Automated Detection of Reduced Ejection Fraction Using an ECG-Enabled Digital Stethoscope

Abstract

BACKGROUND: Asymptomatic left ventricular systolic dysfunction (ALVSD) affects 7 million globally, leading to delayed diagnosis and treatment, high mortality, and substantial downstream health care costs. Current detection methods for ALVSD are inadequate, necessitating the development of improved diagnostic tools. Recently, electrocardiogram-based algorithms have shown promise in detecting ALVSD. OBJECTIVES: The authors developed and validated a convolutional neural network (CNN) model using single-lead electrocardiogram and phonocardiogram inputs captured by a digital stethoscope to assess its utility in detecting individuals with actionably low ejection fractions (EF) in a large cohort of patients. METHODS: 2,960 adults undergoing echocardiography from 4 U.S. health care networks were enrolled in this multicenter observational study. Patient data were captured using a digital stethoscope, and echocardiograms were performed within 1 week of data collection. The algorithm's performance was compared against echocardiographic EF (EF measurements, categorizing EF as normal and mildly reduced [>40%] or moderate and severely reduced [≤40%]). RESULTS: The CNN model demonstrated an area under the receiver operating characteristic curve of 0.85, with a sensitivity of 77.5%, specificity of 78.3%, positive predictive value of 20.3%, and negative predictive value of 98.0%. Among those with an abnormal artificial intelligence screen but EF >40% (false positives), 25% had an EF between 41%-49% and 63% had conduction/rhythm abnormalities. Subgroup analyses indicated consistent performance across various demographics and comorbidities. CONCLUSIONS: The CNN model, utilizing a digital stethoscope, offers a noninvasive and scalable method for early detection of individuals with EF ≤40%. This technology has the potential to facilitate early diagnosis and treatment of heart failure, thereby improving patient outcomes.