Cardiac Autonomic Modulation - IEEE Xplore

CARDIAC RHYTHMS

Cardiac Autonomic Modulation Analyzing Circadian and Ultradian Rhythms ©GETTY IMAGES

BY PHYLLIS K. STEIN, ERIC J. LUNDEQUAM, LEONARDO P.J. OLIVEIRA, DANIEL J. CLAUW, KENNETH E. FREEDLAND, ROBERT M. CARNEY, AND PETER P. DOMITROVICH

T

he amount of time between heartbeats is controlled by the rate of depolarization of the sinoatrial node. This rate is controlled by a combination of sympathetic and parasympathetic inputs. For this reason, analysis of heart rate (HR) and heart rate variability (HRV) derived from intervals between normal heart beats provides insights into changing cardiac autonomic function [1]. Increased sympathetic control of the heart from one time point to the next tends to increase HRs, decrease the amount of vagally modulated sinus arrhythmia measured as high-frequency (HF) spectral power, and increase the ratio between low-frequency (LF) spectral power and HF (the LF/HF ratio). Conversely, a shift toward greater vagal control of the heart tends to produce decreased HRs, increases in HF power, and decreases the LF/HF ratio. Clear circadian cycles of HR and HRV are seen in most subjects and correspond to periods of sleep and activity. The magnitude of these cycles is well captured by traditional HRV indices such as deviation of all normalto-normal (NN) intervals. Cosinor analysis also provides a rough estimate of the magnitude of circadian rhythms [2]. We have observed that HR and HRV fluctuate over time in a cyclic manner in most subjects. Although these cycles are especially prominent during the nighttime, they are seen in the daytime as well. We have developed a number of measures that quantify ultradian properties of HR/HRV cycles and would permit comparisons between subjects [3]. We applied these measures during overnight polysomnography and have shown that ultradian rhythms of cardiac autonomic modulation have a variable and often weak correspondence with traditional sleep stages [4]. In the current study, we applied the methodology developed for the nighttime recordings to 24-h recordings in young healthy subjects and compared these results to those obtained during the nighttime only in the same subjects and in an older group of cardiac patients. Our purpose was to explore the presence and magnitude of ultradian cycles of HRV during the entire 24-h period and during sleep.

Materials and Methods Subjects

Holter data were examined from 30 randomly selected healthy young subjects (16 males, 14 females, aged 38 10 years). Digital Object Identifier 10.1109/MEMB.2007.907092

14

IEEE ENGINEERING IN MEDICINE AND BIOLOGY MAGAZINE

All subjects were in normal sinus rhythm. Bed and wake times were estimated from the Holter recordings based on changes in HR and HRV patterns. Subjects were participating in an in-patient protocol and all followed the same schedule. Sleeptime only data were previously obtained from polysomnograms in 113 subjects (65 males, 48 females, aged 58 10 years) selected from a study of depression and sleep apnea in patients with known coronary artery disease [4]. Extraction of Interbeat Intervals

Holter recordings were scanned to research standards on a MARS 8000 Holter analyzer. Polysomnograms are stored in European Data Format (EDF) and include one channel of electrocardiogram (ECG). ECGs were extracted from Alice 4 overnight sleep studies using a custom-made program written in Unix/C [5]. The ECG signal in EDF format was converted to a two-channel RAW format that can be read by the MARS 8000 Holter scanner using software provided by the manufacturer. To generate the annotated beat-to-beat file for HRV analysis, each RAW file was loaded onto the Holter scanner as a duplicate single-channel recording, and the ECG was analyzed using standard research Holter analysis techniques. Specifically, careful attention was paid to accurate beat notation and to the uniformity of the detection of the onset of each sinus beat. The longest and shortest true NN intervals were identified for each recording, and intervals outside of these limits were excluded from all calculations. Atrial and ventricular premature beats were also excluded. After editing, the labeled QRS data stream was transferred to a Sun Enterprise 450 server for time domain and frequency domain HRV analysis. Usable two-min segments were required to have 80% NN interbeat intervals for frequency domain and 50% NN intervals for time domain HRV calculations. HRV Pattern Analysis

Two-minute averaged HR, standard deviation of NN intervals (SDNN2), and HF power (0.15–0.4 Hz) and the LF power (0.04–0.15 Hz) to LF/HF ratio were calculated. These were plotted in four simultaneous windows using MATLAB. The best-fit curve for each HR/HRV index was separately plotted over the raw data for each window using a custom filter designed around both the filtfilt and filter algorithms built into MATLAB. To keep the ends of the

0739-5175/07/$25.00©2007IEEE

NOVEMBER/DECEMBER 2007

HRV Pattern Measures

For each HRV variable, the number of cycles, number per hour, the average cycle peak amplitude (e.g., in beats/min. for HR), the average and standard deviation of cycle lengths in seconds, the average change in amplitude from start to peak and peak to end, the slope of the line from start to peak (positive slope) and peak to trough (negative slope), and the average area under the curve were measured. Results from 24-h analyses were compared with those obtained from the sleeptime polysomnograms. Statistics

Paired t-tests and correlational analysis were used to compare various HR/HR parameters. The software used was SPSS 13.0. Results 24-H Ultradian HR/HRV Cycles

Table 1 shows selected results of the 24-h ultradian HRV analyses. Patterns similar to those shown in Figure 1 were seen for all subjects. Correlations for the numbers of cycles were

Average HR (Nn)

filtered data from determining the start and end points of the HRV cycles, the first and last sections were filtered separately and then reattached to the original data. This reconstructed series was then filtered using the filtfilt algorithm that processes the signal both forward and backward, using default settings. A trend line was then created using the filtfilt algorithm over one-fourth of the length of the dataset. After the HRV cycle curves were created, the time and amplitude of the beginning, peak, and end of every cycle for each HRV parameter were determined by automatic detection using trend-line crossing criteria. Precrossing minima, postcrossing maxima, and postpeak minima were used to identify the actual start, peak, and end of each cycle. This was followed by manual overreading to add or delete peak or trough markers based on visual inspection. To adjust for incomplete cycles and to permit comparisons between subjects, cycles beginning at a peak at the start of the recording and cycles ending with a peak at the end of the recording were excluded from the analysis. An example of ultradian rhythm detection is shown in Figure 1.

150 100 50 250 200

SdNn

150 100 50

Stage

0

× 104 2.5 2

hf

1.0 1 0.5 0 15

If/hf

10 5 0 00:00 CMIA0183

07:59 Time

OK

00:00:00

23:59:00

00:00:00 Roll 1/4

Filter ES

15:59 CMA0183Jm/bf Load

Save

23:59

Crose

Need Fig

Batch Print

4x or 8x

Save Fig

Zoom

Print-Need Fig

Stage on/off

Fig. 1. Ultradian rhythms of HR, SDNN2, HF, and LF/HF ratio for a typical young, healthy subject. Note that rhythms tend to be of greater magnitude during the nighttime but persist during the entire recording. The sharp peak in HR during the daytime occurred during a bout of maximal exercise.



15

all significant. Correlations ranged for r ¼ 0.40 between the LF/HF ratio and HR to r ¼ 0.57 between LF/HF and HF power. There was no relationship between the magnitude of HR cycles and the magnitude of any HRV cycles. The strongest correlation of cycle magnitudes was between SDNN2 and HF power (r ¼ 0.81, p < 0.001). There was no correlation between any of the cycle durations. Correlations between positive and negative slopes of the ultradian cycles ranged from 0.38 for HR (p ¼ 0.039) to 0.94 for the HRV indices in the Holter cohort. Number of Nighttime Ultradian Cycles

Correlations for the numbers of cycles were all significant among the cardiac patients. They ranged from r ¼ 0.19 between HR and the LF/HF ratio to r ¼ 0.51 between SDNN2

and HF power. Among the healthy subjects, correlations for the number of cycles were moderately strong for SDNN2 and HF (r ¼ 0.61) and for SDNN2 and LF/HF (r ¼ 0.72) but were not otherwise significant. The correlation between LF/HF and HF was similar between groups (r ¼ 0.36 for the cardiac patients and r ¼ 0.32, p ¼ 0.086 among the healthy subjects). Nighttime Ultradian Cycle Amplitudes

Figure 2 is an example of an analysis of an overnight sleep study. In this case, HR cycles had a rough correspondence with wake or REM. Table 2 shows results from overnight polysomnograms in our previous study [2] and from estimated inbed time for the Holter data. When the correlations between cycle amplitudes of the HR/HRV indices during sleep time were explored among the cardiac patients, the strongest

Table 1. Representative ultradian HR/HR patterns measured from 24-h Holter recordings in healthy adults. Number of Cycles (Range) Heart Rate (beats/min.) SDNN2 (ms2) HF Power (ms2) LF/HF ratio

7–24 8–34 6–28 8–31

Number of Cycles/h 0.7 1.1 0.9 1.0

Mean Amplitude (Units)

0.1 0.2 0.2 0.2

12.9 28.4 681 3.8

2.8 10.8 914 1.9

Mean Duration (s) 79.5 52.5 65.5 58.0

11.5 8.2 9.4 8.2

0.4 1.3 26.2 0.16

0.1 0.5 32.5 0.08

66

1,800

64

1,600

62

1,400

60

1,200

58 1,000

56

hf

Average HR (Nn)

Mean Positive Slope (Units/s)

800

54

600

52 50

400

48

200

46

0

200 16

180

14

160

12

140 sdNn

8

100

6

80

4

60

2

40 20 22:47

If/hf

10

120

0 00:32

02:18

04:04

05:50 22:47

00:32

02:18

04:04

05:50

Fig. 2. Ultradian cycles of HR and HRV from overnight polysomnography. Darkened bars correspond to REM periods. Single darker bars reflect a wake period.

16



Table 2. Representative HR/HR pattern measured during sleep. Number of Cycles (Range)

Number of Cycles/h

Mean Amplitude (Units)

Mean Duration (s)

Mean Positive Slope (Units/s)

Middle-Aged Cardiac Patients (N ¼ 113) Heart Rate (beats/min.) 3–12 SDNN2 (ms2) 4–13 HF Power (ms2) 3–12 LF/HF ratio 3–12

0.97 1.11 0.99 1.01

0.21 0.24 0.27 0.23

8.7 61 620 7.1

3.2 29 1,091 4.3

50.2 44.9 47.5 48.2

10.4 9.0 10.7 9.5

0.5 4.1 33.9 0.5

0.3 2.4 57.1 0.4

Healthy Adults (N ¼ 30) Heart Rate (beats/min.) SDNN2 (ms2) HF Power (ms2) LF/HF ratio

1.12 1.44 1.2 1.28

0.33 .27 0.22 0.31

6.1 30.6 761 3.0

2.7 1.1 1,082 2.6

56.2 42.4 50.3 49.3

14.2 14.0 9.9 14.0

0.2 1.7 31.2 0.3

0.1 0.8 50.3 0.2

3–14 5–15 4–14 3–12

correlation in amplitudes (r ¼ 0.41) was between the amplitudes of SDNN2 and HF. In addition, the amplitudes of SDNN2 and HR were correlated (r ¼ 0.35), as were the amplitudes of HF and HR cycles (r ¼ 0.29), but none of the other correlations were significant for cycle amplitude. In contrast, among the healthy younger subjects, there were no significant correlations between cycle amplitudes. Nighttime Ultradian Cycle Durations

As previously reported [2], among the cardiac patients, the correlation in cycle lengths between SDNN2 and HF was the strongest (r ¼ 0.39). A similar correlation was seen in the healthy subjects (r ¼ 0.48 for duration of SDNN2 and HF cycles). Weaker, but significant, correlations were also seen among cardiac patients for the durations of HR and SDNN2 (r ¼ 0.24, p ¼ 0.008) and SDNN2 and LFHF (r ¼ 0.24, p ¼ 0.008) and HF and LF/HF (r ¼ 0.21, p ¼ 0.019). However, among the healthy subjects, correlations for HR and SDNN2 cycle durations (r ¼ 0.70) and HF and LFHF cycle durations (r ¼ 0.74) were far stronger. The magnitude of the correlation between SDNN2 and LF/HF was greater among the healthy subjects (r ¼ 0.34) than the cardiac patients (r ¼ 0.21) but did not attain statistical significance. As seen in Table 2, values for the number of cycles and cycle durations were similar between the two groups studied, although most cycle amplitudes and slopes were different.

when only sleep time was considered. In further analyses, we will explore correlations of measures during awake time and asleep time separately to see that values are more correlated when daytime is evaluated separately. We are in the process of applying this new algorithm to other datasets, including a series of Holter recordings and overnight polysomnograms over a time period of up to 14 years in individuals who were participants in both the Cardiovascular Health Study and the Sleep Heart Health Study. Further developments in curve fitting and cycle identification and an exploration of other HRV measures, including nonlinear ones, and averaging periods should provide additional insight into the utility and optimal measurement of these ultradian rhythms of HR/HRV, their underlying physiology, and their potential use for comparison between subjects. Conclusions

Results suggest that quantifiable ultradian rhythms of HR and HRV, not detected by conventional methods, are present during 24-h, as well as nighttime ECG recordings and may provide significant new information about cardiac autonomic function. Further study will optimize measurement techniques and elucidate the significance of differences between subjects. Acknowledgment

Discussion

This work was supported in part by the NHLBI RO1HL65356 and a grant from the DOD.

Results verify that there are measurable ultradian cycles of HR and HRV during wake as well as sleep periods [4]. The similarity in the mean cycle durations at night (roughly 50 min) between the two groups studied was remarkable. In the younger healthy group, we observed a minimum of six and a maximum of 34 ultradian cycles in 24 h across the HR/HRV indices assessed. As can be seen in Figure 1, these cycles are not synchronous, although relatively strong correspondences were seen for the duration of SDNN2 cycles and HR and LFHF cycles, with a smaller correlation with HF power cycles. Also, as seen in Figure 2, in many individuals during the nighttime, these cycles are not surrogates for sleep stages. Thus, results suggest that ultradian cycles of HR/HRV provide significant new details of cardiac autonomic function that could lead to new insights into normal and abnormal cardiac autonomic control. Relationships between the number, duration, and amplitudes of the HR/HRV cycles in the young healthy group differed when the entire 24-h recording was considered versus

Phyllis K. Stein is a research associate professor of medicine at the Washington University School of Medicine, St. Louis, Missouri, and director of the Washington University Heart Rate Variability Laboratory, specializing in research-quality Holter scanning for both grant- and industryfunded studies, state-of-the-art HRV analysis, and novel ECG-based analysis, including analyses from overnight polysomnography. Her research interests are focused on the use of HRV and other ECG-based measures, including twave alternans as markers of risk and benefit. She is also interested in HR and HRV patterns and their relationship to sleep. Current projects include HRV in the cardiovascular health study, in depressed and nondepressed cardiac patients, and in participants with Gulf War illness. She is the author of more than 80 research papers, invited papers, and book chapters.



17

Eric J. Lundequam received the bachelor’s degree in mechanical engineering from the University of Minnesota in 2002. He worked as a research assistant at Washington University of St. Louis while working on this article. He specialized in programming and data analysis using MATLAB software and has taken a position with Boston Scientific working with drug-eluting stents. Leonardo P.J. Oliveira is a senior year medical student of the Universidade Federal Fluminense in Brazil, where he has previously worked at the Laboratory of Exercise Sciences. Recently, he became a research assistant of the Heart Variability Laboratory of the Washington University in St. Louis. His research interests include autonomic nervous system regulation through the evaluation of HRV parameters and arterial baroreflex sensitivity in addition to electrocardiography and risk stratification in cardiovascular diseases. Daniel J. Clauw received his undergraduate training and attended Medical School, at the University of Michigan. He completed both an internal medicine residency and rheumatology fellowship in 1990 at Georgetown University, where he then became a faculty member. In 1997, he became the chief of the division of rheumatology, immunology, and allergy and director of the rheumatology fellowship training program. He served as the director of Chronic Pain and Fatigue Research Center, and in 2001 he moved the Center to the University of Michigan. He is a professor of medicine in the division of rheumatology, University of Michigan. He also serves as the director of the Center for Advancement of Clinical Research and assistant dean for clinical and translational research. He has been the principal investigator of both NIH and Department of Defense grants studying this spectrum of illness continuously since 1994. He received the Senior Fellow Award from the American College of Rheumatology in 1990 and the Research Breakthroughs Award from the Washington DC Chapter of the Arthritis Foundation in 2001. He is the author of more than 70 peer-reviewed publications in journals such as New England Journal of Medicine, Annals of Internal Medicine, the Lancet, and Journal of the American Medical Association. Kenneth E. Freedland is a clinical psychologist and professor of psychiatry at the Washington University School of Medicine, St. Louis. He has been an investigator and clinical supervisor on numerous observational studies and clinical trials involving depression and related problems in patients with heart disease. He has participated in more than 160 scientific conference presentations and authored or coauthored more than 130 articles for scientific journals, five book chapters, and a recently published book on the assessment and treatment of psychological problems in patients with heart disease. He is a fellow of the Society of

18


Behavioral Medicine and the Academy of Cognitive Therapy, the Councils of the American Psychosomatic Society, and the Academy of Behavioral Medicine Research. He is on the editorial board of Health Psychology. He is an associate editor of Psychosomatic Medicine and for Behavioral Medicine for the Hogrefe series on Advances in Psychotherapy: Evidence-Based Practice. Robert M. Carney, is a professor of psychiatry and the director of the Behavioral Medicine Center at the Washington University School of Medicine. He received his undergraduate degree in psychology from the University of Missouri and his doctorate in counseling psychology from Washington University. He is a licensed psychologist specializing in cognitive behavior therapy for the treatment of depression and anxiety disorders. He is a fellow of the Society of Behavioral Medicine and the Academy of Behavioral Medicine Research and a founding fellow of the Academy of Cognitive Therapy. He is a member of the editorial boards of Psychosomatic Medicine, Journal of General Hospital Psychiatry, and Journal of Consulting and Clinical Psychology. He is an active researcher in the area of depression and comorbid medical illness. He was a member of the Aging and Medical Comorbidity subcommittee for the development of the National Institute of Mental Health’s strategic plan for mood disorders, and a major contributor to the Agency for Health Care Policy and Research Clinical Practice Guidelines for Treatment of Major Depression in Primary Care. His interest in HRV analysis stems from his work, with depression as a risk factor for cardiac mortality. He has written more than 150 scientific papers, book chapters, and editorials. Peter P. Domitrovich is a staff scientist in the Heart Rate Variability Laboratory of the Washington University School of Medicine, St. Louis, Missouri. He graduated from Washington University in St. Louis Department of Physics with a Ph.D. in many-body theory in 1991. He completed a postdoctoral fellowship in many-body theory at the Institute for Theoretical Physics at the University of Tuebingen, Germany, in 1993. His research interests include HRV in various patient populations. Address for Correspondence: Phyllis K. Stein, Washington University School of Medicine, St. Louis, MO 63108 USA. Phone: þ1 314 286 1350. Fax: 314 286 1394. E-mail: pstein@ im.wustl.edu. References [1] P. K. Stein and R. E. Kleiger, ‘‘Insights from the study of heart rate variability,’’ Annu. Rev. Med., vol. 50, pp. 249–261, 1999. [2] W. Nelson, Y. L. Tong, J. K. Lee, and F. Halberg, ‘‘Methods for cosinorrhythmometry,’’ Chronobiologia, vol. 6, no. 4, pp. 305–323, Oct-Dec 1979. [3] P. K. Stein, E. J. Lundequam, D. J. Clauw, K. E. Freedland, R. M. Carney, and P. P. Domitrovich, ‘‘Circadian and ultradian rhythms in cardiac autonomic modulation,’’ in Proc. 28th IEEE EMBS Annu. Int. Conf., New York 2006, pp. 429–432. [4] P. K. Stein, P. P. Domitrovich, E. J. Lundequam, S. P. Duntley, K. E. Freedland, and R. M. Carney, ‘‘Circadian and ultradian rhythms in heart rate variability,’’ Biomed. Tech. (Berlin), vol. 51, no. 4, pp. 155–158, 2006. [5] B. Kemp, A. Varri, A. C. Rosa, K. D. Nielsen, and J. Gade, ‘‘A simple format for exchange of digitized polygraphic recordings,’’ Electroencephalogr. Clin. Neurophysiol., vol. 82, no. 5, pp. 391–393, May 1992.
