and not only cardiology patients, could develop.

The NHS is planning to set up trials for the AIRE in hospitals from the Imperial College Healthcare NHS Trust and Chelsea and Westminster Hospital NHS Foundation Trust starting mid-2025. The trials will recruit patients from outpatient care facilities and inpatient wards to help researchers and the NHS evaluate the benefits of deploying AIRE for use with real patients.