PROJECT SUMMARY
Delayed intervention for congenital heart defect residua and sequelae can lead to heart failure and end-organ
damage. Identifying the optimal time for intervention to avoid both adverse outcomes and to minimize the
number of interventions over a lifetime for any given heart defect relies on routine surveillance with expensive
imaging and centralized expertise. Data from the widely available electrocardiogram (ECG) can be transferred
from point-of-care to remote data analysis centers. Machine learning and time-series analysis of ECG waves
related to atrial depolarization, ventricular depolarization and ventricular repolarization, conduction intervals
and waveform durations can consistently calculate parameters that can be tracked as biomarkers
longitudinally. ECG patterns may reflect dysrhythmia, ischemia, hypertrophy, chamber dilatation, ventricular
fibrosis and dysfunction, and can change over time in response to subclinical changes in the cardiac
chambers. A significant problem in detecting subtle changes in the ECG is the reliance on normal intervals and
pattern descriptions that lack nuance to detect longitudinal changes on an individual basis that may reflect
impending ventricular failure. We propose to apply artificial intelligence-based methods to analyze longitudinal
ECG changes by age, sex, race and ethnicity in an adult congenital heart disease population. Once we identify
and characterize ECG changes over time, we will use the change in ECG parameters to develop a machine
learning algorithm to predict the need for cardiac intervention or occurrence of adverse events.
