Nonlinear heart rate variability and risk stratification in cardiovascular disease.  
Jump to Full Text  
MedLine Citation:

PMID: 16943869 Owner: NLM Status: PubMednotMEDLINE 
Abstract/OtherAbstract:

Traditional time and frequency domain heart rate variability (HRV) have cardiac patients at risk of mortality postmyocardial infarction. More recently, non linear HRV has been applied to risk stratification of cardiac patients. In this review we describe studies of non linear HRV and outcome in cardiac patients. We have included studies that used the three most common nonlinear indices: power law slope, the short term fractal scaling exponent and measures based on Poincare plots. We suggest that a combination of traditional and nonlinear HRV may be optimal for risk stratification. Considerations in using non linear HRV in a clinical setting are described. 
Authors:

Phyllis K Stein; Anand Reddy 
Related Documents
:

8391399  Cardiac sympathetic nervous activity in congestive heart failure. evidence for increase... 12213639  Baroreceptorrelated dysrhythmias in piglets with selective autonomic denervation of th... 9262329  Sympathetic neurotransmission in isolated rat atria after sensorymotor denervation by ... 16330909  Continuous heart rate variability from an implanted device: a practical guide for clini... 1223139  Involvement of adenylate cyclase in mechanisms of denervation supersensitivity followin... 20067509  Autonomic alterations and cardiac changes in epilepsy. 11242559  Cardiopulmonary complications following cardiac surgery. 20059709  Cardiac anatomy for the interventional arrhythmologist: i.terminology and fluoroscopic ... 12796129  Combined epicardial and endocardial electroanatomic mapping in a porcine model of heale... 
Publication Detail:

Type: Journal Article Date: 20050701 
Journal Detail:

Title: Indian pacing and electrophysiology journal Volume: 5 ISSN: 09726292 ISO Abbreviation: Indian Pacing Electrophysiol J Publication Date: 2005 
Date Detail:

Created Date: 20060831 Completed Date: 20070718 Revised Date: 20091118 
Medline Journal Info:

Nlm Unique ID: 101157207 Medline TA: Indian Pacing Electrophysiol J Country: India 
Other Details:

Languages: eng Pagination: 21020 Citation Subset:  
Affiliation:

Washington University School of Medicine HRV Lab, St. Louis, MO 63108, USA. pstein@im.wustl.edu 
Export Citation:

APA/MLA Format Download EndNote Download BibTex 
MeSH Terms  
Descriptor/Qualifier:


Comments/Corrections 
Full Text  
Journal Information Journal ID (nlmta): Indian Pacing Electrophysiol J Journal ID (publisherid): Indian Pacing Electrophysiol J ISSN: 09726292 Publisher: Indian Pacing and Electrophysiology Group 
Article Information Download PDF Copyright: ? 2005 Stein et al. openaccess: This is an openaccess article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. collection publication date: Season: Jul?Sep Year: 2005 Electronic publication date: Day: 01 Month: 7 Year: 2005 Volume: 5 Issue: 3 First Page: 210 Last Page: 220 ID: 1431594 Publisher Id: ipej05021000 PubMed Id: 16943869 
NonLinear Heart Rate Variability and Risk Stratification in Cardiovascular Disease  
Phyllis K Stein, PhD  
Anand Reddy, MD  
Washington University School of Medicine, St. Louis, MO, USA 

Correspondence: Address for correspondence: Phyllis K. Stein, Ph.D., Washington University School of Medicine HRV Lab, 4625 Lindell Blvd, Suite 402, St. Louis, MO 63108, USA Email: pstein@im.wustl.edu 
The association between traditional time and frequency domain measures of heart rate variability (HRV) and cardiovascular outcomes has been the subject of numerous investigations. Although there were earlier reports suggesting the potential clinical importance of HRV, the wellknown paper published by Kleiger and colleagues reporting the results of the MultiCenter PostInfarction Project (MPIP) in 1987 can be viewed as the beginning of serious interest in HRV among cardiologists [^{1}]. This paper reported the remarkable independent association of a single, time domain HRV measure, SDNN, the standard deviation of all normaltonormal (NN) interbeat intervals over 24hours and risk of mortality 1year postMI. With the publication of this paper, the race was on and investigators tried to determine whether other time domain HRV variables were, perhaps better predictors of outcome. By 1991, frequency domain measures, based on the fast Fourier transform or similar mathematical analyses, had been applied to 24hour Holter recordings, with seemingly even better results [^{2}]. Large datasets of Holter recordings began to be collected, especially by the group at St. George?s in London who developed their own geometric measures of HRV that could be applied to Holter recordings without the requirement of painstaking and exacting characterization of all interbeat intervals [^{3}]. By the mid90?s, there were a number of studies verifying the predictive value of HRV in various cardiac patient populations, and the Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology had issued a set of formal recommendations for the measurement and clinical uses of HRV [^{4}].
At about that time, investigators began to consider whether measures of HRV derived from nonlinear mathematics might have some application to clinical populations. It is beyond the scope of this review to detail the various methods that were applied. Rather, we will concentrate in detail on the three bestknown: the power law slope, measures derived from the Poincar? plot and the shortterm fractalscaling exponent. In the sections below we will discuss some of the significant investigations in the field. Table 1 provides a more complete listing of studies in the field.
When HRV is analyzed in the frequency domain, the variance of the signal is decomposed into its sinusoidal components at various underlying frequencies. This can be visualized in a plot of the HRV power spectrum (see Figure 1) in which the area under the curve at each frequency (on the xaxis in Hz) is equal to the variance explained at that frequency. In 1982 Kobayashi and Musha reported on the frequency dependence of the HRV power spectrum a normal young man [^{5}]. That is, they noted that when the power spectrum is plotted on a log:log scale, as in Figure 1, at least below 102 Hz, the amount of power increased, linearly, as the underlying frequency decreased (called a 1/f relationship). The slope of this regression line is now referred to as the power law slope, and its intercept with the yaxis is the power law intercept. In nonlinear terminology, this meant that fractal scaling properties of longerterm HRV were selfsimilar on a scale of minutes to hours.
The potential prognostic value of the power law slope was reported in 1996 when Bigger and his coworkers reanalyzed data from MPIP [^{6}]. In addition, they compared power law slope and intercept between N=274 MPIP patients (patients with a recent myocardial infarction), 274 healthy persons and 19 people with heart transplants. They concluded that MI caused a relative denervation of the heart (steeper slope, decreased intercept) that was significantly more adverse among heart transplant patients. This is illustrated in Figure 2 which compares power law slope for a healthy subjects and one with severe cardiac disease. Individually and especially combined, power law regression parameters were stronger predictors of death from any cause or arrhythmic death than traditional measures of time or frequency domain HRV.
In a subsequent prospective study by Huikuri and his coworkers, N=347 people >65 years of age were followed for 10 years [^{7}]. Power law slope (<1.50) was the best univariate predictor of all cause mortality (odds ratio, 7.9; 95% CI 3.7 to 17.0; p<0.0001). Decreased power law slope predicted both cardiac and cerebrovascular death, suggesting that altered long term behaviour of HR reflects an increased risk of vascular causes of death, rather than being a marker of any disease or frailty leading to death.
Similarly, HRV indices, including power law slope was compared in 2 groups of patients after acute myocardial infarction, one with normal and one with reduced ejection fraction. A more negative power law slope was observed in patients with reduced ejection fraction after myocardial infarction but not in those with a normal ejection fraction [^{8}]. This study was small (N=35) and no relationship with outcome was reported.
In 1995, CK Peng and colleagues proposed a novel nonlinear measure of HRV called the shortterm fractal scaling exponent [^{9}]. Unlike many of the other proposed nonlinear measures, this new measure was relatively straightforward to compute and not hampered by any mathematical assumptions. The mathematical algorithm for calculating the shortterm fractal scaling exponent (referred to as DFA1, alpha1 or ?1) can be downloaded from PhysioNet www.PhysioNet.org. DFA1 has been calculated from all interbeat intervals, regardless of their aetiology, and it has been calculated from normaltonormal interbeat intervals only. DFA1, which can be calculated from a minimum of 1000 beats and then averaged, reflects the amount of randomness in the heart rate time series, with the lowest values (~0.5) associated with a completely random time series and the highest values (1.5) associated with a time series that is totally correlated. It is becoming clear that increased randomness of heart rate patterns, i.e., a high degree of sinus arrhythmia of nonrespiratory origin, is associated with a worse prognosis [^{10}].
It should be noted that the algorithm for the calculation of DFA1 has recently been improved, and although values calculated using the old and new algorithms correlate >0.95, the old value for normal DFA1 was a bit over 1 and the new one is about 1.2. All of the studies referred to below use the old algorithm but future studies will increasingly use the newer one.
In the TRACE (TRAndopril Cardiac Evaluation study) which enrolled 159 patients with acute MI and with LV wall motion score <1.2 or LVEF <35%, Holter recordings were obtained 3.7 days postMI. Patients were followed for up to 4 years, at which time 45% had died. Among all analysed variables, reduced shortterm fractal scaling exponent (?1<0.85) was the best univariable predictor of mortality (relative risk 3.17, 95% CI 1.965.15, p<0.0001) with positive and negative predictive accuracies of 65% and 86% respectively. Even after adjustment for clinical covariates and LV function, in the Cox proportional hazards analysis, decreased ?1 was an independent predictor of mortality (p<0.001) [^{11}].
More recently, ?1 was shown to predict mortality in patients with heart failure. In a sub study of the DIAMOND CHF trial which enrolled patients with left ventricular wall motion index ?1.2 (equivalent to ejection fraction 35%), N=499 eligible patients were followed up for mean period of 665 ? 374 days, during which time N=210 died. After adjustment for age, functional class, medication and left ventricular ejection fraction in the multivariate proportional hazards analysis, decreased ?1 predicted mortality (RR=1.4, 95%CI 1.01.9, p<0.05). HRV was a better predictor of outcome among those with moderate heart failure (Class II) but HRV was not an independent predictor of mortality among those with severe heart failure (class III or IV).
Recently Laitio and his coworkers [^{12}] reported that reduced alpha1 is predictive of postoperative myocardial infarction in elderly patients having emergent surgery for traumatic hip fracture. N=32 patients, aged 60, with preoperative nighttime and daytime Holter recordings were studied. The short term fractal scaling exponent of RR intervals was assessed from 2:00 AM to 5 AM and 7AM to 12 noon for each patient. Preoperative ?1 was significantly lower during the nighttime compared with the daytime (0.92?0.08 vs. 1.03?0.06, p=0.002) in patients with postoperative MI. Results suggest that preoperatively decreased night time fractal correlation properties of HR dynamics may identify elderly patients at risk for developing myocardial ischemia in the early postoperative phase after nonvascular surgery.
The Poincar? plot is a graphic method for representing the underlying structure of the RR interval time series. In a Poincar? plot, also known as a return map plot, the time between each pair of beats is plotted against the time between the next pair of beats, i.e., the x, y coordinates of each point are rr_{i}, rr_{i+1}. An example is see in Figure 3. As is the case with the shortterm fractal scaling exponent, some investigators plot every interbeat interval in the heart rate time series and others use only the normaltonormal intervals. When all RR intervals are plotted, the assumption is made that only the beats in the central portion of the plot are normal and that outlier beats are ectopic. Plots of morphologically normaltonormal interbeat intervals have proved that this assumption is simplistic and misleading [^{10}].
In one of the earliest studies involving Poincar? plots in a cardiac patient population, Woo et al [^{13}] constructed Poincar? plots of sinus RR intervals from 24hour Holter recordings in N= 24 healthy control subjects and N=24 patients with heart failure. In the control group, plots had a comet shape resulting from an increase in beattobeat dispersion at slower heart rates. No heart failure patient displayed a cometshaped pattern. Instead, three distinctive patterns were identified: (1) a torpedo shaped pattern resulting from low RR interval dispersion over the entire range of heart rates, 2) a fanshaped pattern resulting from restriction of overall RR interval ranges with enhanced dispersion and 3) complex pattern with clusters of points characteristic of sudden changes in RR intervals. Thus, the Poincar? plots revealed a complexity in heart rate patterns, which could not be perceived from standard heart rate variability measures like SDNN (the standard deviation of all normaltonormal interbeat intervals). Indeed SDNN was not different in patients with different HR patterns. However, serum norepinephrine levels were significantly higher among heart failure patients with complex Poincar? plots, suggesting that these plots are a marker for increased sympathetic activation and might provide additional prognostic information and insight into autonomic alterations in patients with heart failure [^{14}].
We previously alluded to the presence of increased randomness in heart rate patterns and their potential as a marker of increased risk. As can be appreciated from Figure 3, because each point on the plot represent the relationship of one beat to the next, increased randomness of heart rate patterns can easily be recognized in the Poincar? plot as well. Figure 4 shows examples of normal and abnormal heart rate patterns seen on hourly Poincar? plots. These plots illustrate that an increased dispersion of points results in an increase in the relative width of the plot compared with its length. The nonlinear index SD12 quantifies the shape of the Poincar? plot. SD12 is the ratio of the lengths of the axes of an imaginary ellipse which has its center at the average RR (or NN) interval of the time series and is fitted to the Poincar? plot. As shown in Figure 3, the longitudinal axis of this ellipse is along a line beginning at the origin and with a slope of 1 and has a length called SD2. SD2 reflects intermediateterm variability. The length of the transverse axis of the ellipse is SD1 and reflects shortterm variability. SD12 is the ratio of these measures.
In our lab, HRV from 740 tapes recorded before antiarrhythmic therapy in the Cardiac Arrhythmia Suppression Trial was studied [^{15}]. CAST was an historic post MI study with impaired left ventricular ejection fraction and high grade ventricular arrhythmias that were suppressed with one of three antiarrhythmic therapies (encainide, flecainide or moricizine) [^{16}]. Holter recordings were obtained at a large range of times post MI. Subjects in the HRV substudy were patients with ventricular premature contractions (VPCs) suppressed on the first randomised antiarrhythmic treatment. They were 70?121 post MI and follow up was 362?241days (70 deaths). Traditional time and frequency domain HRV were determined from the qualifying, prerandomization Holter recording. Nonlinear HRV measures included shortterm fractal scaling exponent and SD12, each calculated from an average of 1000 normaltonormal interbeat intervals and power law slope calculated from the entire recording. As mentioned, this group of patients was at a broad range of times postMI whereas in the studies discussed earlier HRV was measured shortly after MI. As a result, some patients had undergone CABG surgery between their MI and their first Holter recordings. Decreased HRV among postCABG patients is caused by the surgery itself and has been shown to be of no prognostic value. Therefore, postCABG patients were excluded from the analysis [^{17}]. Similarly, in the CAST population, the decreased HRV associated with diabetes was not prognostic. Thus, when the analysis was limited to those without CABG surgery postMI or diabetes, power law slope had no association with mortality. A significant association was seen for DFA1 (p<0.001). However, increased SD12 had the strongest association with mortality. Importantly, however, in this study, we showed that traditional time and frequency domain HRV and nonlinear HRV were independent predictors of mortality. That is, when all clinical and demographic covariates were considered, the combination of increased SD12 (<55%), decreased ultra low frequency power (a measure of the circadian rhythm of HR), history of prior MI and history of CHF identified the patients at highest risk of mortality.
We also determined the predictive value of DFA1 in association with traditional frequency domain HRV in the Cardiovascular Health Study, a prospective, populationbased study of risk factors for heart disease and stroke among communitydwelling adults aged >65 year [^{18}]. Of these participants, 1227 had 24hour recordings eligible for frequency domain and nonlinear analysis that were obtained at the beginning of the study. After 8 years of follow up, 267 had died. On univariate analysis, decreased values for all frequency domain HRV indices except for high frequency power (HF, a measure of the power of vagallymodulated respiratory sinus arrhythmia), and also decreased DFA1, were significantly associated with mortality. When HRV indices were combined, even after adjustment for age, gender, presence of subclinical or clinical cardiovascular disease and diabetes, the combination of decreased DFA1 and decreased HF was the strongest predictors of mortality. In the adjusted model, the risk ratio for DFA1 was 0.15 (95% CI 0.080.29) and for (ln) HF it was 0.78 (95% CI 0.0690.88) all p<0.001. We hypothesize that HF power is significantly exaggerated by sinus arrhythmia of nonrespiratory origins, thus eliminating its association with mortality. However, when DFA1 is added to the model, it essentially controls for this randomness, permitting the association of decreased vagal modulation of HR and mortality to be determined.
Both increased SD12 and decreased DFA1 reflect increased randomness in the heart rate time series and are therefore highly correlated. As mentioned, DFA1 is dependent on accurate Holter scanning, but SD12 is less sensitive to scanning error. This might help explain its better prognostic value in the CAST study, mentioned above; where Holters were scanned on older equipment where accurate labelling of each beat would have taken days in some cases. Also, Poincar? plots can be examined visually and additional insights gained from their structure. While most studies that have used Poincar? plots examined patterns from 24hour recordings, 24hours of data represents >100,000 beats and details patterns may be obscured by looking only at the entire recording. In our lab, we examine hourly Poincar? plots and have noted that periods of increased randomness of HR patterns can occur in recordings that have many hours of normal patterns. Finally, power law slope, because it is based on a regression line, is relatively insensitive to scanning errors, but its prognostic value has not been as consistently verified.
It is clear that nonlinear HRV analysis has the potential to substantially improve the utility of HRV for risk stratification. The circumstances under which nonlinear HRV is applicable have not been fully elucidated. Nonlinear HRV is not available from commercial Holter scanning systems, but most export an annotated beattobeat file that, with a little bit of computer expertise, can be used to generate nonlinear HRV indices. Considerable care must be taken to accurately label the morphology of each beat. The Holter scanning system must provide a way to ensure that the interbeat intervals reflect a consistent detection of beat onset at the same point of each QRS, or a least of each series of QRS. Unfortunately, at the moment this is often not the case. Filters that automatically label beats by prematurity are not sufficient to insure accurate enough scanning, certainly for calculation of DFA1, at least when it is based on normaltonormal interbeat intervals. Computer technology is advancing rapidly. The need to riskstratify patients for implantable cardiac defibrillators has enlivened the search for Holterbased risk markers of increased risk. Because of these factors, it is likely that future Holter scanners will have far more sensitive beatdetection and beat classification algorithms, making it possible to obtain sufficiently accurate scanning to permit routine calculation of nonlinear HRV for risk stratification in the clinical setting.
References
1.  Kleiger RE,Miller JP,Bigger JT Jr,et al. Decreased Heart Rate Variability and its association with increased mortality after acute myocardial infarctionAm J Cardiol 1987;59:256–262. [pmid: 3812275] 
2.  Bigger JT Jr,Fleiss JL,Rolnitzky LM,et al. Time course of recovery of heart period variability after myocardial infarctionJ Am Coll Cardiol 1991;18:1643–1649. [pmid: 1960309] 
3.  Cripps TR,Malik M,Farrell TG,et al. Prognostic value of reduced heart rate variability after myocardial infarction: clinical evaluation of a new analysis methodBr Heart J 1991;65:14–19. [pmid: 1704246] 
4.  Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Heart rate variability standards of measurement, physiological interpretation, and clinical useCirculation 1996;93:1043–1065. [pmid: 8598068] 
5.  Kobayashi M,Musha T. 1/f fluctuation of heartbeat periodIEEE Trans Biomed Eng 1982;29:456–457. [pmid: 7106796] 
6.  Bigger JT Jr,Steinman RC,Rolnitzky LM,et al. Power law behaviour of RRinterval variability in healthy middleaged persons, patients with recent acute myocardial infarction, and patients with heart transplantsCirculation 1996;93:2142–2151. [pmid: 8925583] 
7.  Huikuri HV,M?kikallio TH,Airaksinen KE,et al. Powerlaw relationship of heart rate variability as a predictor of mortality in the elderlyCirculation 1998;97:2031–2036. [pmid: 9610533] 
8.  Lombardi F,Sandrone G,Mortara A,et al. Linear and nonlinear dynamics of heart rate variability after acute myocardial infarction with normal and reduced left ventricular ejection fractionAm J Cardiol 1996;15:1283–1288. [pmid: 8677867] 
9.  Peng CK,Havlin S,Stanley HE,et al. Quantification of scaling exponents and crossover phenomena in nonstationary heartbeat time seriesChaos 1995;5:82–87. [pmid: 11538314] 
10.  Stein PK,Domitrovich PP,Hui N,et al. Sometimes Higher Heart Rate Variability is Not Better Heart Rate Variability: Results of Graphical and NonLinear AnalysesJournal of Cardiovascular Electrophysiology 2005;16:1–6. 
11.  M?kikallio TH,Hoiber S,Kober L,et al. Fractal analysis of heart rate dynamics as a predictor of mortality in patients with depressed left ventricular function after acute myocardial infarction. TRACE Investigators. TRAndolapril Cardiac EvaluationAm J Cardiol 1999;83:836. [pmid: 10190395] 
12.  Laitio TT,Huikuri HV,M?kikallio TH,et al. The breakdown of fractal heart rate dynamics predicts prolonged postoperative myocardial ischemiaAnesth Analg 2004;98:1239–1244. [pmid: 15105194] 
13.  Woo MA,Stevenson WG,Moser DK,et al. Patterns of beattobeat heart rate variability in advanced heart failureAm Heart J 1992;123:704–710. [pmid: 1539521] 
14.  Woo MA,Stevenson WG,Moser DK,et al. Complex heart rate variability and serum norepinephrine levels in patients with advanced heart failureJ Am Coll Cardiol 1994;23:565–569. [pmid: 8113535] 
15.  Stein PK,Domitrovich PP,Huikuri HV,et al. Traditional and nonlinear heart rate variability are each independently associated with mortality after myocardial infarctionJ Cardiovasc Electrophysiol 2005;16:113–120. 
16.  Epstein AE,Bigger JT Jr,Wyse DG,et al. Events in the Cardiac Arrhythmia Suppression Trial (CAST): mortality in the entire population enrolledJ Am Coll Cardiol 1991;18:14–19. [pmid: 1904891] 
17.  Demirel S,Akkaya V,Oflaz H,et al. Heart rate variability after coronaryartery bypass graft surgery: a prospective 3year followup studyAnnals of Noninvasive Electrocardiology 2002;7:247–250. [pmid: 12167187] 
18.  Stein PK,Domitrovich PP,Gottdiener JS,et al. The Combination of Decreased ShortTerm Fractal Scaling Exponent and Decreased High Frequency Power Provides a Powerful Independent Predictor of Mortality in the ElderlyCirculation 2004;110(suppl III):III457. 
19.  Brouwer J,van Veldhuisen DJ,Man in 't Veld AJ,et al. Prognostic value of heart rate variability during longterm followup in patients with mild to moderate heart failure. The Dutch Ibopamine Multicenter Trial Study GroupJ Am Coll Cardiol 1996;28:1183–1190. [pmid: 8890814] 
20.  Huikuri HV,M?kikallio TH,Peng CK,et al. Fractal correlation properties of RR interval dynamics and mortality in patients with depressed left ventricular function after an acute myocardial infarctionCirculation 2000;101:47–53. [pmid: 10618303] 
21.  Kamen PW,Tonkin AM. Application of the Poincar? plot to heart rate variability: a new measure of functional status in heart failureAust N Z J Med 1995;25:18–26. [pmid: 7786239] 
22.  Laitio TT,Huikuri HV,Kentala ES,et al. Correlation properties and complexity of perioperative RRinterval dynamics in coronary artery bypass surgery patientsAnesthesiology 2000;93:69–80. [pmid: 10861148] 
23.  M?kikallio TH,Ristimae T,Airaksinen KE,et al. Heart rate dynamics in patients with stable angina pectoris and utility of fractal and complexity measuresAm J Cardiol 1998;81:27–31. [pmid: 9462601] 
Alpha1 (?1)  shortterm fractal scaling exponent, a nonlinear measure of the randomness or correlatedness of heart rate patterns 
CABG  coronary artery bypass grafting 
DFA1  same as alpha1 
HRV  heart rate variability 
SDNN  standard deviation of all normaltonormal interbeat intervals in ms, a traditional measure of heart rate variability 
SD1  short axis of ellipse fitted around a Poincar? plot of NN intervals in ms 
SD2  long axis of an ellipse fitted around a Poincar? plot of NN intervals in ms 
SD12  SD1/SD2 
Article Categories:
Keywords: heart rate variability, risk stratification, non linear, post myocardial infarction, congestive heart failure. 
Previous Document: Magnetic resonance imaging in patients with ICDs and pacemakers.
Next Document: Methods of assessment and clinical relevance of QT dynamics.