Electrocardiogram-based deep learning improves outcome prediction following cardiac resynchronization therapy

Comprehensive Summary

This study by Popat et al. aims to identify and visualize electrocardiogram (ECG) features using a deep learning-based algorithm to predict cardiac resynchronization therapy (CRT) outcome. A deep algorithm trained on 1.1 million ECGs from over 250,000 patients was used to summarize most ECG features into FactorECG–21 explainable features. This model was then applied to pre-implant ECGs from 1306 CRT patients across three centers to predict clinical outcomes. Based on the results, FactorECG had great discriminative ability to predict clinical outcomes without needing extra clinical data and outperformed contemporary guideline criteria.

Outcomes and Implications

This research is important in patients with dyssynchronous heart failure (HF), because CRT can effectively restore left ventricular electrical activation and therefore, improve clinical outcomes. Existing ECG criteria are limited in predicting outcomes, so better prediction tools can guide more effective therapy decisions. AI applied to existing 12-lead ECGs could become a practical tool to enhance CRT outcome prediction. While this can be a great tool to potentially accelerate personalized medicine, further validation is needed for widespread clinical implementation.