A Novel ECG-Based Deep Learning Algorithm to Predict Cardiomyopathy in Patients With Premature Ventricular Complexes.

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Medical Diagnostics

Background:

• Premature ventricular complexes (PVCs) are common arrhythmias that can lead to PVC-induced cardiomyopathy.
• Early identification of patients at risk for left ventricular ejection fraction (LVEF) reduction is crucial for timely intervention.
• Current methods for predicting cardiomyopathy in PVC patients may not fully leverage readily available diagnostic data.

Purpose of the Study:

• To develop and validate a deep-learning algorithm for predicting LVEF reduction in patients with PVCs.
• To assess the accuracy of the deep-learning model using a large, multi-center dataset.
• To identify key electrocardiogram (ECG) features contributing to the model's predictions.

Main Methods:

• Utilized a dataset of 383,514 ECGs from 5 hospitals, with 14,241 patients diagnosed with PVCs.
• Developed a deep-learning model trained and tested internally, with external validation across multiple institutions.
• Primary outcome was diagnosis of LVEF ≤40% within 6 months; analyzed using area under the receiver operating curve (AUC) and explainability plots.

Main Results:

• The deep-learning model achieved an AUC of 0.79 (95% CI: 0.77-0.81) in predicting LVEF reduction to ≤40% within 6 months.
• Explainability analysis highlighted the QRS complex and ST segment in sinus rhythm as key predictors, independent of PVC burden.
• In patients undergoing PVC ablation, 89% showed LVEF improvement and cardiomyopathy resolution.

Conclusions:

• Deep-learning analysis of 12-lead ECGs can accurately predict new-onset cardiomyopathy in PVC patients, irrespective of PVC burden.
• The model's performance was consistent across different demographic groups (sex, race).
• The algorithm relies on sinus rhythm ECG features (QRS complex/ST segment), not PVC morphology, for prediction.