Detection of Left Ventricular Hypertrophy Using Bayesian Additive Regression Trees: The MESA.

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Left ventricular hypertrophy (LVH) is a significant risk factor for cardiovascular disease.
• Traditional electrocardiogram (ECG) criteria for diagnosing LVH have limitations in sensitivity and specificity.
• Development of novel, accurate methods for LVH detection is crucial for risk stratification.

Purpose of the Study:

• To develop and validate a new LVH criterion using Bayesian Additive Regression Trees (BART), a machine-learning technique.
• To compare the diagnostic and prognostic performance of the new BART-LVH criterion against traditional ECG-LVH criteria and cardiac MRI.

Main Methods:

• Utilized the Multi-Ethnic Study of Atherosclerosis (MESA) cohort (n=4714) free of cardiovascular disease at baseline.
• Employed BART to predict LV mass from ECG and participant characteristics, with cardiac MRI serving as the gold standard.
• Randomly split the cohort into training (n=3774) and validation (n=940) sets for model development and testing.

Main Results:

• In the validation set, BART-LVH exhibited the highest sensitivity (29.0%) for LVH detection compared to traditional ECG criteria (e.g., Sokolow-Lyon-LVH at 21.7%).
• All criteria demonstrated high specificity (>93%).
• During a median follow-up of 12.3 years, BART-LVH and cardiac MRI-LVH were significantly associated with increased risk of mortality, cardiovascular disease events, and coronary heart disease events, outperforming traditional ECG criteria.

Conclusions:

• The novel BART-LVH criterion offers superior diagnostic and prognostic accuracy for identifying LVH compared to conventional ECG-based methods.
• BART-LVH demonstrates comparable performance to cardiac MRI in predicting adverse cardiovascular outcomes.
• This machine-learning approach represents a promising advancement in cardiovascular risk assessment and management.