Revisiting Abnormalities of Ventricular Depolarization: Redefining Phenotypes and Associated Outcomes Using...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Traditional ECG categorization of abnormal ventricular depolarization (broad QRS) into LBBB and RBBB may miss disease subtypes.
• Current classifications lack granularity for predicting cardiovascular disease (CVD) risk and mortality.

Purpose of the Study:

• To identify and characterize novel phenogroups of broad QRS complexes using unsupervised machine learning.
• To assess the predictive value of these phenogroups for cardiovascular outcomes and treatment response.

Main Methods:

• Trained a variational autoencoder on 1.1 million ECGs to extract 51 latent features.
• Applied reversed graph embedding to 42,538 ECGs with QRS duration >120 ms to model population heterogeneity.
• Identified six distinct phenogroups based on ECG features.

Main Results:

• Six phenogroups were identified, including distinct RBBB and LBBB subtypes.
• A higher-risk RBBB phenogroup showed significantly increased risk of cardiovascular and all-cause mortality.
• Within LBBB phenogroups, tree position predicted CVD risk and cardiac resynchronization therapy response.

Conclusions:

• Novel phenogroups derived from machine learning offer a more nuanced understanding of broad QRS complexes.
• These phenogroups can enhance patient selection for cardiac resynchronization therapy (CRT) in LBBB patients.
• Findings suggest improved investigation and follow-up strategies for RBBB patients with higher-risk phenogroups.