Respiratory sinus arrhythmia as a predictor of sudden cardiac death after myocardial infarction

Abstract

Background. Measurement of high-frequency (HF) spectral power of heart rate (HR) variability has not been able to identify the patients at risk of sudden cardiac death (SCD) despite the experimental evidence of protective role of vagal activity for fatal arrhythmias.

Aim. We developed a novel respiratory sinus arrhythmia (RSA) analysis method and tested its ability to predict SCD after an acute myocardial infarction.

Method. The RSA analysis method was developed in 13 subjects from simultaneous recordings of respiration and R-R intervals. An adaptive threshold was computed based on the zero-phase forward and reverse digital filtering in the analysis of RSA. With this method, only respiration-related R-R interval fluctuations are included. The prognostic power of RSA, analyzed from 24-hour electrocardiographic recordings, was subsequently assessed in a large postinfarction population including 1631 patients with mean follow-up of 40±17 months.

Results. Depressed RSA was a strong predictor of SCD (hazard ratio 7.4; 95% CI 3.6–15.1; P <0.0001) but only a weak predictor of non-SCD. The RSA index remained an independent predictor of SCD after adjustments for ejection fraction and other clinical risk variables (RR 4.7; 95% CI 2.28–9.85).

Conclusions. Reduced respiratory-related HR dynamics, detected by RSA index, are a specific marker of an increased risk of SCD among postinfarction patients.

• Heart rate
• respiratory sinus arrhythmia
• sudden cardiac death

Several indexes of heart rate (HR) variability have been used in the assessment of cardiac autonomic regulation 1, 2. High-frequency (HF) oscillations of HR have been shown to be most directly related to cardiac vagal outflow 1, 3–6. Animal studies have demonstrated the protective role of cardiac vagal activity in preventing fatal arrhythmias 7. In the light of these observations, it has been surprising that measurement of HF spectral power from 24-hour Holter recordings has not been able to provide information on the risk of sudden cardiac death (SCD) or life-threatening arrhythmic events in previous studies 8–13. In fact, it has had the lowest predictive value among various HR variability indexes 8–13.

Methodological issues may explain why an association between the reduced 24-hour HF spectral power and the increased risk of SCD has not been apparent. Firstly and importantly, it is evident that HF spectral power measured from 24-hour electrocardiographic (ECG) recordings includes a lot of nonrespiratory R-R interval fluctuations 14–18. Secondly, editing methods of random oscillations of R-R intervals and ectopic beats have a significant influence particularly on HF spectral power 19.

The aim of the first part of the present study was to develop a method to quantify the respiratory-specific HR fluctuation from ECG recordings with simultaneous respiration signal. The aim of the second part of the study was to test the developed method in 24-hour ECG recordings without information of respiration signal, which is not recorded in standard Holter recordings. We quantified the respiratory sinus arrhythmia (RSA) by designing and implementing an algorithm that emphasizes only the respiratory-associated R-R interval variations during the 24-hour ECG recording. We then assessed the prognostic power of this new algorithm as a predictor of SCD and non-SCD in a large population of patients with myocardial infarction (MI).

Methods

Development of algorithm

For the development of the RSA quantification algorithm, ECG and breathing signals of 13 studied subjects of healthy male volunteers were recorded simultaneously. Electrocardiographic signals were recorded with Cardiolife recorder (TEC-7721K, Nihon Kohden, Tokyo, Japan) and simultaneous breathing signals were obtained with a temperature sensor (Hewlett Packard, Germany), so that the maximum of the respiration signal appeared at the end of expiration. Signal recordings were obtained with a sampling frequency of 512 Hz in laboratory conditions. Mean length of the recordings was 38±4 minutes, and the recordings were performed so that both metronome-guided (15 breaths/minute) and spontaneous breathing patterns occurred during the recordings. Various physiological stimulations, such as cold face, handgrip, and cold hand tests, were also included in the recordings to obtain possible irregularities in the R-R interval fluctuation during the metronome-guided breathing.

The respiratory segment validation criteria were formed as follows: First, R-R interval oscillations were examined during the metronome-guided expiration and inspiration. Peak detection was performed for the R-R interval time series to obtain local R-R interval maximums and minimums. The local minimums of the R-R interval time series occurred at the minimum of the breathing signal (end of inspiration). Correspondingly, the local maximums of the R-R interval time series occurred at the maximum of the breathing signal (end of expiration). The distances between adjacent local maximums and minimums were examined. Then, the forward and reverse zero-phase low-pass digital filtering was performed to obtain an adaptive threshold for the R-R interval time series, i.e. the filtering was done twice to avoid the phase distortion, first from the beginning to the end of the R-R signal and next in the reverse direction starting from the end of the signal. The obtained threshold is adaptive, and therefore it changes along the variation of the R-R interval time series. The R-R interval fluctuation around the adaptive threshold was observed so that the local R-R interval maximums and minimums corresponding to maximums and minimums of the breathing signal appear alternately around the adaptive threshold. Figure 1 illustrates the RSA quantification algorithm with an example of the peak detection and filtering process of the R-R interval time series. To obtain an adequate number of R-R intervals for the short-term spectrum analysis of the HF power the minimum size of the respiratory R-R interval segments was defined as 256 R-R intervals, corresponding to 3–5 minutes of ECG recordings. The minimum number of respiratory R-R interval segments that one ECG recording had to include was defined as 10. In addition, the developed algorithm was tested on the R-R intervals during the spontaneous breathing and during the physiological stimulations to study its functionality with possible irregular variation of the R-R intervals. Lastly, the R-R interval segments matching the above-defined criteria were separated for the analysis of the HF power and the new RSA index. Figure 2 illustrates an example of nonrespiratory R-R interval changes being discarded from the RSA analysis. The average value for the HF power of the respiratory-associated R-R interval segments was computed with the fast Fourier method to obtain the power of the respiratory-linked HF fluctuation, called Respiratory Sinus Arrhythmia index (RSA index).

Key messages

• Reduced respiratory-related heart rate dynamics, detected by the respiratory sinus arrhythmia index, are a specific marker of an increased risk of sudden cardiac death among postinfarction patients.

Figure 1.  Example of the R-R interval episode illustrating the variation around the adaptive threshold (dotted line). The minimums (star) and maximums (diamond) of the local R-R intervals are marked.

Figure 2.  Example of the nonrespiratory sudden changes in the R-R interval time series that are discarded from the respiratory sinus arrhythmia (RSA) analysis (light gray segment). At top an R-R interval tachogram, in the middle the ECG printout, and below the corresponding breathing signal.

Subjects and measurements

The developed RSA algorithm was tested in a large patient population with MI 10. The details of this study population have been described elsewhere 10. From total of 2130 patients, 499 patients were excluded due to a large amount of technical and biological disturbances, such as periods of atrial fibrillation, large amounts of ectopy, and lack of periodic R-R interval fluctuations that did not fulfill the above-mentioned RSA algorithm criteria. The average number of the obtained RSA segments were 33±23 corresponding to approximately 165±115 minutes of the ECG recording. In addition to RSA index computation, traditional HR variability indexes, such as the standard deviation of normal-to-normal R-R interval, standard high-frequency, low-frequency, and very-low-frequency spectral powers were computed from the 24-hour R-R interval time series with previously described methods 10. R-R interval time series containing ectopic beats and other disturbances were edited with the interpolation of degree zero.

Patients were followed up for 40±17 months after the MI. The follow-up time was extended from the previous reports of this population to include larger number of events 9, 10. In cases of death, the causes of death were verified from the hospital and autopsy records and from either the primary physicians or those who had witnessed the death. The end point committees of the participating institutes defined the modes of death. Deaths were defined as cardiac and noncardiac deaths. In addition, the cardiac deaths were further classified as sudden or nonsudden 9, 10. Cardiac death was defined as sudden if it was 1) a witnessed death occurring within 60 minutes from the onset of new symptoms unless a cause other than cardiac was obvious, 2) an unwitnessed death (within <24 hours) in the absence of preexisting progressive circulatory failure or other causes of death, or 3) death during attempted resuscitation.

Statistics

The data were analyzed using the SPSS software (SPSS 11.0, SPSS Inc., Chicago, Illinois, USA). The primary end point of this study was SCD. The selection of a sample size of more than 1500 patients with an average follow-up of over 36 months was based on the assumption that the annual incidence of the primary end point varies from 0.5% to 4% in the current treatment era. A reliable SCD analysis was evaluated to require a minimum of 30 end points. Survival curves were estimated by Kaplan-Meier method with comparisons of cumulative end points based on logarithmic transformations. Multivariate analyses were performed using the Cox proportional hazards model and adjusted with the significant clinical risk factors (diabetes, advanced age, and depressed left ventricular function). The effects of the factors investigated are given as hazard ratios with 95% confidence intervals. All statistical tests, including log-rank tests in the Cox model, were two-sided and assessed at a 5% significance level. Cox regression analysis was performed using the optimum cutoff points based on the receiver operating characteristic curves.

Results

The performance of the RSA quantification algorithm was manually examined for 13 healthy subjects. Visual analysis revealed that the algorithm could extract the respiratory-related R-R intervals with high accuracy.

The characteristics of the patient population are shown in Table I. During a mean follow-up of 40 months, the total number of deaths was 147. Of these deaths, 89 were cardiac and 58 were noncardiac. Of the cardiac deaths, 47 were non-SCDs and 42 were SCDs. The annual incidence of SCD was 0.8%. The hazard ratios of the various HR variability risk variables for SCD and non-SCD are listed in Table II.

Table I.  Clinical characteristics of the study population.

Clinical:	Age (years)	59±10
	Gender (male/female)	1279/352
	Smoking (ex or current)	984 (61%)
	Diabetes	288 (18%)
	Prior MI	199 (12%)
	Hypertension	936 (57%)
Treatment:	PCI/CABG	1142 (70%)
	Q-MI no revasc	82 (5%)
Medical:	ASA/warfarin	1567 (96%)
	Beta-blockers	1541 (95%)
	Statins	1152 (71%)
	ACE/AT II	1206 (74%)
	Diuretic	529 (32%)

Data are presented as n (%).

ACE = angiotensin-converting enzyme; AT II = angiotensin II receptor; CABG = coronary artery bypass grafting; MI = myocardial infarction; PCI = percutaneous coronary intervention.

Q-MI no revas = Q wave myocardial infarction without revascularization; ASA = acetylsalicyclic acid.

Table II.  Relative risk variables as predictors of SCD and non-SCD.

	SCD		Non-SCD
	RR	P-value	RR	P-value
Univariate:
EF < 45 (%)	5.76 (2.89–11.47)	0.000	4.57 (2.44–8.57)	0.000
SDNN < 90 (ms)	2.14 (0.51–8.91)	0.296	3.55 (1.27–9.90)	0.015
HF < 70 (ms^2)	3.79 (1.89–7.57)	0.000	2.36 (1.20–4.63)	0.013
LF < 170 (ms^2)	4.23 (1.85–9.67)	0.001	2.66 (1.13–6.28)	0.025
VLF < 700 (ms^2)	3.85 (1.92–7.69)	0.000	2.62 (1.47–4.67)	0.001
RSA index < 200	7.40 (3.64–15.06)	0.000	2.09 (1.18–3.71)	0.012
Multivariate:
SDNN < 90 (ms)	1.25 (0.29–5.29)	0.004	2.07 (0.72–5.90)	0.176
HF < 70 (ms^2)	2.35 (1.13–4.89)	0.022	1.17 (0.57–2.35)	0.689
LF < 170 (ms^2)	1.74 (0.72–4.19)	0.216	0.84 (0.34–2.05)	0.693
VLF < 700 (ms^2)	1.90 (0.90–4.00)	0.093	1.03 (0.55–1.94)	0.928
RSA index < 200	4.74 (2.28–9.85)	0.000	1.08 (0.59–1.99)	0.805

EF = ejection fraction; HF = standard 24-hour high frequency power of the heart rate variability; LF = low frequency power of the heart rate variability; RR = relative risk; RSA index = respiratory sinus arrhythmia index; SCD = sudden cardiac death; SDNN = standard deviation of normal-to-normal R-R intervals; VLF = very low frequency power of the heart rate variability.

In univariate analysis, various Holter-based risk indexes, including standard deviation of R-R intervals and spectral measures of HR variability, were associated with an increased risk of SCD and non-SCD (Table II). Depressed respiratory-associated HF fluctuation, expressed by a reduced RSA index (<200 ms²), was the measure with specific predictive value for SCD, indicating a 7-fold risk of a sudden death during the follow-up (RR 7.4; 95% CI 3.6–15.1; P < 0.0001). In a multivariate analysis after adjustment for age, left ventricular systolic function, and history of diabetes, abnormal RSA index remained a predictor of particularly SCD (RR 4.7; 95% CI 2.3–9.9; P < 0.0001) but no longer predicted non-SCD (Table II). The other indexes of HR variability were strong predictors of both non-SCD and SCD without the capability to predict SCD in particular (Table II).

The Kaplan-Meier survival curve in Figure 3 shows that especially for the patients with a preserved RSA index (>200 ms²), the SCD rate was very low, approximately 1% up to 3 years, and <2% even after 4 years.

Figure 3.  Kaplan-Meier survival curves for sudden cardiac death (SCD) among patients with depressed respiratory sinus arrhythmia (RSA) index and RSA index over 200 ms².

Discussion

The main finding of the present study is that depressed RSA, measured by the RSA index, is a specific risk marker for the SCD among patients with prior MI. Although other risk markers, such as conventional HR fluctuation variables and left ventricular function, have been shown to be associated with an increased risk of cardiac mortality in this population 10, these indexes have only limited value for sorting out the risk for either SCD or non-SCD. The RSA index was the only measure that appeared to be a specific indicator of increased risk for the SCD even after adjustments with significant clinical risk factors, such as advanced age, diabetes, and depressed left ventricular systolic function.

Previous observational follow-up studies have shown that the 24-hour HF spectral component has the lowest predictive accuracy for mortality compared to other spectral indexes among patients with MI 8. Consistently with these findings, HF spectral component has not been able to differentiate between the patients with and without life-threatening events in case-control studies, nor has it been able to predict the onset of imminent life-threatening arrhythmic events 20. Methodological differences may well explain the differences in the results between previous studies and the present one. For one thing, despite the HF spectral component being mostly associated with respiration, the HF spectral area analyzed from the standard 24-hour Holter recordings includes R-R interval oscillations that are not caused by respiration. Nonautonomic, erratic oscillations that affect the HF spectral area produce a bias in the quantification of the RSA with the standard 24-hour HF spectral analysis methods. Such nonrespiratory fast changes in the R-R interval time series increase the power of the HF spectral component. This type of nonrespiratory fast R-R interval oscillations has been most commonly observed in patients with congestive heart failure, resulting in fan-shaped Poincaré plots 21, 22 and increased indexes describing short-term HR variability by standard analysis methods 15. For another thing, most of the 24-hour Holter recordings include both biological and technical disturbances. It has been shown that spectral analysis of the HR variability is sensitive to editing, and careful attention should be paid to the selection of the method of editing the disturbances in the R-R interval time series. The commonly used deletion method of R-R interval editing had a clear effect on the HF fluctuation spectral values 23. The deletion method increases the HF spectral component in high-risk patients with frequent ectopic beats. Indeed, HF spectral component, analyzed by the standard method, performed somewhat better in the present study than in previous studies, perhaps owing to an improved editing method. Furthermore, physical activity also influences the level of 24-hour HF-power, so that patients who are more active paradoxically have a reduction in 24-hour HF band because the HF oscillations are reduced during physical activity 17.

All traditional HR variability indexes predicted cardiac mortality in the present study. However, these indexes lack the ability to predict specifically SCD. Long-term HR variability indexes, which have been most commonly used in previous studies, predicted non-SCD far better than they predicted SCD. These indexes describe partly the activity level of the patients and thereby seem to reflect frailty and poor overall health of the patients. Short-term HR variability measures, such as RSA, seem to be better indexes of specific autonomic disturbances that increase the vulnerability to SCD.

Potential limitations

A few limitations are obvious in the present study. First, the quantification of the RSA is based on the algorithm developed for such electrocardiographic recordings that are obtained without the information of the respiration signal (standard Holter recordings). Inclusion of the simultaneous respiration signal in the analysis of the R-R interval time series would increase the likelihood of the selected R-R interval segments being affected only by the respiration. Second, RSA could not be analyzed in all patients due to the lack of artifact-free R-R interval data for a relatively large patient sample. This fact reflects a potential weakness of the developed analysis method. Future studies should investigate whether an analysis of short-term RSA during controlled conditions, which could be determined for the majority of patients, could provide even better prognostic information.

Implications

The present findings confirm the results of the experimental studies that vagal activity plays an important role in preventing fatal arrhythmias 7. Patients with well preserved RSA had an extremely low risk for SCD despite depressed left ventricular function after MI. Preserved RSA index had a high negative predictive accuracy for SCD, and its predictive power for SCD was better than that of ejection fraction. This suggests that the RSA index could be a useful tool in selecting patients for implantable cardioverter-defibrillator and, especially, in excluding patients with depressed left ventricular function from costly and sometimes cumbersome therapy.

Several indexes of HR variability have been used and tested in various clinical settings since the original observation of Ewing et al. 24 that short-term HR variability quantifies predominantly the cardiac vagal outflow, and the finding of Kleiger et al. 11 that predicts mortality after MI. Respiratory sinus arrhythmia does not always reflect cardiac vagal activity 25–27, and none of the HR variability indexes have become a routinely used clinical tool. This is mostly due to the fact that none of the HR variability indexes have been able to predict specific modes of death. Therefore, the HR variability indexes have not been found to be very useful for tailoring specific therapy, such as implantation of cardioverter-defibrillator. The present study suggests that modern signal processing of R-R interval time series may offer some advantages over previously described methods in specific prediction of SCD. The role of the RSA index, measured from standard Holter recordings, should be tested in comparison to other risk markers in future trials.

Acknowledgements

The authors wish to acknowledge Mrs Anne Lehtinen, Pirkko Huikuri, RN, and Päivi Karjalainen, RN, for their assistance.

Supported by the Foundation for Cardiovascular Research, the Paulo Foundation, Instrumentarium Foundation for Science and the Research and Science Foundation of Farmos, Helsinki, Finland, Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie, Berlin, Germany, Kommission für Klinische Forschung, Munich, Germany, and Deutsche Forschungsgemeinschaft, Bonn, Germany, and the Sixth Framework Programme of the European Union (DAPHNet, 01847–2).

None of the authors have any conflict of interest.