Logo image
SURFACE ECG-BASED MACHINE LEARNING MODEL FOR PREDICTING PATIENT SUBGROUP AT A HIGH RISK FOR MAJOR ADVERSE CARDIAC EVENTS
Abstract   Open access   Peer reviewed

SURFACE ECG-BASED MACHINE LEARNING MODEL FOR PREDICTING PATIENT SUBGROUP AT A HIGH RISK FOR MAJOR ADVERSE CARDIAC EVENTS

Heenaben Patel, Naveena Yanamala, Marton Tokodi, Nobuyaki Kagiyama, Marco Piccirilli, Sirish Shrestha, Peter Farjo, Grace Casaclang-Verzosa, Wadea Tarhuni, Negin Nezarat, …
Journal of the American College of Cardiology, Vol.77(18 Supplement 1), pp.3227-3227
05/11/2021
DOI: 10.1016/S0735-1097(21)04582-4
url
https://doi.org/10.1016/S0735-1097(21)04582-4View
Published (Version of record) Open Access

Abstract

Background This study explores the feasibility of using surface electrocardiograms (ECG) as a diagnostic tool for predicting echocardiographically defined patients subgroups at a high-risk of major cardiac adverse events (MACE). Methods A total of 1,166 ECG features including signal-processed (spECG) and conventional ECG data were obtained in 1,461 patients as part of multicenter prospective study conducted at 4 institutions in North America. Patients from 3 institutions (n = 943) formed an internal cohort and data from the fourth institution was reserved as an external test set (n = 518). Patients subgroups delineated using high-throughput cardiac imaging data were used to train an ECG based bagged decision tree (DT) model to identify patient groups at low or high risk for MACE. Results The DT model showed robust classification of patients with an area under the receiver operating curve (AUC) of 76% (precision = 73% & recall = 72%) and 78% (precision = 71% & recall = 68%) for the internal and external testset respectively (Fig 1A). A combination of spECG and conventional ECG features such as depolarization average measures in lead V6, V1, I, and aVR, and LVH score were important for the classification and provided a robust prediction of MACE in the external test set (Fig 1B). Conclusion ECG-derived machine-learning models provide a cost-effective strategy for predicting echocardiographically defined patients subgroups at a high-risk of MACE and may aid in optimizing intervention strategies.

Details

Metrics

10 Record Views
Logo image