Valvular heart disease continues to go undiagnosed while being a significant cause of cardiovascular morbidity and death. Electrocardiography (ECG) deep learning analysis may help identify mitral regurgitation (MR), aortic regurgitation (AR), and aortic stenosis (AS). For a study, researchers sought to develop deep learning ECG algorithms to detect both individual and combined AS, AR, and MR in mild to severe cases.

Between 2005 and 2021, 77,163 individuals had an ECG within a year after having an echocardiogram. These patients were divided into train (n = 43,165), validation (n = 12,950), and test sets (n = 21,048; 7.8% with any of AS, AR, or MR). The area under the receiver-operating characteristic (AU-ROC) and precision-recall curves were used to evaluate the performance of the model. An independent data set was subjected to external validation. To mimic screening effectiveness using the deep learning approach, test accuracy was modeled using various disease prevalence levels.

The following deep learning algorithm models had similar accuracy in external validation: AS (AU-ROC: 0.88), AR (AU-ROC: 0.77), MR (AU-ROC: 0.83), and any of AS, AR, or MR (AU-ROC: 0.84; sensitivity 78%, specificity 73%). Test features were based on underlying prevalence and predetermined sensitivity thresholds in screening program modeling. The positive and negative predictive values were 20% and 97.6%, respectively, for a prevalence of 7.8%.

In the multicenter cohort, deep learning analysis of the ECG can reliably identify AS, AR, and MR, which may form the basis for creating a valvular heart disease screening program.

Reference: jacc.org/doi/10.1016/j.jacc.2022.05.029

Author