An artificial intelligence (AI)-enhanced electrocardiography model may effectively predict the development and progression of moderate or severe mitral, tricuspid or aortic regurgitation—MR, TR, or AR—before the appearance of symptoms or physical changes that can be detected by ultrasound, research published in the European Heart Journal has shown.
The study was led by Arunashis Sau and Fu Siong Ng (Imperial College London, London, UK) alongside researchers at Zhongshan Hospital (Shanghai, China), who trained AI models using nearly one million electrocardiogram (ECG) and transthoracic echocardiogram (TTE) records from over 400,000 patients in China. These were tested on a separate group of more than 34,000 patients from a secondary care dataset from Beth Israel Deaconess Medical Center (Boston, USA).
Using a deep learning approach known as a convolutional neural network (CNN) to analyse the data, the AI was trained to spot patterns in the ECG with signs of valve disease on the TTE. Classification models learned to identify if someone had moderate or severe valve disease at the time of the ECG, and survival models learned to predict if someone without significant disease at baseline may go on to develop it in the future.
The researchers showed that the AI could accurately predict who would go on to develop significant regurgitant valvular heart diseases, with the model able to correctly identify the risk of disease in the years following the ECG (from high to low) in around 69–79% of cases.
“By the time symptoms and structural changes appear in the heart, it may be too late to do much about it,” said Sau. “Our work is harnessing AI to detect subtle changes at the earliest stage from a simple and common test, and we think this could be really transformative for doctors and patients.
“Rather than waiting for symptoms, or relying only on expensive and time-consuming imaging tests, we could use AI-enhanced ECGs to spot those most at risk earlier than ever before. This means that many more people could get the care they need before their hidden condition affects their quality of life or becomes life-threatening.”
People flagged as ‘high-risk’ by the algorithm were up to 10 times more likely to develop these diseases than those classed as lower risk.
“AI has enormous potential for improving healthcare around the world, but it requires huge amounts of data to train and test these algorithms,” said Ng, the senior author. “Our work is an example of the benefits of international collaboration in this fast-growing area. By training the model in an almost exclusively Chinese population and then testing in a US cohort, we can show that our AI tool has the potential to be applied in various countries and settings around the world. This ultimately means it has the potential to help even more patients.”
In their discussion of the research in the European Heart Journal, the study team writes that their work addresses a “notable gap in the clinical management of regurgitant valvular heart diseases” and anticipates that advances in AI-ECG technology could greatly facilitate the screening of these conditions. AI models are poised to streamline the surveillance of patients with clinically insignificant regurgitant valve disease, they state, by providing enhanced diagnostic capabilities and facilitating personalised treatment plans.
There are several limitations to the study, they add, including that the ECG and TTE data were retrospectively collected from the hospital-based cohorts, which may introduce some inherent patient selection bias, as well as noting that the model for predicting AR had reduced performance compared with those for MR and TR.
“Artificial intelligence-enhanced electrocardiography can accurately diagnose prevalent and predict future development of significant MR, AR, and TR,” they conclude, nevertheless. “This approach may serve as the basis for the development of a regurgitant valvular heart disease prediction programme, to facilitate early detection, timely intervention, and potentially preventative therapeutics, while also optimising resource utilisation by tailoring the frequency of echocardiographic surveillance based on risk of progression.”
