Clin. Cardiol. 13, 570-576 (1990)
Electrophysiology, Pacing, and Arrhythmia This section edited by A. John Camm, M.D . , F. R. C.P . , F. A . C,C.
Heart Rate Variability M. MALIK.M.D., Ph.D.. A . J . CAMM. M.D..F.R.C.P., F.A.C.C. Department of Cardiological Sciences, St. George's Hospital Medical School, London, England
Summary: Reduced heart rate variability cames an adverse prognosis in patients who have survived an acute myocardial infarction. This article reviews the physiology, technical problems of assessment, and clinical relevance of heart rate variability. The sympathovagal influence and the clinical assessment of heart rate variability are discussed. Methods measuring heart rate variability are classified into four groups, and the advantages and disadvantages of each group are described. Concentration is on risk stratification of postmyocardial infarction patients. The evidence suggests that hean rate variability is the single most important predictor of those patients who are at high risk of sudden death or serious ventricular arrhythmias.
diseases of the heart. Clinical cardiologists were therefore surprised that an absolutely regular sinus rhythm is also an indication of malfunction of cardiac regulatory processes and that reduced variability of sinus rhythm is a serious warning sign in patients who survive the acute stage of myocardial infarction. The first observation that reduced hean rate variability (HRV) correlates with mortality and severe arrhythmic complications during the postinfarction period was published in 1978.' Since that time, several clinical studies have been completed which fully confirm this finding (Fig. 1). This brief review summarizes the information that has been accumulated on the irregularity and variability of sinus rhythm and on its clinical usefulness, especially for the stratification of postinfarction risk.
Key words: sympathovagal cardiac control, myocardial infarction, long-term electrocardiograms
Introduction Although recognized and widely known, the variations of cardiac rhythm have been virtually ignored in practical cardiology. Cardiologists have generally believed that irregularity of cardiac function is a pathological phenomenon. This was probably the result of the large clinical experience with atrial fibrillation and ventricular ectopic beats which indicate impaired intracardiac regulatory mechanisms and are negative prognostic factors in many
Supported by the British Heart Foundation Address for reprints: Prof. Marek Malik. Ph.D., M.D. Department of Cardiological Sciences St. George's Hospital Medical School Cranmer Terrace London SW I7 ORE, England Received: June 1 1 , 1990 Accepted: June 15, 1990
Physiology of Heart Rate Variability The physiological background of HRV has been attributed to the sympathovagal system. Because of the dramatic changes in the sympathovagal system in cold-blooded vertebrates, several laboratory studies have examined changes in hean rate and HRV provoked by different stimuli in these animals, especially in lizards (e.g., Ref. 2). In mammals, HRV has also been examined in situations which are known to be associated with marked changes in the tone of the autonomic nervous ~ y s t e m ,including ~,~ investigations into specific diseases which impair the autonomic system, such as d i a b e t e ~ HRV .~ has also been used to demonstrate the development of tonic vagal influence on the heart.6 HRV has been studied in humans in order to assess sympathovagal balance.' HRV provoked by atropine administration has been reported as a method of the diagnosis of brain death. * Many neurological and psychological investigations have used HRV to evaluate the effects of stress, emotion, and work on the autonomic nervous s y ~ t e r n , ~ - l ~ including very specialized investigations. l 4 Quantification of HRV has also been used as the standard for evaluating other psychological methods. I 5 Physiological studies in humans have reported an increase of HRV, interpreted as an increase of cardiac va-
M. Malik and A. J. Camm: Heart rate variability
From a purely technical point of view, the methods used for HRV assessment can be approximately divided into four groups. Simple Methods
i-I III 21 I/
6 1 4
FIG.I Typical differences in HRV between high- and low-risk patients who have survived acute myocardial infarction. The plots show sample density distributions of RR intervals from a continuous 24-h Holter recording made on Day 7 after myocardial infarction in two patients (labeled D and S in the figure). Both patients were male of similar age (D=73and S=71 years), suffering from anterior infarction with pathological Q waves on the electrogram. Their LVEF was similar (D=35% and S=30%);neither was treated with beta blockers. Patient S had an uncomplicated postinfarction course during a 2-year follow-up, Patient D died suddenly on Day 27 after the infarction. Note that the duration of RR intervals (i.e., the rate of the sinus rhythm) was virtually constant in Patient D while it was very variable in Patient S. The horizontal axes RR show duration of normal-to-normal intervals in seconds, the veltical axes NN show the number of normal-to-normal intervals in thousands.
gal tone at rest, following intensive exercise for several days. l 6 A similar study performed in dogs” showed that the postexercise increase of vagal tone was lower in animals who were at higher risk of ventricular fibrillation induced by artificial coronary occlusion performed the following day. These and other similar studies create a continuous spectrum between purely physiological and mainly clinical investigations. There appears to be a “prima facie” case that HRV, sympathovagal tone, and cardiac arrhythmias are linked and that a mechanical relationship is probable.
Measurement of Heart Rate Variability Both research and clinical studies have reported a variety of recording techniques and different methods for actual measurement and numerical quantification of HRV. For example, special Computerized systems were developed for the assessment of fetal heart rate (FHR) variability,I8 including automatic recognition and differentiation of fetal and maternal cardiac depolarizations.l9 Unfortunately, many of these methods were designed for very specific purposes and cannot be used or easily adapted for the measurement of HRV in other situations, such as those for stratification of patients who are at high risk after acute myocardial infarction.
In studies examining very short-term HRV, such as the sudden changes in heart rate provoked by instantaneous stimuli, a very simple measurement of HRV can be used. Instantaneous change in heart rate can be expressed as the ratio (min RR)/(max RR)20(e.g.. the Valsalva ratio), or as the difference (max RR)-(min RR).21 Similar simple methods have been used for the assessment of FHR variability. For instance, simple counts of marked accelerations of FHR occurring during 30 min intervals have been used to express FHR variability.22 These methods are unsuitable for the assessment of HRV in studies oriented to postinfarction risk stratification. Most of these studies are based on elaborating the signals obtained by computerized recognition of long-term electrocardiograms. The first observation that postmyocardial infarction risk was associated with reduced HRV was based on short-term and manually measured electrocardiograms.I However, more recent studies have shown that the most significant and powerful distinction between patients at low and high risk following acute myocardial infarction must be based on HRV assessed from Holter records. In the majority of these studies, 24-h electrocardiograms were used, and both commercial and research algorithms were employed to analyze the tapes and identify the sinus rhythm beats. Spectral Analysis and Autocorrelation
Spectral analysis of electrocardiograms is an important field of both theoretical and practical biomedical engineering. Often, spectral analysis is performed on signalaveraged electrocardiograms where it can reveal late potentials and serve as one of the methods for the assessment of the homogeneity and synchrony of cardiac excitation. Analogous methods can also be used to analyze the sequence of RR interval^.^^-^^ Broadly speaking, spectral analysis evaluates and quantifies periodicities which can be found in the analyzed data. For each length of time, the results of spectral analysis express the range of changes in the variable data corresponding to the given period. In this way, spectral analysis allows superimposed periodicities to be unravelled. A similar spectrum of periodical components is provided by the so-called autocorrelation and autoregression methodsz6which are based on comparisons of individual and variable segments of the analyzed data. The major advantage of spectral methods lies in their accuracy. These methods can be used to evaluate physiological variations of sympathovagal tone in individualsand are also able to distinguish cases in which only one of the mechanisms of a known periodicity is impaired. For instance, a selective depression of vagal tone with preser-
Clin. Cardiol. Vol. 13, August 1990
vation of sympathetic cardiac control has been diagnosed using the spectral methods. 27 However, the accuracy of spectral methods also accounts for their disadvantages. The accuracy of results makes the methods largely dependent on the accuracy and quality of the analyzed data. Unfortunately, very accurate recording of long-term electrocardiograms is difficult to guarantee. The exact shape of electrocardiographic signals including the QRS patterns vary during ambulatory recording and there is much artefact in the record. The results of spectral methods can also be affected by ‘behavioral’ noise. If the state of the patient varies during recording, the dominant results of spectral methods correspond to the changes in patient behavior rather than to the changes in sympathovagal tone. Therefore, a precise validation of spectral results requires that recordings are made during a ‘steady state.’ This is impossible to achieve in long-term electrocardiograms. Statistical Methods The evaluation of spectral data in large populations of patients leads to a comparison of selected parameters such as the spectral components associated with one particular form of periodicity of the cardiac cycles. This may not necessarily be the optimum approach to the stratification of postinfarction risk, as the general overall HRV which comprises all spectral components may also be relevant. The usual approach to the evaluation of the overall HRV is the application of a standard statistical method to express the variability within the sequence of durations of normalto-normal RR (NN) intervals. Different statistical methods have been employed, including the standard deviation of NN interval durations2*and the standard deviation of the differences between neighboring NN intervals (expressing the so-called beat-to-beat HRV).29 These general approaches have also been modified to include only selected portions of the spectrum of heart rate changes. These modifications enable both short- and long-term HRV to be separately quantified and evaluated, but their definition and distinction is much less accurate when compared with the standard spectral methods. However, these combined components of HRV may be medically important. Various adapted methods have therefore been proposed, mainly for the analysis of FHR variability. Several such methods were validated on both clinical data and computer-generated series of known predefined variability.I0 Large population studies by the North American Multicenter Post-Infarction utilized these statistical methods, but also showed their dependence on the validity of the data provided by algorithmic computer analysis of Holter tapes. Statistical formulae expressing standard deviation and similar parameters have a certain preference in the values which are far from the mean of the analyzed sequence. In practical terms this means that the results of standard statistical formulae are seriously affected by such misrecognition artefacts as the omission
of QRS complexes because of their varying voltages or the misinterpretation which classifies premature ventricular depolarizations or even T waves as supraventricular patterns. The authors of the multicenter studies employed filtering of the computer recognized N N intervals in order to reduce the level of the artefact but also reported the use of visual checking and manual corrections of the analyzed data. Technically oriented studies show that even sophisticated filtering of NN sequences does not necessarily lead to the desired results.’l This acknowledges the dependence of statistical methods on the quality of the computer recognition of N N intervals. Similar dependence is seen with methods analyzing the beat-to-beat HRV which are based on counting the sudden changes in N N interval over a given threshold.I2 Evaluation of RR Interval Distributions From a practical clinical point of view, methods that provide an approximation of HRV can be clinically useful. Several such methods are based on the evaluations of the statistical distribution of NN interval durations. Even methods based on simple analysis of NN interval di~tributions~l can offer medically relevant evaluation of HRV (Fig. 2).I3.l4 The simplest of these methods expresses HRV as the relative number of N N intervals with the most frequent duration (i,e., as the fraction totlmar, where m a r is the maximum number of computer-recognized N N intervals with the same duration; and tor is the number of all NN intervals recognized in a Holter record). When this simple method was applied to analysis of standard quality 24-h recordings HRV data were provided which proved to be of very significant value (Fig. 3). These methods can be adapted to approximate spectral analysis, the results of which are not affected by a low level of noise and artefact in the analyzed data.Is At the same time, the proof of the validity of these methods requires further investigation, including theoretical mathematical examinations. For artefact-free data, the standard statistical methods and the methods that combine several power spectrdl components provide strongly correlating results. l6 Future studies should investigate similar intramethod correlations for data which include a low level of artefact.
Clinical Studies Although postinfarction risk stratification has been the first cardiological need for HRV assessment, it was not the first attempt to quantify both the beat-to-beat and circadian variations in cardiac cycles. Studies in Noncardiological Medical Fields HRV assessment has been widely used in clinical investigations, for example in the evaluation of alcoholic
M. Malik and A. J. Camm: Heart rate variability
HRV (ms) 400
10 12 14 16 18 20 22
Time of day
Time of day FIG.2 Differences between short-term HRV in two groups of 40 postinfarction patients matched with regard to age, gender, infarct site, ejection fraction, and beta-blocker treatment. Patients in the positive group (A) experienced serious arrhythmic events (sudden death or sustained ventricular tachycardia) during a 6-month follow-up; patients in the negative group (B) remained free of complications for more than 6 months after discharge. Short-term HRV (baseline width of the distribution curve of the unfiltered sequence of NN intervals) was measured in separate 40 min intervals. For both groups, the mean values f SD are shown.
ne~ropathy,~' familial amyloid polyneuropathy,38diabetic and hyperten~ion.~' neuropathy,32.39 thyrotoxi~osis,~~ Assessment of FHR variability is an acknowledged method for stratification of pre and intralabor risk. Reduced variability of FHR has been observed prior to fetal death.42 On the contrary, a physiological level of FHR variability indicates an uncomplicated labor and normal adaptation of neonates.43 In infants, short-term HRV has also been used to examine the reaction to external stimuli, such as in tests of hearing.44Similar studies have been performed examining FHR variability following external vibratory stimulus.45
n = 366
I I c
n = 27 200
FIG.3 Risk stratification in a group of 414 consecutive myocardial infarction patients who survived until discharge. HRV (baseline width of the N N intervals distribution curve) was assessed in 24-h Holter records made at a median of 7 days (range 5-9 days) following the hospital admission. The plot shows the differences (mean f SE) of three groups of patients: (A) uncomplicated 12-month follow-up; (B) serious arrhythmic events (sudden death or sustained ventricular tachycardia); ( C ) patients who died due to other reasons (reinfarction or cardiac failure).
Studies correlating low HRV with the risk of sudden death are not only related to postinfarction adult patients. Reduced HRV has also been associated with the sudden infant death ~ y n d r o m e . ~ ~ - ~ ~ Stratification of Postinfarction Risk Utilization of reduced HRV as a major factor for the stratification of postinfarction risk is attractive because the long-term electrocardiographic recording which enables the assessment of HRV is one of the least invasive methods of investigation. Detailed case-matched studies as well as investigations involving large populations of postinfarction patients have repeatedly proved that a reduced HRV is an important negative prognostic factor correlating with both the so-called all cause mortality and with anhythmic complications and sudden death.28.29.3'.34.50-52 The possibility of repeated measurement of HRV in patients after myccardial infarction has also been examined53in order to assess the state of the heart during postinfarction healing and recovery. The findings of these studies suggest that the reduced vagal and increased sympathetic activity accounts for low HRV. High sympathetic and low vagal tone decreases the electrical threshold for ventricular fibrillati0n2~and the probability of ventricular fibrillation during myocardial ischemia is higher in patients with reduced HRV. In a study of 177 consecutive patients34the relative risk of low HRV for the prediction of sudden death and /entricular tachycardia was 7.0, compared with lower relative risks of in-hospital complications (4.3),Killip I1 classification at admission (4.0),the presence of late potentials ( 3 . 3 , and left ventricular ejection fraction