Artificial Intelligence for Determining Genetic Risk of Sudden Cardiac Arrest using the Electrocardiogram

PROJECT SUMMARY Sudden cardiac arrest (SCA) is a leading cause of cardiovascular deaths in the United States. Despite its prevalence, predicting risk of SCA is challenging due to limited methods for early detection. It has been known that risk of SCA is mediated by genetic contributions, which can be inherited as monogenic and polygenic factors. Although there exist techniques for assessing clinical risk, there is currently no viable approach to deduce genetic risk unless through direct genotyping or genetic sequencing. Furthermore, current clinical genomic testing is constrained in terms of accessibility, cost, and expert interpretation requirements, leading to a lack of scalable methods for early determination of SCA genetic risk. The electrocardiogram (ECG) is a non-invasive, widely used tool for measuring cardiac electrical activity. As biological mechanisms influence waveform patterns that define certain arrhythmias, the ECG has potential to reveal the genetic underpinnings of the electrical function of the heart. While subtle waveform morphology may escape human observation, ECG interpretation has been greatly augmented by deep learning (DL) techniques, allowing for the discernment of subtle waveform patterns of even subclinical disease for which diagnostic guidelines are lacking. DL has already been leveraged on ECG to identify various cardiac pathologies, such as hypertrophic cardiomyopathy, low left ventricular ejection fraction, ST elevated myocardial infarction, and valvular disease, but has yet to be applied to genetic predisposition to SCA. As such, the applications of DL to ECG waveform promise to provide insight into genotypic foundations of SCA to enable prevention and prophylactic strategies in cardiac electrophysiology and personalized medicine. This proposal seeks to develop DL models that bridge the gap between phenotypic ECG patterns and disease genotypes. To improve the identification of monogenic inherited syndromes, a DL model will be leveraged on electronic health record (EHR) and ECG data to classify those with risk genotypes (Aim 1). To facilitate identification of those at high polygenic risk for SCA, a separate DL algorithm using EHR and ECG data will be developed for classification of individuals with high polygenic risk score (PRS) for coronary artery disease and SCA (Aim 2). For Aims 1 and 2, by integrating data from large repositories of genetic data such as the Mount Sinai Million Health Discoveries Program, All of Us Research Program, the United Kingdom Biobank, we hypothesize that DL has potential to enable identification of monogenic and polygenic risk of SCA. Supported by resources and renowned faculty from the Charles Bronfman Institute for Personalized Medicine at the Icahn School of Medicine at Mount Sinai, this multimodal study will provide data-driven insight into preventing SCA and establishing a clinical and research foundation for an MD/PhD candidate.