AI-enabled ECG platform can predict future health risks, including early death, study finds

1st November 2024

Artificial intelligence (AI)-enabled electrocardiography (ECG) can accurately predict an individual patient’s risk of future cardiovascular events as well as their short and long-term risk of dying, a study finds.

Existing AI-enabled ECG could predict disease and mortality but could not give sufficient information to guide clinical decisions for individual patients and were not adopted into practice, UK researchers wrote in The Lancet Digital Health.

To address the limitations of previous AI models, the team from Imperial College London and Imperial College Healthcare NHS Trust developed the AI-ECG risk estimator (AIRE) platform using data from a secondary care dataset of 1,163,401 ECGs taken from 189,539 patients.

Using a deep learning and a discrete-time survival model, researchers were able to create a patient-specific survival curve with a single ECG, allowing the AIRE platform to predict not only risk of mortality but also time-to-mortality.

The platform was validated in five diverse, transnational cohorts from the USA, Brazil, and the UK, including volunteers, primary care patients, and secondary care patients.

Researchers found the model was able to identify the risk of death in the ten years following the ECG (from high to low) in 78% of cases, and was also able to predict future ventricular arrhythmia, future atherosclerotic cardiovascular disease, and future heart failure.

‘Through phenome-wide and genome-wide association studies, we also identified candidate biological pathways for the prediction of increased risk, including changes in cardiac structure and function, and genes associated with cardiac structure, biological ageing, and metabolic syndrome,’ they wrote.

They concluded that clinicians could act on AIRE’s predictions to provide targeted, personalised and earlier intervention.

‘AIRE is an actionable, explainable, and biologically plausible AI-ECG risk estimation platform that has the potential for use worldwide across a wide range of clinical contexts for short-term and long-term risk estimation,’ they said.

Lead author Dr Arunashis Sau, an academic clinical lecturer at Imperial College London’s National Heart and Lung Institute, and cardiology registrar at Imperial College Healthcare NHS Trust, said compared with cardiologists the AI model could detect more subtle detail in the ECGs.

‘It can ‘spot’ problems in ECGs that would appear normal to us, and potentially long before the disease develops fully,’ he said.

‘Our analysis shows that the AI can tell us a lot about not only about the heart but also what is going on elsewhere in the body and may be able to detect accelerated ageing.’

Dr Sau acknowledged it was necessary to see how the model performed in the healthcare system, but suggested it was possible that in the future, the technology could be used in a wearable device that provided doctors with continuous remote monitoring and a potential alert system.

Senior study author Dr Fu Siong Ng, Reader in Cardiac Electrophysiology at the National Heart & Lung Institute at Imperial College London, said the work had shown the AI model was a credible and reliable tool that could, in future, be programmed for use in different areas of the NHS to provide doctors with relevant risk information.

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

Dr Ng, who is also a consultant cardiologist at Imperial College Healthcare NHS Trust and Chelsea and Westminster Hospital NHS Foundation Trust, said the technology could also reduce waiting lists and allow more efficient allocation of resources.

Trials of the model are planned to start by mid-2025 in hospitals across Imperial College Healthcare NHS Trust and Chelsea and Westminster Hospital NHS Foundation Trust.

Participants are set to be recruited from outpatient clinics and from inpatient medical wards.

AI-guided screening improves detection of atrial fibrillation

10th October 2022

An AI-guided screening device applied to an ECG could potentially identify patients at high risk of atrial fibrillation compared to usual care.

The worldwide prevalence of atrial fibrillation (AF) was estimated to be 37,574 million cases (0.51% of worldwide population) in 2017 and the authors noted how this has increased by 33% during the last 20 years. However, AF is often asymptomatic and a US-based study estimated that of a total AF prevalence of 5.3 million (in 2009), 0.7 million (13.1% of AF cases) were undiagnosed.

Moreover, AF is a major risk factor for strokes and it has been estimated that around 20% of all strokes are caused by the arrhythmia, highlighting the need to identify those affected.

AF can be detected with an ECG although patients require intermittent and prolonged monitoring which is labour intensive. In a 2019 study, it was found that an AI-guided screening tool enabled an ECG, acquired during normal sinus rhythm, to identify individuals with atrial fibrillation.

Nevertheless, while this important advance was potentially of great clinical value, prior to widespread implementation, there were two further and important questions. Firstly, could the AI-guided screening tool enable risk-stratification that was superior to currently available approaches and secondly, how often or how much monitoring would be required for those deemed to be at a high-risk for AF using the device.

In trying to answer these questions, a team of US researchers, undertook a non-randomised, interventional trial and prospectively recruited patients who had risk factors for a stroke but without AF and who had an ECG.

The AI-guided screening tool using just the raw ECG data, determined an individual’s AF risk score which was categorised as either high or low-risk. Eligible participants wore a continuous ambulatory heart rhythm monitor all the time for 30 days and were matched 1:1 to real-world control patients (i.e., a group not wearing the monitor).

The primary outcome of interest was newly diagnosed atrial fibrillation, defined as an episode lasting 30 seconds or longer.

AI-guided screening tool and detection of atrial fibrillation

A total of 1,003 individuals with a mean age of 74 years (61.8% male) were included in the study, of whom 370 were deemed to be at a low AI assessed risk of AF.

Over a mean of 22.3 days, AF was detected in 1.6% of those deemed to be at low risk and 7.6% deemed to be at high risk (odds ratio, OR = 4.98, 95% CI 2.11 – 11.75, p = 0.0002).

The researchers calculated that AI-guided screening was associated with a significantly higher detection of AF compared to usual care in those deemed at high risk (10.6% vs 3.6%, p < 0.0001). However, the difference was not significant for those deemed at low risk (2.6% vs 1.1%, p = 0.12).

The authors concluded that the AI-guided screening tool was able to risk-stratify patients for AF in the short-term and to provide a higher rate of detection among patients deemed to be at high risk.

Citation
Noseworthy PA et al. Artificial intelligence-guided screening for atrial fibrillation using electrocardiogram during sinus rhythm: a prospective non-randomised interventional trial Lancet 2022.