Photo Credit: Hanna Sova

A research team from the Mayo Clinic externally validated an AI electrocardiography algorithm that detects hypertrophic cardiomyopathy with high accuracy.

An AI model trained to detect hypertrophic cardiomyopathy (HCM) from 12-lead electrocardiogram (ECG) raw data demonstrated high accuracy when tested in diverse international cohorts, according to a study published in the European Heart Journal – Digital Health.

“This study is one component of the validation efforts of the Mayo Clinic AI-ECG HCM algorithm,” wrote corresponding author Konstantinos C. Siontis, MD, “including a previous study… where the algorithm demonstrated excellent discrimination performance and an internal validation where the algorithm was applied in tandem with clinical factors to optimize detection of new HCM cases in routine clinical practice.”

This external validation study assessed the performance of the AI-ECG algorithm in three cohorts in Bern, Switzerland; Oxford, England; and Seoul, South Korea. Participating centers at the University of Bern, Oxford University, and Seoul National University contributed raw 12-lead ECG data for patients with HCM and patients without HCM who served as controls.

The merged cohort included 773 patients with HCM and 3,867 controls. The HCM patient sample was 54.6% East Asian, 43.2% White, and 2.2% Black. The prevalence of obstructive HCM was 15.7%, and the prevalence of apical HCM was 21.9%.

According to the study, the median AI-ECG probabilities of HCM were 85% for patients with HCM and 0.3% for control subjects. The AI-ECG algorithm had an area under the receiver operating characteristic curve (AUC) of 0.922 in the overall cohort. The AI-ECG model had a diagnostic accuracy of 86.9%, a sensitivity of 82.8%, and a specificity of 87.7% for HCM detection.

Subgroup analyses showed the model’s performance was better in women (AUC 0.94) than in men (AUC 0.91). AUC and sensitivity were higher for patients with apical HCM than non-apical HCM.

In an age- and sex-matched analysis that included 773 patients with HCM matched 1:2 with 1546 controls, the AUC was 0.921, accuracy was 88.5%, sensitivity was 82.8%, and specificity was 90.4%.

“The value of this algorithm in improving HCM detection in clinical practice and screening settings requires prospective evaluation,” the researchers advised.