Loading ...

An AI algorithm that can diagnose heart failure using an ECG has shown promise as a screening tool in a trial in Kenya.

The burden of heart failure is particularly high in sub-Saharan Africa, with patients presenting at younger ages and facing worse outcomes — and the prevalence of risk factors continues to rise. Early detection of left ventricular systolic dysfunction (LVSD) would help with prevention and treatment but is a challenge in resource-limited settings with limited access to echocardiography and natriuretic peptide biomarkers.

“In sub-Saharan Africa a lot of patients present much later and there is a lot of fatal index presentation with heart failure, so early diagnosis and screening is all the more essential for this part of the world,” said Ambarish Pandey, MD, a cardiologist at UT Southwestern Medical Center in Dallas.

AI-enhanced ECG models that have been shown to accurately diagnose LVSD could help with that screening. But those models were primarily trained and tested on data from high-income countries, and they need to be validated in other settings.

So Pandey and his colleagues ran a prospective trial with patients from eight hospitals to test the algorithm’s performance against the gold-standard of echocardiogram. The work was published in JAMA Cardiology.

Inexpensive and Scalable Model

Among the 1444 participants who had both an ECG and echocardiogram, 14.1% were found to have LVSD. The AI algorithm was found to have a sensitivity of 95.6%, specificity of 79.4%, positive predictive value of 43.2%, negative predictive value of 99.1%, and area under the receiver operating characteristic curve of 0.96.

This means negative ECG results can be treated with high confidence, whereas positive ones should be combined with other risk scores and possibly followed up with an echocardiogram to confirm. Pandey sees a lot of scope for clinical use in lower-income countries.

“ECG is a fairly cheap and scalable model of screening, and the AI-ECG model we used has been well validated so this has immediate potential for scalability in screening and identifying high-risk patients who can then be started on effective therapies,” he said.

New Triage Tool?

Fu Siong Ng, PhD, a cardiologist at Imperial College London, London, England, and co-author of an accompanying editorial, said it was important to validate the potential of AI-ECG models with prospective studies such as this in a variety of settings to ensure they work reliably. Even in this case, the high burden of disease detected means these results may not be generalizable outside Kenya, he said.

In cases where the models are well validated, Ng foresees them being used initially as a triage tool to determine which patients should get further testing when that testing capacity is limited. “You can prioritize and triage those who are most in need of urgent testing,” he said.

The next step, Pandey said, is to follow up and see how early diagnosis or early identification can improve disease management in Kenya.

“It’s not just about giving patients a diagnosis but actually providing them the resources to manage it and then get them on therapies early,” he said. “So that is our next goal, to get these patients on effective therapies and see if early detection and treatment can improve outcomes.”

Pandey reported receiving consulting fees from Medical AI, Tricog, and Ultromics. Ng reported having a pending patent on artificial intelligence-enhanced ECG and being a cofounder of Cardiovolt.ai. 

Brian Owens is a freelance journalist based in New Brunswick, Canada.