Heart Rate Variability During Short-Term Electroencephalography

Figure 1. Marking state changes in an EEG. Relevant events and state changes were manually marked on de-identified EEGs by the investigators, and the marks, times, and ECG trace were exported for further analysis. (A) EEGs were de-identified. (B) Events of interest in the EEG recording were marked with a standardized comment beginning with “#”. (C) In this EEG page, the transition from sleep to awake was marked with the comment “# AWAKE”. (D) The single-channel ECG was exported to a text file in American Standard Code for Information Interchange (ASCII) format.

Figure 2. Measurement of NN intervals. (A) A modified Pan Tompkins algorithm was used to place fiducials (maroon circles) at the peak of the R waves. All the beats are normal, so the measured RR intervals are NN (normal-to-normal) intervals. (B) This person had a premature ventricular contraction (PVC, the third QRS complex), and the RR intervals surrounding it are therefore not normal. A fiducial that our computer algorithm placed at the PVC peak was manually deleted by the investigator, and only the remaining NN intervals were measured. The scale in the top panel is 500 µV x 200 ms, and in the bottom panel is 200 µV x 200 ms.

Figure 3. Examples of heart rate variability (HRV) in five-minute tachograms of awake EEG. Standard HRV measurements are calculated over five-minute epochs. These HRV examples are from the first five-minute awake epoch. (A) This example is from a 21-year-old woman who had the highest HRV of all our subjects. Her minimum heart rate (HR) during these five minutes was 46 beats per minute, maximum was 88, and RMSSD was 133 ms. (B) This example is from a 61-year-old woman who had the lowest HRV of all our subjects. During this five-minute epoch, her minimum HR was 81, maximum was 85, and RMSSD was 4.3 ms. (C) A plot of the successive NN interval differences for the subject in (A) and (D) the subject in (B) emphasizes the marked differences in HRV in these two subjects. The 21-year-old woman had 218 occurrences of a greater-than-50 ms difference between successive NN intervals (NNI 50 = 218), while the 61-year-old woman had 0 occurrences (NNI 50 = 0). RMSSD: root mean square of successive differences; NN: normal-to-normal; NNI: normal-to-normal interval.

Figure 4. Effect of wakefulness and sleep on HRV. (A) There was a statistically significant increase (p < 0.01) in NN intervals (decrease in heart rate) during sleep (N = 95 awake, 76 N1 sleep, 42 N2 sleep). (B) There was a statistically nonsignificant trend for an increase in HRV (RMSSD) during sleep. Box and whisker plots in this and other figures show the mean (black diamond), median (horizontal line), 25th percentile (box bottom), 75th percentile (box top), 25th percentile – 1.5x interquartile range (lower whisker), 75th percentile + 1.5 x interquartile range (top whisker), and outlier data points (circles). HRV: heart rate variability; RMSSD: root mean square of successive differences; NN: normal-to-normal.

Figure 5. HRV during continuous EEG wakefulness and sleep is relatively stable. (A) Normalized RMSSD is stable during continuously maintained wakefulness. (B) In stage N1 sleep, RMSSD increased slightly, but (C) there was no appreciable change during continuously maintained stage N2 sleep. HRV was analyzed in five-minute segments (0-5 minutes, 5-10 minutes, and so on) and was normalized to the value during the initial 0-5 minutes. Error bars are + and - SEM, and if not shown are smaller than symbol size. HRV: heart rate variability; RMSSD: root mean square of successive differences.

Figure 6. Age and sex relationship for time domain measures of HRV. The values shown in each graph are for the first five minutes of the EEG. There were declines in both (A) SDNN and (B) RMSSD with age. A linear trendline for our subjects is shown with a dotted line. There is a highly significant negative correlation (r) between age and SDNN and between age and RMSSD. In (A) and (B), female subjects are indicated by red and male by blue symbols. (C) The age-dependent results per decade from our subjects having a normal EEG are comparable to those of normal subjects studied by Agelink et al. [15] and Abhishekh et al. [16]. (D) More subjects were female (54) than male (49), and the average age of females (53.4) was greater than that of males (average age = 48.8 years). Female subjects had an overall lower RMSSD than male subjects. Two-factor ANCOVA of age and sex effects on RMSSD showed a highly significant effect of age, but the effect of sex was not statistically significant. HRV: heart rate variability; SDNN: standard deviation of normal-to-normal intervals; RMSSD: root mean square of successive differences; ANCOVA: analysis of covariance.

Figure 7. Illustration of frequency domain measures of HRV. These are Welch periodograms for the same two subjects as in Figure 3. (A) For the subject with high heart rate variability (HRV), power in the low frequency (LF, green) band is greater than power in the high frequency (HF, orange) band. The LF/HF ratio is normal. Very low frequency (VLF, blue) power is even higher, but VLF measurement is only considered reliable in 24-hour ECG recordings (VLF power greater than 0.5 is not shown). (B) For the subject with abnormally low HRV, power is very low in all frequency bands. Note the 10-fold difference in scale for the two graphs.

Figure 8. Frequency domain measures of HRV awake and asleep. (A) Mean LF power and (B) mean HF power tend to increase with sleep, but the changes are not statistically significant. There is no significant change in (C) the LF/HF ratio with sleep. N = 95 awake, 76 N1 sleep, and 46 N2 sleep for each graph. The data are highly positively skewed, and outliers are not shown. HRV: heart rate variability; LF: low frequency; HF: high frequency.