Predicting Hypertension-mediated Subclinical Left Ventricular Hypertrophy using Machine Learning Techniques

Abstract

Hypertension is the commonest cause of left ventricular hypertrophy (LVH). Using cardiac magnetic resonance (CMR) imaging, four hypertension-mediated LVH phenotypes have been reported: normal left ventricle (LV), LV remodelling, eccentric LVH and concentric LVH, with varying prognostic implications. An accessible and cost-effective approach to identify LVH in hypertensives is desirable. The electrocardiogram (ECG) is routinely used to detect LVH, however its capacity to differentiate between the hypertension-mediated LVH phenotypes and its associations with cardiovascular (CV) outcomes is unknown. I developed and validated three machine learning (ML) algorithms to detect LVH and hypertension-mediated LVH phenotypes from the ECG and performed an exploratory analysis to determine if ECG-predicted LVH groups were associated with CV outcomes. I also investigated CMR structural and functional differences across the hypertension-mediated LVH phenotypes and assessed whether there were differences in the risk of major adverse CV events and heart failure. In 37,534 UK Biobank participants, classification of LVH had comparable accuracies with all models, all > 0.80 AUC values. QRS amplitude and blood pressure (P<0.001) were the features most strongly associated with LVH. Classification of hypertension-mediated LVH phenotypes was also comparable among the ML methods, with superior prediction of eccentric LVH (AUC=0.86) and a 3.2 times increased hazard rate of heart failure (HR 3.24, CI: 1.06-9.86) in the ECG-predicted group. ECG biomarkers enhanced model performance of all classifiers, compared to using clinical variables alone with external validation in an independent study demonstrating robustness of the model. My results also showed hypertensives with CMR defined eccentric LVH had the worst LV function, and this group had a 9.5 times higher risk of heart failure (HR 9.49, CI: 6.03-14.94). In summary, my results support using ECG biomarkers for predicting hypertension-mediated LVH phenotypes. The results provide important new data for improving risk stratification of hypertensives and prediction of heart failure.