Assessment of cardiovascular risk from ambulatory blood pressure measurements enabled by deep learning

Left ventricular hypertrophy (LVH), clinically defined as an elevated left ventricular mass index (LVMI), is a strong predictor of adverse cardiac outcomes1,2. Currently, measurements of LVMI are performed using transthoracic echocardiography, accessible only in clinical settings. In this study, we use data from an ambulatory blood pressure monitor, combined with laboratory test results and demographics data collected once, to investigate whether LVMI and LVH may be predicted and classified accurately. We developed and tested machine learning models using clinical data. For LVMI prediction, a mean absolute error of 10.4 ± 1.5 with our test set, and 15.6 ± 1.2 with an external test set, were achieved. For the classification of LVH and concentric LVH, mean areas under the curve of ≥0.95 and ≥0.93 were achieved, respectively. This study presents a method for accurately assessing cardiovascular risk using ambulatory measurements and highlights the use of waveform data for assessing hypertrophy.