Twelve-lead electrocardiography-based artificial intelligence predicts the upcoming future of patients with heart failure with mildly reduced ejection fraction

Heart failure with mildly reduced ejection fraction (HFmrEF) has emerged as the predominant subtype of heart failure (HF), but there is a paucity of data for identifying it using artificial intelligence (AI) based electrocardiogram (ECG).

This study aimed to develop artificial intelligence (AI)-ECG to identify and predict the prognosis of patients with HFmrEF.

We collected 104,336 12-lead electrocardiography (ECG) datasets from April 2009 to December 2021 in a tertiary center. The AI-ECG encompasses a novel model that combines an automatic labelling preprocessing method with a transformer architecture incorporating a triplet loss for HFmrEF analysis.

The receiver operating characteristic analyses revealed that the area under the curve of AI-ECG for identifying all types of HF were acceptable (0.87, 95% confidence interval [CI]: 0.86-0.89), but that for identifying those with HFmrEF was relatively lower (0.83, 95% CI: 0.78-0.86) than those with HF with reduced ejection fraction (HFrEF) (0.89, 95% CI: 0.85-0.82) and those with normal ejection fraction (EF) (0.87, 95% CI: 0.85-0.89). The analysis of ECG features showed significant increases in QRS duration (p=0.001), QT interval (p=0.045), and corrected QT interval (p=0.041) with increasing ‘Severity by Euclidean distance’. Following the predictability analysis with another group of 1,134 patients for improvements of follow-up EF in HFmrEF, the patients were grouped into three clusters based on the AI- Euclidean distance; Cluster 1 had the most severe cases and poorer outcomes than Clusters 2 (p<0.001) and 3 (p<0.001).

AI-ECG presents an innovative approach for the prognostic stratification of cardiac contractility in patients with HFmrEF. By using AI-ECG, patients with HFmrEF will be able to predict upcoming disease progression.