Effect of Sleep on Heart Rate Variability - Semantic Scholar

1

Effect of Sleep on Heart Rate Variability Payel Ghosh1

Abstract— The paper compares the heart rate variability, of seven subjects in two different states: healthy young adults when awake, and asleep. The spectral analysis of their heart rates showed that the low frequency power as well as the high frequency power was lower when the subjects were asleep. However, the ratio of low to high frequency power was found to be comparable for both the states. This was expected because the subjects were in a resting state either lying down (when asleep) or sitting (when awake), which reduced the heart rate and increased the dominance of the low frequency power. Further investigation of heart rate variability of subjects in different postures should be performed to better understand the physiological effects of sleep. Index Terms— Spectral Analysis, Heart Rate Variability, Sleep.

I. INTRODUCTION

S

LEEP is a dynamic state of consciousness characterized by rapid variations in the activity of the autonomic nervous system. The understanding of autonomic activity during sleep is based on observations of the heart rate and blood pressure variability. However, conflicting results exist about the neural mechanisms responsible for heart rate variability (HRV) during sleep. Zemaityte and colleagues [1], found that the heart rate decreased and the respiratory sinus arrhythmia increased during sleep. On the other hand, two other studies [2, 3] revealed high sympathetic activity during deep sleep. Thus the complete understanding of the autonomic activity during sleep is still elusive. Moreover, there are five different sleep stages and the HRV varies differently with these stages of sleep [4]. The comparison of HRV during different sleep stages is outside the scope of this paper. The causes of heart rate variability have been widely researched because of its ability to predict heart attack and survivability after an attack. The power spectral density (PSD) of the heart rate has been found to vary with the rate of respiration, changes in the blood pressure, as well psychosocial conditions such as anxiety and stress [5, 6]. All these phenomena are in turn related to the activity of the autonomic nervous system. The objective of this project was to understand how sleep affects the body in terms of what is already understood about the heart rate PSD. This was performed by comparing the shapes of the heart rate PSDs of seven subjects in awake and asleep states.

1 This work was completed as part of a class project for Statistical Signal Processing at Portland State University during Fall 2005.

II. BACKGROUND A. Description of Data The data for this analysis was obtained from Physionet data archives [7], an online archive of a wide variety of physiological signals. The data consisted of heart rate time series of seven subjects in awake and asleep states, for 30- to 60-minutes duration each. The subjects were sitting when awake, and were lying down when asleep. The original electrocardiogram (ECG) signatures for the data were not available. All the subjects under observation were healthy with comparable overall health conditions. B. Heart Rate Heart rate is defined as the number of times per minute that the heart contracts. The heart rate fluctuates with two branches of the autonomic nervous system, the parasympathetic and sympathetic. The parasympathetic activity strives to conserve the energy of the body by reducing the blood pressure, heart rate etc., and becomes more dominant at resting states. The sympathetic activity dominates the body functions during increased physical activity and results in an increase in respiration, heart rate, and blood pressure. Figures 1 shows the PSD of the heart rate at resting conditions. Note that the two peaks, one at the low frequency band (0.04-0.15 Hz) and the other at the high frequencies (0.15-0.5 Hz), correspond to the parasympathetic and sympathetic activities respectively.

Fig. 1. Power spectral density of the heart rate at resting conditions. The low frequency peak corresponds to parasympathetic activity, while the high frequency peak is due to sympathetic activity of the autonomic nervous system.

2 III. PREPROCESSING THE DATA A. RR to HR Conversion The raw data collected from the ECG contains inter beat interval signatures (RR). The RR data is converted to the heart rate (HR) using the formula, 1 (1) HR = [ ET (n) − ET (n − 1)] * 60 where ET is the elapsed time between two successive beats in seconds. However, this step was not performed in this analysis because the heart rate time series was already available. B. Uniform Sampling The original heart rate time series was non-uniformly sampled at approximately 1 Hz. This data was uniformly resampled at 2 Hz using the csaps function of Matlab [8] that performs a cubic-spline interpolation between existing data points (code provided by Dr. McNames). Figure 2 shows the uniformly sampled points for a 1-minute segment of data.

window. The variance and bias of the estimate depend on the choice of the window as well as the window length. The window length used here was chosen after experimenting with several other window lengths until an acceptable bias variance tradeoff was achieved. The Matlab code provided in [8] has been used for this analysis. The PSD for each of the 7 subjects in two different states were found. The mean estimate and confidence intervals (90%) were derived using 5-minute non-overlapping segments of the data. Low frequency power (LFP) was calculated as the area under the mean PSD curve within the frequency range of 0.04-0.15 Hz. The high frequency power (HFP) was calculated similarly in a range of 0.15-0.5 Hz. The ratio of the low frequency power to the high frequency power was also calculated because it shows the balance in the sympathetic and parasympathetic activities of the body. B. Discussion The processes responsible for the variation in heart rate are stochastic, therefore the data cannot be considered stationary or ergodic. However within a small time range (5-minutes for this analysis) the processes can be assumed to remain unchanged. Thus, the data can be considered locally stationary in this time range. However the data is still not ergodic. V. RESULTS

Fig. 2. Heart rate time series uniformly re-sampled data at 2 Hz (in blue). The red dots show the original data points.

IV. METHODOLOGY A. Blackman-Tukey Periodogram Estimate The power spectral density estimate was found using the Blackman-Tukey method of periodogram estimation. It was performed in three steps. At first the autocorrelation sequence of the un-windowed data was estimated. Then the autocorrelation sequence was windowed with a Parzen window of 75 data points (40s duration). The Fourier transform of the windowed auto-correlation sequence gave the periodogram estimate of the data. Specifically, the BlackmanTukey spectrum estimate is given by [8, 9]

PˆBT (e jw ) =

M

∑ rˆ(k )w(k )e

− jkw

(2)

k =− M

where w(k) is the lag window that is applied to the autocorrelation estimate rˆ(k ) , and M is the length of the lag

Figures 3 through 12 show the PSD estimates for five subjects in awake and asleep states. Due to space limitations the PSD estimates for two other subjects could not be shown here. The mean estimate, confidence intervals and a single estimate are shown in each figure. Note that the low frequency component (parasympathetic activity) is dominant in all of the images. This is expected because the subjects were in the resting state, (either sitting or lying down) when the data was acquired. Table 1 shows the LFP and HFP values for each of the seven subjects. The ratio of LFP and HFP is also tabulated. Observe that the LFP and the HFP values are higher for all the subjects when awake. The average value for the LFP and HFP ratio is comparable for the awake and asleep states. This is again due to the resting posture of the subjects. TABLE I LFP HFP VALUES FOR SEVEN SUBJECTS

3

Fig. 3. Heart rate PSD estimates (single and average estimate) and confidence intervals for subject 1 in asleep state.

Fig. 6. Heart rate PSD estimates (single and average estimate) and confidence intervals for subject 1 in awake state.





4





VI. CONCLUSION The low frequency power as well as the high frequency power was observed to reduce during sleep. This can be inferred as the slowing down of bodily processes during sleep. The balance in the sympathetic and parasympathetic activity seemed to remain almost unchanged which was due to the resting posture of the subjects in awake as well as asleep states. Experiments with more number of subjects (in different postures) needs to be performed before any physiological effects can be correlated with sleep.

REFERENCES [1] [2] [3] [4] [5]

[6]

ACKNOWLEDGMENT I am thankful to Dr. James McNames for the insightful lectures on statistical signal processing. The code provided by him for pre-processing the data was very helpful for this analysis. I am thankful to the anonymous reviewers for their valuable comments on this paper.

[7] [8] [9]

Zemaityte D, Varoneckas G, Sokolov E. Heart rhythm control during sleep. Psychophysiology. 1984;21:279-289. Hornyak M, Cejnar M, Elam M, Matousek M, Wallin BG. Sympathetic muscle nerve activity during sleep in man. Brain. 1991;114:1281-1295. Somers VK, Dyken ME, Mark AL, Abboud FM. Sympathetic nerve activity during sleep in normal subjects. N Engl J Med. 1993,303-307. Kawachi I, Sparrow D, Vokonas PS, Weiss ST., “ Decreased heart rate variability in men with phobic anxiety.” Am J Cardiol 1995; 75:882-885. Mittleman MA, Maclure M, Sherwood JB, Mulry RP, Tofler GH, Jacobs SC, Friedman R, Benson H, Muller JE. “Triggering of acute myocardial infarction onset by episodes of anger,” Circulation 1995, 92: 1720-1725. Baust W, Bohnert B. , “The regulation of heart rate during sleep,” Exp Brain Res. 1969;7:169-180. “Physiologic signal archives for biomedical research”, http://www.physionet.org/physiobank/database/ M. H. Hayes, “Statistical digital signal processing and modeling”, John Wiley & Sons, 1996. D. Manolakis, V. Ingle, S. Kogon, “Statistical and adaptive signal processing,” Artechouse Pub. 2005.