Session S82.2

Implicit Comparison of Accuracy of Heart Rate Variability Spectral Measures Estimated via Heart Rate and Heart Period Signals

AI Maistrou*

Technische Universität Munich
Munich, Germany

Heart rate variability (HRV) is widely used as a non-invasive informative marker of activity of the autonomic nervous system. Its spectrum analysis is suitable for short-term analysis and spectrum indexes most commonly used in practice. HRV could be analyzed via heart period (HP) and heart rate (HR) time series. Power spectral density (PSD) estimation methods usually used for HRV spectrum analysis. Number of papers showed that HRV spectrum indexes could seriously very depending on chosen time series used for PSD estimation. Detailed analysis of HRV indexes acquired using HR and HP signals was performed by J. Mateo and P. Laguna. They showed that HR signals generated by an integral pulse frequency modulation (IPFM) model were more adequate for HRV spectrum measures estimation. Some authors published results on real data, but unfortunately no analysis on real data and direct instructions for choosing between HR and HP signals were presented. This analysis is objective of current research.
Theoretical analysis as well as numerical simulation of HR and corresponding HP signals were used to find dependency between its PSD functions. The HR and HP series' PSDs were normalized using Euclidean norm and its ratio was calculated. 1000 runs Monte-Carlo simulations for various HR and HP series were used and it was detected that ratio function of HR to HP normalized PSDs has statistically significant trend. A slope of linear component of the trend changes the sign depending on time series initially chosen for modeling: if we initially model HP signal and HR signal is mathematically derived from originally formed HP signal, then the slope of linear component of the trend of ratio function has positive sign; if we initially model HR signal, then the slope has negative sign. It was shown that this trend was caused by high-frequency harmonics arose because of nonlinear transformation from HR to HP signal.
For analysis on real data we used “Fantasia” database (Physionet.org). For all records ectopic beats were excluded, spline interpolation and 10 Hz discretization were applied, 5 minutes windows with 30 seconds shifts between it were extracted and Welch FFT PSD estimation was used. Statistical analysis resulted in negative sign of analyzed slope. It means that HR signal is originally modulated signal. Differences in spectral measures increased with overall HRV increasing. LF/HF index had maximal difference in between HR and HP analysis – standard deviation of difference was 30.8% for group of young volunteers.
In the research implicit way of comparing accuracies of HRV spectral measures estimation derived from HR and HP signals was presented. Results on real data shows that HR signal is more accurate then HP signal that agrees with previous studies on artificial data. Statistical analysis of indexes' differences shows that it is impossible to interchange results of HRV spectrum analysis acquired using HR and HP signals.

(Abstract Control Number: 176)