Practical Approaches to Care in Emergency Syncope (PACES)

Columbia University Health Sciences, New York NY

Investigators

Linked publications & trials

Abstract

Project Summary/Abstract Syncope (or transient loss of consciousness) is a common reason to present to the ED, representing over 1.3 million visits per year in the United States. Although syncope is most often benign, it can occasionally be caused by serious cardiac diseases such as cardiac dysrhythmia, valvular heart disease, or other structural heart disease. Despite thorough evaluation in the ED, the cause of syncope remains unknown in over 50% of cases. The goal of this project is to use artificial intelligence-electrocardiogram (ECG) models to improve the diagnosis of cardiac disease for patients who present to the emergency department (ED) with syncope, by better delineating which patients require further cardiac testing, such as echocardiography or prolonged cardiac monitoring. Artificial intelligence (AI) models, using machine learning approaches, have been developed using retrospective ECG data to predict valvular heart disease and, more broadly, any structural heart disease. The first model, known as ValveNet, is highly accurate at predicting mitral regurgitation, aortic stenosis, and aortic regurgitation. The second model, known as EchoNext, is highly accurate at predicting all forms of structural heart disease as diagnosed by echocardiography, including valvular heart disease, ventricular systolic dysfunction, left ventricular hypertrophy, and significant pericardial effusions. While promising, these two AI models require external validation prior to clinical implementation. In Aim 1 of this proposal, we use prospectively collect data on ~1,012 ED patients with syncope/pre- syncope to validate the predictive accuracy of these two AI models in detecting valvular and structural heat disease, including mitral valve prolapse, using echocardiography as our gold standard. In Aim 2, we will assess the whether baseline valvular heart disease is an independent risk factor for serious cardiac events, such as acute cardiac dysrhythmias, at 30 days among ED patients with syncope. If validated and shown to accurately predict valvular and structural heart disease, these artificial intelligence models could play a major role in improving emergency syncope care by rapidly identifying patients who require echocardiography and/or prolonged cardiac monitoring. This would, in turn, lead to expedited medical and surgical therapy to reduce cardiac morbidity and mortality. This study, entitled SyncopeNet, will help improve clinical care for patients with syncope and advance the field of syncope research.