Implementation of a Deep Learning Model Using Electrocardiogram Tracings to Address Disparities in the Diagnosis of Structural Heart Disease
Presentation Time: 08:15 AM - 08:30 AM
Abstract Keywords: Machine Learning, Health Equity, Population Health, Deep Learning, Mobile Health
Primary Track: Applications
Programmatic Theme: Public Health Informatics
The under-diagnosis and delayed diagnosis of structural heart disease (SHD) remains a significant public health concern, particularly among medically underserved populations who experience excess morbidity and mortality. Many of those with undiagnosed SHD disproportionately use the emergency department (ED) as their primary source of care. We describe the derivation, validation and implementation of EchoNext, a deep learning model that uses electrocardiogram (ECG) tracings to identify patients at risk for undiagnosed SHD and would most benefit from a transthoracic echocardiogram (TTE). This artificial intelligence (AI) algorithm, EchoNext, was developed and tested on over one million ECG-TTE pairs within our health system with excellent area under the receiver operating characteristic curve and area under the precision-recall curve (85.2% and 78.5%, respectively), which remained consistent when externally validated using data from three external health systems. We will deploy EchoNext coupled with an electronic health record alert for patients receiving an ECG in the ED. At-risk patients will be referred to an existing mobile integrated health program, Community Tele-Paramedicine (CTP), to coordinate post-ED discharge TTE and necessary follow-up cardiology care. We will study the impact of this implementation across six EDs within our urban, integrated health system with a mix of academic and community sites. This will provide a successful use-case for AI in clinical medicine that highlights the critical role the ED plays in this significant public health concern and demonstrate the feasibility of leveraging AI to improve SHD diagnosis and reduce health disparities without burdening frontline providers.
Speaker(s):
Brock Daniels, MD, MPH
Weill Cornell Medicine
Presentation Time: 08:15 AM - 08:30 AM
Abstract Keywords: Machine Learning, Health Equity, Population Health, Deep Learning, Mobile Health
Primary Track: Applications
Programmatic Theme: Public Health Informatics
The under-diagnosis and delayed diagnosis of structural heart disease (SHD) remains a significant public health concern, particularly among medically underserved populations who experience excess morbidity and mortality. Many of those with undiagnosed SHD disproportionately use the emergency department (ED) as their primary source of care. We describe the derivation, validation and implementation of EchoNext, a deep learning model that uses electrocardiogram (ECG) tracings to identify patients at risk for undiagnosed SHD and would most benefit from a transthoracic echocardiogram (TTE). This artificial intelligence (AI) algorithm, EchoNext, was developed and tested on over one million ECG-TTE pairs within our health system with excellent area under the receiver operating characteristic curve and area under the precision-recall curve (85.2% and 78.5%, respectively), which remained consistent when externally validated using data from three external health systems. We will deploy EchoNext coupled with an electronic health record alert for patients receiving an ECG in the ED. At-risk patients will be referred to an existing mobile integrated health program, Community Tele-Paramedicine (CTP), to coordinate post-ED discharge TTE and necessary follow-up cardiology care. We will study the impact of this implementation across six EDs within our urban, integrated health system with a mix of academic and community sites. This will provide a successful use-case for AI in clinical medicine that highlights the critical role the ED plays in this significant public health concern and demonstrate the feasibility of leveraging AI to improve SHD diagnosis and reduce health disparities without burdening frontline providers.
Speaker(s):
Brock Daniels, MD, MPH
Weill Cornell Medicine
Implementation of a Deep Learning Model Using Electrocardiogram Tracings to Address Disparities in the Diagnosis of Structural Heart Disease
Category
Podium Abstract
Description
Date: Wednesday (11/13)
Time: 08:15 AM to 08:30 AM
Room: Continental Ballroom 1-2
Time: 08:15 AM to 08:30 AM
Room: Continental Ballroom 1-2