Entropy-Based Analysis of Heart Rate Variability in Congestive Heart Failure

Abstract

Heart rate variability (HRV) is a physiological phenomenon in which the time between heartbeats fluctuates as the body adjusts to maintain homeostasis. HRV is evaluated with a wide variety of metrics. In this thesis, we focus on entropy-based metrics that attempt to describe the complexity and regularity of the data. The methods of interest are approximate entropy, sample entropy and multiscale entropy. Recent studies suggest that these metrics can support the assessment of pathological diseases such as congestive heart failure. However, these metrics have not yet been thoroughly compared to other HRV metrics to differentiate healthy subjects from those with congestive heart failure.

In this thesis, we evaluated the binary classification performance of multiple HRV metrics using the receiver operating characteristic and the area under curve (AUC) value. The metrics were evaluated for healthy and congestive heart failure patient data from PhysioNet. The entropy metrics obtained significantly lower AUC values than most of the other analyzed metrics. Contrary to multiple studies, the entropy metric values were higher for congestive heart failure patients compared to healthy subjects in this study. We suspect that these results and the poor classification performance are a result of the unsupervised 24-hour measurement setting. Thus, the results highlight the importance of well-defined, consistent and comparable protocol or baseline measurement. This aspect will be further explored in future studies.