Association of cardiac automatic function and the development of hypertension

AJH

1996;9.1147-1156

ORIGINAL

COMMUNICATIONS

Association of Cardiac Autonomic Function and the Development of Hypertension The ARIC Study Duanping Liao, Jianwen Cai, Ralph W. Barnes, Herman A. Tyroler, Pentti Rautaharju, Ingar Hohne, and Gerardo Heiss

To relate cardiac autonomic function measured by heart rate variability (HRV) with prevalent and incident hypertension at the population level, the authors examined a stratified random sample of 2,061 examinees from the biracial Atherosclerosis Risk in Communities (ARIC) cohort. Baseline, supine, resting beat-to-beat heart rate data were collected. High frequency (HF, 0.15 to 0.35 Hz), low frequency (LF, 0.025 to 0.15 Hz) spectral powers, and LF/HF ratio, estimated from spectral analysis, and standard deviation of all normal RR intervals ( S D N N ) , calculated from time domain analysis, were used as the conventional indices of cardiac autonomic function. From this sample, 650 prevalent hypertensives were identified. Of those normotensive at baseline (n = 1,338), 64 participants developed hypertension during 3 years of follow-up. In the cross-sectional analysis, the adjusted geometric means of HF were 1.26, 1.20, and 1.00 (beat/min) 2 for normotensives, untreated hypertensives, and treated hypertensives, respectively; means of LF were

3.24, 3.26, and 2.58; means of LF/HF ratio were 2.57, 2.70, and 2.56; and means of SDNN were 39, 34, and 35 (ms) respectively. In the prospective analysis, a statistically significant, graded inverse association between baseline HF and the risk of incident hypertension was observed: the adjusted incident odds ratios (95% CI) were 1.00, 1.46 (0.61, 3.46), 1.50 (0.65, 3.50) and 2.44 (1.15, 5.20) from the highest to the lowest quartile of HF. No clear pattern of association was observed for LF. Significant trends of association for LF/HF and SDNN and incident hypertension were also found. These results suggest that cardiac autonomic function is associated with prevalent hypertension, and that reduced vagal function and the imbalance of sympatho-vagal function are associated with the risk of developing hypertension. © 1996 A m e r i c a n Jo u r n al of

Received January 12, 1996. Accepted June 3, 1996. From the Department of Epidemlology I DL. HAT, GH) and the Department of Biostahstics (JC), School of Public Health, University of North Carohna at Chapel Hill, Chapel Hill, North Carohna, Department of Neurology (RWB) and ECG Reading Center--EPICARE Center, Department of Public Health Sciences (PR). Bowman-Gray Medical School, Wake Forest University, Winston-Salem, North Carohna; and Life Insurance Companies' Institute for Medical Statistics, Ullevaal Hospital, Ullevaal, Norway (IH) This arhcle was supported in part by National Heart, Lung, and

Blood Institute Contracts N01-HC-55015, N01-HC-55016, N01-HC55018, N01-HC-55019, N01-HC-55020, N01-HC-55021, N01-HC55022. The work was completed while the lead author was a postdoctoral fellow in the Cardaovascular Disease Epidemiology Traming Program supported by National Institutes of Health, National Heart, Lung, and Blood Inshtute NRSA grant qT32HL07055. Address correspondence and reprint requests to Dr. Duanping Liao, Department of EpldemJology, School of Pubhc Health, University of North Carolina at Chapel Hill, 137 E Franklin Street, 306 NationsBank Plaza, Chapel Hill, NC 27514

© 19q6 by the American Journal ot Ht/pertcnsltm. Ltd Published by Elsevtcr Sclencc, lnc

H y p e r t e n s i o n , Ltd. A m J H y p e r t e n s 1996;9:1147-1156

KEY WORDS: Heart rate variability, autonomic function, hypertension, cohort study, race.

0895-7061/96/$15.00 PII c,¢lSq5-7061( 96 )OO249-X

1148

LIAO ET AL

p

hysiological and clinically based studies have demonstrated the involvement of autonomic function in the development and the maintenance of human hypertension. 1 Recently, analysis of beat-to-beat heart rate variability has become established as one of the noninvasive methods to assess the influence of the sympathetic and parasympathetic systems on cardiac function. As a result of the interaction between sympathetic and parasympathetic activity, beat-to-beat heart rate shows periodicities over time. These periodicities can be identified through spectral analysis, whereby the observed heart rate is expressed mathematically by a function of time as the sum of a series of sine and cosine functions of varying amplitudes and frequencies (Hz). A plot of the square of the amplitudes of these sine and cosine functions against their cycle frequencies is known as the power spectral density for beat-to-beat heart rate, or the heart rate variability (HRV) power spectrum. When the heart r a t e / t i m e data are re-expressed in this mathematical form, previous work has shown that cycles with a frequency of 0.025 to 0.15 Hz are under the influence of both the sympathetic and parasympathetic nervous systems. Cycles with a frequency of 0.15 to 0.35 Hz are under the influence of the parasympathetic system only, and have been regarded as a marker of cardiac vagal function. The area under the power spectrum for cycles in the 0.025 to 0.15 Hz range is called low frequency power (LF), and in the 0.15 to 0.35 Hz range is called high frequency power (HF). LF represents the contribution to heart rate variability from the sympathetic and parasympathetic system, HF represents the contribution to the variability from the parasympathetic, and the ratio L F / H F represents a measure of the balance of sympathetic and parasympathetic function. 2 10 Several clinically based studies have found that hypertensive patients have a higher HRV-LF and a lower HF, indicating reduced parasympathetic activity and sympathetic predominance. 6'7'1~ However, none of these results can be considered as conclusive. No population based study in this area has been published, nor have data been available to investigate the temporal association of autonomic function and the development of hypertension. Therefore, this study is designed to assess the association of cardiac autonomic function, measured by time domain and spectral analysis of heart rate variability, and prevalent hypertension in a population based sample, and to test the hypothesis that cardiac autonomic function is related to the development of incident hypertension in a prospective mode. POPULATION AND METHODS

Study Population The population for this study was drawn as a stratified random sample from the 15,800

AJH-DECEMBER 1996-VOL 9. NO. 12, PART 1

individuals who participated in the baseline examination of the Atherosclerosis Risk in Communities (ARIC) study. ARIC is a longitudinal study of cardiovascular and pulmonary diseases sponsored by the National Heart, Lung, and Blood Institute. It includes Community Surveillance and a Cohort component. The ARIC cohort was selected as a probability sample of men and women between the ages of 45 and 64 years at four study centers in the United States, three of which enumerated and enrolled an ethnically diverse population (selected Minneapolis suburbs, MN; Washington County, MD; and Forsyth County, NC). The fourth quarter of the ARIC cohort was sampled from black residents of Jackson, MI. Details of sampling, study design, and cohort examination procedures have been published. 12 Eligible participants were interviewed at home, and then invited to a baseline clinical examination. The baseline examination of the ARIC cohort was conducted in 1987 to 1989. Three years after the baseline examination, all participants were invited to a followup clinical examination, which was conducted in 1990 to 1992. The sample for this study was drawn so as to include a high proportion of individuals with intima-media wall thickness indicative of carotid atherosclerosis, a group of individuals with thin walls suggesting that they were free of carotid atherosclerosis but comparable in other attributes to the former, and a random sample of remaining examinees. The sampling process involved dividing the entire ARIC cohort into strata defined by field center, race, and carotid atherosclerosis status. All individuals with a maximum far wall intimamedia thickness suggestive of atherosclerosis according to criteria previously reported 13 were selected, as were individuals of the same field center and race, whose Bmode ultrasound scan of the carotid arteries and one popliteal artery showed no evidence of atherosclerosis, ie, all intima-media measurements were below the 75th percentile of the population distribution. Also included as a sampling stratum were two individuals examined closest in time to each of the individuals identified previously as either having atherosclerosis or being free of atherosclerosis. Because induction cycles for the examination at the ARIC Field Centers were carried out in a randomized order, the latter group, identified as one of the sampling strata, was considered as a random sample of the remainder of the ARIC cohort. From this total of 15 strata, 2,618 individuals were selected as the study population for this analysis. Therefore, the sample for this analysis is treated as a stratified random sample, and the stratum-specific sampling weights were calculated by dividing the total number of the ARIC participants in each stratum by the total number of participants sampled from that stratum. A total of 557 of the 2,618 observations were excluded

AJH-DECEMBER 1996-VOL. 9, NO 12, PART 1

AUTONOMIC FUNCTION AND HYPERTENSION 1149

from this analysis due to one or more of the following reasons: age < 44 years (n = 4); race other than black or white (n = 6); heart rate data collected before May 15, 1987 (n = 7), indicating possible data collection error; prevalent CHD or frequent ventricular/atrial premature beats (n = 361); and a high degree of artifacts in the heart rate record (n = 173), defined by total spectral power > 321 or spectral power at high frequency band (HF) > 27 (beats/minute) 2, (the former corresponds to three times the variance of heart rate in the ARIC cohort, and the latter approximates the upper 96th percentile of the study population); hypertension status missing (n = 6). Furthermore, 723 individuals were excluded from the follow-up analysis either because of hypertension at baseline (n = 650); or a repeat examination could not be scheduled (n = 73; 13 were deceased and 60 did not come to the scheduled follow-up visit). The final sample size for this report is 2,061 (650 prevalent hypertensives and 1,411 normotensives ) for the cross-sectional analysis, and 1,338 ( 64 incident hypertensives and 1,274 normotensives) for the prospective analysis.

squared as the predictors, so that any time trend in the data was removed by taking the residuals. From the residuals, Fast Fourier Transformation (FFT) was performed to estimate the Fourier transformation of the heart rate residuals, from which the power spectral density (PSD) was computed. An example of the PSD curve, following the FFT, with corresponding time domain beat-to-beat heart rate data inserted, is shown in Figure 1 for one participant. From the PSD curve, the high frequency spectral power component (HF) and the low frequency spectral power component (LF), defined as the power (area) between 0.15 and 0.35 Hz, and 0.025 and 0.15 Hz, bands under the PSD curve, respectively, were calculated based on a rectangular method. The ratio of LF to HF ( L F / H F ) was also derived. The standard deviation of all normal RR intervals (SDNN) was calculated from the time domain data after elimination of artifacts. The short-term intraparticipant reliability coefficients were 0.82, 0.56, 0.64, and 0.70 for HF, LF, L F / H F , and SDNN, respectively. The intra- and interreader reliability coefficients for all four HRV indices used in this study were greater than 0.95.

Data C o l l e c t i o n At the baseline examination, study participants had three ECG electrodes placed on the epigastrium. Resting, supine, 2-rain beat-to-beat heart rate data were collected after participants remained comfortably in the supine position for 20 min during the B-mode ultrasound and arterial distensibility studi e s . 14 A dedicated computer and software program were used for continuous detecting and recording of ECG R waves. The system then converted the R-to-R interval into beat-to-beat heart rate, including a record of the clock time for each beat. TM

Blood Pressure and Hypertension At both visits, sitting blood pressure was measured three times on each participant with a random zero sphygmomanometer, after a 5-min rest, by trained technicians following a standardized protocol. 14 The systolic and fifth phase diastolic blood pressure measurements used in this report are the average of the second and the third readings. Study participants were asked to bring all medications, vitamins, and supplements taken in the 2 weeks prior to the examination. The information on pharmacological treatment of hypertension is based on the participant's self-reported use of any medication to treat high blood pressure, and the transcription and coding of all medication names. 14 Prevalent hypertension at the baseline examination was defined as diastolic blood pressure _~ 90 mm Hg, or systolic blood pressure 140 mm Hg, or use of antihypertensive medication. Incident hypertension after 3 years of follow-up was defined as diastolic blood pressure -> 95 mm Hg, or systolic blood pressure -> 160 mm Hg, or use of antihypertensive medication. Individuals with prevalent hypertension were defined as treated hypertensives if they reported the use of antihypertensive medication prior to the baseline examinations, and otherwise as untreated hypertensives.

I-IRV Analysis The 2-min raw heart rate data were first subjected to a filter program to remove artifacts under visual control by a single, trained operator. For each record, a plot of the beat-to-beat heart rate over time was displayed on screen. If segments of the beatto-beat record were interrupted by signals indicating artifacts, the operator used a mouse to draw an upper and a lower boundary within which to impute the heart rate for the sections of the record affected by artifacts. Details of the computer imputation algorithm will be published elsewhere and are available on request. A plot of the smoothed version of the heart rate data over time was then superimposed on the plot of the raw data, to confirm a good fit of any segment of smoothed data. The procedure could be repeated until a satisfactory plot was obtained. After smoothing, linear interpolation was applied to neighboring heart rate data points, and 256 heart rate data points were resampled with equal distance of 0.4685 sec. These 256 data points were used to fit a quadratic least squares model with time and time

Statistical Analysis To make inferences to the population from which this study sample was drawn, weighted analyses were performed taking into consideration the sampling weight in each stratum. Survey Data Analysis Software (SUDAAN, Research Triangle Institute, Research Triangle Park, NC) and SAS (SAS Institute, Cary, NC) were used. Mean values and stan-

1150 LIAOET AL

A/H-DECEMBER 1996 VOL 9. NO. 12, PART I

14 A

/V

N

-!- 12

2-Minute Heart Rate (HR)

C

m~

10 m "-"

,,r

m

8

v

re •- ( -

w ~

5oi

ca

6

20

i

' 40

i 60

i 80

100

i 120

Time (Second) ~e ~

4

r

if)

o~ 2 0

0

0.15

0.35

0.45

1.0

Frequency (Hz) FIGURE 1. An example of a power spectral density curve ~ollowntg fast fourier transformation, derived fpom time domain heart rate data (inserted panel).

dard errors / proportions and 95% confidence intervals for major covariates were obtained stratifying by hypertension status in both cross-sectional analysis and prospective analysis. Since HRV indices showed a distribution skewed to the right, natural logarithm transformation was used to normalize their distributions. Statistical comparisons were made on the natural logarithm scale. Weighted analysis of covariance was used to estimate adjusted geometric mean values of HRVHF, LF, L F / H F , and SDNN, and to test the mean differences comparing prevalent hypertensives with normotensives, and treated hypertensives with untreated hypertensive in the cross-sectional analysis. Weighted logistic regression was used to estimate the association of HRV indices and the risk of developing incident hypertension over 3 years of follow-up. RESULTS Cross-sectional Analysis The characteristics of the population from which the study sample for the cross-

sectional analysis was drawn are presented in Table 1. There were 650 prevalent hypertensives defined by BP >140/90 mm Hg or using antihypertensive agents, with a prevalence of 31.5%. Seventy-three percent (n = 477) of prevalent hypertensives were using antihypertensive medication (treated hypertensives). Among the treated hypertensives, 267 (56%) used diuretics, 157 (31%) used fl-blockers, 57 (13%) used antiadrenergics, 52 (10%) used ACE inhibitors, 33 (7%) used calcium channel blockers, and 75 (16%) used other types of antihypertensive medications. Sixty-nine percent of the treated hypertensives used only one type of antihypertensive medication, and 31% used at least two types of the above mentioned medications. The most common combination was fl-blockers and diuretics, which accounted for 11% of all treated hypertensives, or 36% of those who used at least two types of antihypertensive medications. Compared to normotensives, the hypertensives (both treated and untreated) were older, more likely to be African-American, less educated, more

AJH-DECEMBER 1996-VOL 9, NO 12, PART 1

AUTONOMIC FUNCTION A N D HYPERTENSION

1151

TABLE 1. SELECTED C H A R A C T E R I S T I C S , M E A N * (SE*) O R P R O P O R T I O N * (95% CI*), OF T H E B A S E L I N E STUDY POPULATION--THE A R I C B A S E L I N E S U R V E Y (1987 T O 1989) Untreated Hypertensives

Normotensives

Age (years) Race (% black) Gender (% men) Current smoker (%) Education (%) HS BMI (kg/nl 2) Diabetes mellitus (%) Heart rate (beats/rain) Systolic BP Diastolic BP Total cholesterol (mmol/L) HRV High frequency power (HF) (beats/min)2t Low frequency power (LF) (beats/min)e~Low/high frequency power ratio (LF/HF)~Standard deviation of RR interval (SDNN, msec)-t

Treated H y p e r t e n s i v e s

n = 1,411

N* = 7,575

n = 173

N* = 976

n = 477

N* = 2,649

53 18.9 45.7 23.7 18.7 40 1 41.2 26.4 4.6 67 114 70 5.48

(0.16) (18.5, 19.2) (43.0, 48.5) (21.4, 26.0) (16.6, 20.8) (37.4, 42.7) (38.5, 43.8) (0.13) (3.4, 5.7) (0.27) (0.34) (0.23) (0.03)

55 41.4 53.2 27.1 30.5 42.4 27.1 27.5 10.4 71 148 87 5.77

(0.45) (40.6, 42.1) (45.1, 61.2) (20.4, 33.7) (23.5, 37.5) (34.7, 50.0) (20.3, 34.0) (0.45) (5.4, 15.4) (0.92) (l .23) (0.83) (0.11)

56 43.7 37.5 22.1 33.6 38.6 27.8 29.6 17.0 68 131 78 5.58

(0.26) (43.1, 44.3) (32.9, 42.0) (18.1, 26.0) (29.0, 37.9) (34.1, 43.2) (23.6, 32.0) (0.27) (13.3, 20.6) (0.56) (0.98) (0.51) (0.06)

1.29

(0.04)

1.23

(0.12)

1.02

(0.06)

3.36

(0.09)

3.18

(0.29)

2.34

(0.13)

2.60

(0.06)

2.58

(0.17)

2.3(I

(0.10)

45

(0.71)

40

(1.93)

39

(1.13)

* Adjusted for samphng weight. t Expressed as geometric mean and standard error of the geometric mean BP, blood pressure, BML body mass index, HRV, heart rate variablhty; SDNN. standard devlat~on of all normal RR Intervals, LF, low frequency spectral component; HF, high frequency spectral component, HS, high school.

likely to be diabetics, and also had higher b o d y mass indices (BMI) and higher total s e r u m cholesterol. It is also s h o w n in Table 1 that hypertensives (both treated and untreated) had a lower m e a n value of HF and LF, and a similar m e a n value of LF / HF ratio, c o m p a r e d to normotensives. Age, race, gender, current smoking, diabetes, and education-adjusted geometric m e a n values of HRV-HF were 1.26 and 1.05 ( b e a t s / r a i n ) 2 for normotensive and hypertensive groups, respectively (P for m e a n difference < .01); those of LF were 3.24 and 2.75 ( b e a t s / min) 2 respectively (P for m e a n difference < .01 ); those of L F / H F were 2.57 and 2.65 (P for m e a n difference = .67), and those for S D N N were 39 and 35 (msec) (P for m e a n difference = .001). The prevalent hypertensives were divided into treated and untreated groups, and the adjusted geometric m e a n values of HRV indices by hypertension and treatment status, and the test of differences in m e a n values are presented in Table 2. The adjusted m e a n s of HRV-HF were 1.26, 1.20, and 1.00 ( b e a t s / m i n ) 2 for normotensive, untreated hypertensive, and treated hypertensive groups, respectively;

those of HRV-LF were 3.24, 3.26, and 2.58 for normotensive, untreated hypertensive, and treated hypertensive groups, respectively; those of L F / H F were 2.57, 2.72, and 2.58, respectively; and those of S D N N for these three groups were 39, 34, and 35 (msec), respectively. The test of the m e a n differences indicated that, comp a r e d to normotensives, treated hypertensives had significantly lower HF, LF, and S D N N (all P < .001 ). Untreated hypertensives had significantly lower S D N N than normotensives (P < .001). The test of the m e a n differences also indicated that c o m p a r e d to untreated hypertensives, treated hypertensives had slightly lower LF (P < .05). Analysis A total of 1,338 of 1,411 baseline normotensives were available for follow-up for at least 3 years, with a follow-up rate of 94.8%. The baseline characteristics of the population from which the study sample for the incident, prospective analysis was d r a w n are presented in Table 3. C o m p a r e d to normotensives, the incident hypertensives were more likely to be African-American, male, diabetic, and a current Prospective

1152

LIAO ET AL

AJH-DECEMBER 1996-VOL. 9, NO. 12, PART 1

TABLE 2. ADJUSTED*t GEOMETRIC M E A N VALUES OF HRV-HF, LF, LF/HF RATIO, A N D S D N N BY PREVALENT HYPERTENSION A N D TREATMENT S T A T U S - - T H E ARIC BASELINE SURVEY (1987 TO 1989) Heart Rate Variability Indices (HRV)

Normotensives

Untreated Hypertensives

Treated Hypertensives

n = 1,411 N t -- 7,575

n = 173 N t = 976

n = 477 N t = 2,649

High frequency power (HF) (beats/min) 2 Low frequency power (LF) (beats/rnin) 2 L o w / h i g h frequency power ratio (HF/LF) Standard deviation of RR intervals (SDNN)

1.26

1.20

1.00~:

3.24

3.26

2.58~:§

2.57

2.72

2.58

39

34:~

351:

* Adjusted for age, race, gender, current smoking, diabetes mellitus, and education level. t Adjusted for samphng wezght. :~Normotensive versus untreated or treated hypertenslw, s P < .01. § Untreated hypertensive versus treated hypertenswe" P < .05.

TABLE 3. BASELINE CHARACTERISTICS, MEAN* (SE*) OR PROPORTION* (95% CI*), A C C O R D I N G TO STATUS AT REEXAMINATION Incident Hypertensives

Normotensives

Age (years) Race (% black) Gender (% men) Current smoker (%) Education (~) < H S = HS >HS BMI ( k g / m 2) Diabetes mellitus (%) Heart rate (beats/rain) Systolic BP Diastolic BP Total cholesterol (retool/L) Heart rate variability High frequency power (HF) (beats/min)at Low frequency power (LF) (beats/min)2¢ L o w / h i g h frequency power ratio (LF/HF)t Standard deviation of RR intervals (SDNN, msec)t

Lost to Follow-Up

n = 1,274

N* = 6,829

n = 64

N* = 356

n = 73

N* = 390

53 16.8 44.9 22.3 17.6 40.5 41.6 26.2 3.8 66 113 69 5.51

(0.16) (16.4, 17.2) (42.1, 47.8) (19.9, 24.7) (15.6, 19.8) (37.8, 43.5) (38.8, 44.4) (0.13) (2.7, 4.9) (0.27) (0.35) (0.23) (0.03)

53 35.1 59.9 31.8 19.7 41.0 39.3 28.2 13.4 68 126 76 5.22

(0.79) (33.9, 36.4) (46.7, 73.1) (19.0, 44.5) (8.6, 30.9) (27.5, 54.5) (26.1, 52.6) (0.60) (3.7, 23.2) (1.53) (1.28) (1.32) (0.14)

54 39.9 46.8 40.6 36.5 29.7 33.7 27.9 10.1 69 117 73 5.45

(0.80) (38.3, 41.5) (34.3, 59.4) (28.6, 52.5) (24.7, 48.4) (18.5, 41.0) (23.5, 44.0) (0.73) (1.90, 18.3) (1.20) (1.43) (1.21) (0.14)

1.31

(0.04)

0.90

(0.14)

1 46

(0.22)

3.40

(0.10)

2.77

(0.39)

3.30

(0.45)

2.60

(0.06)

3.08

(0.37)

2.26

(0.28)

40

* Adjusted for samphng weight t Expressed as geometric mean and standard error. BP, blood pressure; BML body mass index: HS, high school.

(0.57)

34

(2.38)

38

(2.164)

AJH-DECEMBER 1996-VOL 9, NO. 12, PART 1

A U T O N O M I C F U N C T I O N A N D HYPERTENSION

smoker. They also had higher body mass index (BMI). They also exhibited higher baseline systolic and diastolic blood pressures, although these values were below the cut-off point for hypertension (by definition). Age and education level were similar between these two groups. A repeat examination could not be scheduled for 73 individuals (13 were deceased and 60 did not come to the scheduled follow-up visit). Their baseline characteristics are presented in Table 3. Overall, they were similar to those who had the follow-up visit. There were 64 individuals who developed hypertension, defined by BP > 160/95 mm Hg or using antihypertensive medication, over 3 years of follow-up, with a cumulative incidence of 5.0%. The 3-year weighted cumulative incidence by quartiles of baseline HRV indices is presented in Figure 2. It is apparent from this figure that baseline HRV-HF was inversely associated with the development of incident hypertension. A similar pattern of association can be seen for L F / H F ratio and SDNN, and there was no clear pattern of association between HRV-LF and the development of hypertension. The age, race, gender, current smoking, diabetes, and education-adjusted odds ratios (_+ 95% CI) of incident hypertension by quartiles of HRV indices are presented in Table 4, using the highest quartiles as the reference group. As shown in this table, a lower baseline HRVHF was associated with a graded increase in the risk of incident hypertension. The adjusted odds ratios (95% CI) were 1.00, 1.46 (0.61, 3.46), 1.50 (0.65, 3.50), and 2.44 (1.15, 5.20) from the highest quartile to the lowest quartile of HRV-HF (P for linear trend < .001). No clear, nor statistically significant pattern of association was observed for the HRV-LF and incident hyperten-

1153

sion, with adjusted odds ratios (95% CI) of 1.00, 1.41 (0.65, 3.07), 1.03 (0.45, 2.33), and 1.35 (0.64, 2.85) from the highest quartile to the lowest quartile of HRV-LF. Similar to HF, there was also a trend of significant association between LF / HF and the incident hypertension. The adjusted odds ratios (95% CI) were 1.00, 0.59 (0.28, 1.26), 0.69 (0.32, 1.46), and 0.47 (0.21, 1.05) from the highest quartile to the lowest quartile of L F / H F (P for linear trend = .05). A trend of inverse association between SDNN and incident hypertension was also found. The adjusted odds ratios (95% CI) were 1.00, 1.36 (0.60, 3.10), 1.36 (0.58, 3.21), and 1.74 (0.78, 3.85) from the highest quartile to the lowest quartile of SDNN (P for linear trend < .05). We also analyzed the prospective data by using a lower cut point criterion of BP > 140/90 (or using antihypertensive agents) to define incident hypertension. From this, 126 individuals were identified as having developed incident hypertension over 3 years of follow-up, with a cumulative incidence of 9.6%. The relation of HRV indices and incident hypertension was similar to that found by using the higher cut point definition described above (data not shown). DISCUSSION Analysis of beat-to-beat HRV data has been widely used as a method to evaluate sympathetic and parasympathetic influence on the cardiac system in various scientific applications. 2-~° Traditionally, HRV analyses have been based on heart rate data recorded over several hours. This provides a wealth of data (and high degrees of precision), but also introduces costs and logistical complexities that are particularly burdensome for population based, epidemiologic re-

A

~

10.001

•

O ,n

= 1::

i 8.00 -

O~

O.

Q1 [ ] Q 2

[]Q3

[] Q4

7 36

.I-

645

"6 8 . o o -

FIGURE 2. Cumulative incidence of hypertension (%) by quartiles of heart rate variabihty ( HRV ) radices, 3 years of follow-up--The ARIC Study. HF, HRV high frequency spectral po~;er component: LF, HRV low frequency spectral power component; SDNN, standard deviation of all normal RR intervals.

6.34

5 86 5 38

=

4.00 -

Q)

"10 r. m

>

2.00

m E : O

0.00 HF

LF

LF/HF Ratio

SDNN

1154

LIAO ET AL

AJH-DECEMBER 1996-VOL. 9, NO. 12, PART 1

TABLE 4. ADJUSTED*t O D D S RATIO (95% CI) OF INCIDENT HYPERTENSION DURING 3 YEARS OF FOLLOW-UP, BY QUARTILES OF HRV INDICES AT BASELINE. THE UPPERMOST QUARTILE IS USED AS REFERENCE G R O U P - - T H E ARIC STUDY OR (95% CI) Quartile of HRV Indices HRV Indices

High frequency power (HF, beats/min) 2 Low frequency power (LF, beats/min) 2 Low/high frequency power ratio (LF/HF) Standard deviation of RR intervals (SDNN)

Q4

Q3

Q2

Q1

P for Linear Trend

1.00

1.46 (0.61, 3.46)

1.50 (0.65, 3.50)

2.44 (1.15, 5.20)

<.001

1.00

1.41 (0.65, 3.07)

1.03 (0.45, 2.33)

1.35 (0.64, 2.85)

>.10

1.00

0.59 (0.28, 1.26)

0.69 (0.32, 1.46)

0.47 (0.21, 1.05)

.05

1.00

1.36 (0.60, 3.10)

1.36 (0.58, 3.21)

1.74 (0.78, 3.85)

<.05

* Weighted logistic regresswn models, adjusted ~or samphng weight. 4; Adjusted for age, race. gender, current smoking, education, and dzabetes melhtus.

search. In this report, we applied spectral analysis to 2-min heart rate data collected according to a standardized protocol, by trained and certified technicians, subject to well established quality control monitoring. Although caution needs to be exercised when estimating low frequency component (LF) of HRV from heart rate records of short duration, sufficient documentation exists today to support the use of records of 5-min duration, or less. For frequency domain analysis, Bigger 3 indicates that estimates of high frequency power centered around 0.25 Hz should be based on records of approximately 1-min duration; LF power estimates (0.04 to 0.15 Hz) require about 2.5 min of beat-to-beat data. Additional, supportive findings have been reported by other investigators in this field. ~'~ Based on the work of Bigger, 3 our 2-min records collected on the ARIC cohort during its baseline examination are informative, since they included at least 10 to 15 times the period of the heart rate fluctuation being estimated. A single, trained technician processed all HRV data for this study. Standardized data collection and processing protocols were used in this study, as reflected in reasonably high short-term intraparticipant reliability coefficients, intra-, and interreader reliability coefficients for all HRV indices used in this study. Our cross-sectional analysis of the baseline data indicate that hypertensives have significantly lower HF, LF, and SDNN than normotensives (all P < .001); treated hypertensives have significantly lower LF than untreated hypertensives (P < .05); and no statistically significant differences were observed for the L F / H F ratio. These findings may be indicative of a reduced parasympathetic function in hypertensives. However, the differences between hypertensives and normotensives in the cross-sectional data is largely due to the lower HF and lower LF among treated

hypertensives. Caution should be exercised in the interpretation of the cross-sectional differences of HRV between hypertensives and normotensives, and between treated and untreated hypertensives, since these results could be reflective of hypertension per se, the effect of antihypertensive medication, or a mix of these effects. There may also be differences by severity and for duration of hypertension, which we are not able to ascertain from these data. The profile of antihypertensive medication used in our study population is similar to that found in the United States during the examination period (1987 to 1989).1, However, small numbers for each type of antihypertensive medication, and the lack of information on duration of use, prevent us from comparing HRV indices by type of antihypertensive medication at baseline. Seated blood pressure at baseline is a strong predictor of the development of hypertension in this cohort, and it is also associated (weakly) with HRV. Seated blood pressure could therefore be considered as a covariate, to be included in the multivariable analytic models predicting incident hypertension, for purposes of statistical adjustment. However, the authors adhere to the view that seated blood pressure measured at the baseline examination reflects the autonomic regulation of blood pressure, which in turn contributes to its sustained elevation over time. Therefore, statistical adjustment of blood pressure measured at baseline at intake could be reasonably regarded as "over adjustment" instead of exercising control for a "nuisance variable." Consequently, we did not think it appropriate to adjust our findings by including baseline blood pressure values in the multivariable estimation of the risk of incident hypertension. We wish to point out to the interested reader, however, that the addition of baseline levels of blood pressure to the analyses reduced the magnitude of the

AJH-DECEMBER 1996-VOL. 9, NO. 12, PART I

associations between HRV and the risk of developing hypertension, but did not change the direction or trend of the findings. A u t o n o m i c function plays an important role in the d e v e l o p m e n t and maintenance of the elevated blood pressure, t h r o u g h (at least) three possible mechanisms1: direct effect on the heart to increase stroke volume, output, and heart rate; direct effect on peripheral arterial resistance; and possibly by directly and indirectly increasing renin level and renin activity to increase blood pressure. Although these mechanisms are plausible, the hypothesis that an altered autonomic cardiac function is associated with the risk of developing h y p e r t e n s i o n has not been tested at the general population level. Our results showing a statistically significant, graded, inverse association between HRV HF and the cumulative incidence of hypertension p r o v i d e support for a link between red u c e d vagal function and the risk of developing hypertension. Similar associations are present in our data between L F / H F ratio, SDNN, and incident hypertension (P for trend < .01 and < .05, respectively). By contrast, no clear pattern of association was observed for HRV-LF and incident h y p e r t e n s i o n in our data, a finding that differs from that reported in the cross-sectional studies. 7'11 This difference could be due to several reasons. First, our study protocol required participants to rest in a quiet environment, in the supine position for at least 20 min prior to beatto-beat heart rate data collection. Under these conditions, the vagal function predominates, and sympathetic function is at its minimum. This m a y be relevant in our particular case, since the LF is influenced by both the sympathetic and parasympathetic function, and our LF m e a s u r e m e n t m a y reflect strong vagal influence. Further, LF p o w e r estimation requires about 2.5 rain of beat-to-beat data, 3 and the 2-min beat-tobeat heart rate data collected in our study are only long enough to capture a few cycles of low frequency heart rate fluctuation. Therefore, the LF estimated in our s t u d y m a y not be an accurate reflection of the sympathetic function expressed in the heart rate variability. Finally, it should also be considered that the two studies cited above e m p l o y e d autoregressive methods, opening the possibility that methodological differences m a y underlie the discrepancy with the findings reported here. The same considerations can be applied to the cross-sectional association of hypertension treatment status and LF discussed earlier. In the context of heart rate variability records of short duration, it was deemed important to supplement the information in spectral components by computing the standard deviation of RR intervals (SDNN), in assessing autonomic cardiac modulation. SDNN, a conventional time domain index of overall heart rate variability, was statistically

AUTONOMIC FUNCTION AND HYPERTENSION

1155

lower in both treated and untreated hypertensives than normotensives at the baseline examination. SDNN was also associated with the incidence of hypertension over the course of 3 years (P for linear trend < .05). This s t u d y is based on a short follow-up (3 years) and the n u m b e r of individuals w h o d e v e l o p e d incident h y p e r t e n s i o n is small (n = 64). Consequently, its generalizability to long-term follow-up is limited. Nevertheless, this s t u d y represents the first population-based prospective s t u d y to identify HRV as a potentially i m p o r t a n t risk factor of hypertension, s u p p o r t i n g the role of a u t o n o m i c function in the dev e l o p m e n t of h u m a n hypertension.l Additional, and long-term, population-based follow-up studies are n e e d e d to confirm our findings, before it can be concluded that r e d u c e d HRV plays a causal role in the d e v e l o p m e n t of hypertension. ACKNOWLEDGMENTS The authors wish to acknowledge the valuable contributions made by the ARIC staff at the five collaborating institutions: The University of North Carolina at Chapel Hill, Chapel Hill: Catherine C. Paton and Thomas G. Goodwin; The University of Mississippi Medical Center, Jackson: Patricia F. Martin; The University of Minnesota, Minneapolis: Gail Murton; The Johns Hopkins University, Baltimore: Sunny Harrell; Bowman-Gray School of Medicine, Winston-Salem: Robert Ellison, Fontaine Gervassi, and Teresa Crotts. REFERENCES

1. Julius S, Esler MD, Randall OS: Role of autonomic nervous system in mild human hypertension. Am Heart J 1975;48:243s-252s. 2. Pfeifer MA, Cook D, Brodsky J, et al: Quantitative evaluation of cardiac parasympathetic activity in normal and diabetic man. Diabetes 1982;31:339-345. 3. Bigger JT, Jr.: Spectral analysis of R-R variability to evaluate autonomic physiology and pharmacology and to predict cardiovascular outcomes in humans, in Zipes D, Jalife J (eds): Cardiac Electrophysiology: From the Cell to the Bedside (2nd ed). WB Saunders Co, Philadelphia, 1995, pp 1151-1170. 4. Akselrod S, Gordon D, Ubel FA, et al: Power spectrum analysis of heart rate fluctuation: a quantitative probe of beat-to-beat cardiovascular control. Science 1981; 213:220-222. 5. Pomeranz B, Macaulay RJB, Caudlll MA, et al: Assessment of autonomic function in humans by heart rate spectral analysis. Am J Physiol 1985;248:H151-H153. 6. Hayano J, Sakakibara Y, Yamada A, et al: Accuracy of assessment of cardiac vagal tone by heart rate variability in normal subjects. Am J Cardiol 1991;67:199-204. 7. Pagani M, Lombardi F, Guzzetti S, et al: Power spectral analysis of heart rate and arterial pressure variabilities as a marker of sympatho-vagal interaction in man and conscious dog. Circ Res 1986;59:178-193. 8. Kamath MV, Ghista DN, Fallen EL, et al: Heart rate variability power spectrogram as a potential noninva-

1156

LIAO ET AL

sive signature of cardiac regulatory system response, mechanisms, and disorders. Heart Vessels 1987;3:3341. 9.

Malik M, Camm AJ: Heart rate variability. Clin Cardiol 1990; 13:570-576. 10. Ori Z, Monir G, Weiss J, et al: Heart rate variabilityfrequency domain analysis. Cardiol Clin 1992; 10:499537. 11. Guzzetti S, Piccaluga E, Casati R, et al: Sympathetic predominance in essential hypertension: a study employing spectral analysis of heart rate variability. J Hyperten 1988;6:711-717. 12. ARIC Investigators: The Atherosclerosis Risk in the Communities (ARIC) study: design and objectives. Am J Epidemiol 1989; 129:687-702. 13. Heiss G, Sharrett AR, Barnes R, et al: Carohd atherosclerosis measured by B-mode ultrasound in populations: associations with cardiovascular risk factors in the ARIC study. Am J Epidemiol 1991;134:250-256. 14. National Heart, Lung and Blood Institute: The ARIC

AJH-DECEMBER 1996-VOL 9, NO. 12, PART 1

Manuals of Operation: no. 1, General description and study management; no. 2, Cohort Component; no. 3, Surveillance component procedures; no. 4, Pulmonary function assessment; no. 5, Electrocardiograph; no. 6, Ultrasound assessment; no. 7, Blood collection and processing; no. 8, Lipid and lipoprotein determinations; no. 9, Hemostasis determinations; no. 10, Clinical chemistry determinations; no. 11, Sitting blood pressure and postural changes in blood pressure and heart rate; no. 12, Quality assurance and quality control for ARIC cohort study. ARIC Coordinating Center, School of Public Health, University of North Carolina, Suite 203. NCNB Plaza, 137 E. Franklin St., Chapel Hill, NC 27514. 15. Freed LA, Stein KM, Gordon M, et al: Reproducibility of power spectral measures of heart rate variability obtained from short-term sampling periods. Am J Cardiol 1994; 74:972-973. 16. Manolio TA, Cutler JA, Furberg CD, et al: Trends in pharmacologic management of hypertension in the United States. Arch Internal Med 1995; 155:829-837.