Patient Identification with ECG and SaO2 Time Series
Embargo until
2020-05-01
Date
2019-05-13
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
Sudden cardiac death is the most common cause of death in United States. Primary prevention implantable cardioverter defibrillators (ICDs) have been the first line to reduce mortality for high-risk patients. Previous work of identifying subjects at greater risk is neither sensitive nor specific. The development of more reliable predictors that could help identify patients that could benefit from these devices is of both academic and public health interest.
In this thesis, we study the time series data of both electrocardiogram (ECG) and oxygen saturation (SaO2) signals from patients who received ICD implantation. This study is part of Prospective Observational Study of Implantable Cardioverter Defibrillators (PROSE-ICD).
The features for each subject are generated from some statistics of the ECG and SaO2 signals respectively. For ECG signal, the analysis is from both geometry and dynamics perspective. For SaO2 signal, multivariate and dynamics analysis is applied. Our results showed an overall accuracy of 93.2% for patient classification, with no bias towards healthy or HF patients. Further analysis does not show a clear relationship between ECG and SaO2 signals.
Description
Keywords
Dynamics, Time series, Machine learning, Statistics