Author + information
- Received February 3, 2016
- Revision received March 2, 2016
- Accepted March 7, 2016
- Published online May 17, 2016.
- Daniel Sinnecker, MDa,
- Michael Dommasch, MDa,
- Alexander Steger, MDa,
- Anna Berkefeld, MDa,
- Petra Hoppmann, MDa,
- Alexander Müller, MSca,
- Josef Gebhardt, MSca,
- Petra Barthel, MDa,
- Katerina Hnatkova, PhDb,
- Katharina M. Huster, MDa,
- Karl-Ludwig Laugwitz, MDa,c,
- Marek Malik, MD, PhDb and
- Georg Schmidt, MDa,c,∗ ()
- a1. Medizinische Klinik und Poliklinik, Klinikum rechts der Isar, Technische Universität München, Munich, Germany
- bSaint Paul's Cardiac Electrophysiology, University of London and Imperial College, London, London, United Kingdom
- cDZHK (German Center for Cardiovascular Research), partner site Munich Heart Alliance, Munich, Germany
- ↵∗Reprint requests and correspondence:
Dr. Georg Schmidt, 1. Medizinische Klinik und Poliklinik, Klinikum rechts der Isar, Technische Universität München, Ismaninger Strasse 22, 81675 München, Germany.
Background Respiratory sinus arrhythmia (RSA), a measure of cardiac vagal modulation, provides cardiac risk stratification information. RSA can be quantified from Holter recordings as the high-frequency component of heart rate variability or as the variability of RR intervals in individual respiratory cycles. However, as a risk predictor, RSA is neither exceptionally sensitive nor specific.
Objectives This study aimed to improve RSA determination by quantifying the amount of sinus arrhythmia related to expiration (expiration-triggered sinus arrhythmia [ETA]) from short-term recordings of electrocardiogram and respiratory chest excursions, and investigated the predictive power of ETA in survivors of acute myocardial infarction.
Methods Survivors of acute myocardial infarction (N = 941) underwent 30-min recordings of electrocardiogram and respiratory chest excursions. ETA was quantified as the RR interval change associated with expiration by phase-rectified signal averaging. Primary outcome was 5-year all-cause mortality. Univariable and multivariable Cox regression was used to investigate the association of ETA with mortality.
Results ETA was a strong predictor of mortality, both in univariable and multivariable analysis. In a multivariable model including respiratory rate, left ventricular ejection fraction, diabetes mellitus, and GRACE score, ETA ≤0.19 ms was associated with a hazard ratio of 3.41 (95% confidence interval: 1.10 to 5.89, p < 0.0001). In patient subgroups defined by abnormal left ventricular ejection fraction, increased respiratory rate, high GRACE score, or presence of diabetes mellitus, patients were classified as high or low risk on the basis of ETA.
Conclusions Expiration-triggered sinus arrhythmia (ETA) is a potent and independent post-infarction risk marker.
- heart rate variability
- phase-rectified signal averaging
- respiratory sinus arrhythmia
- risk stratification
Respiratory sinus arrhythmia (RSA)—first described by Carl Ludwig more than 150 years ago (1)—is caused by phasic changes of vagal neural discharge directed to the sinus node. During inspiration, vagal activity is inhibited, causing a prompt heart rate increase. During expiration, the pattern of vagal discharge resumes, causing a prompt heart rate decrease (2,3). Thus, the extent of the corresponding heart rate changes is largely dependent on vagal modulation.
RSA is therefore a major determinant of heart rate variability (HRV), particularly when considering short-term HRV. RSA decrease is commonly observed in conditions characterized by altered sympathovagal balance with reduced vagal and increased sympathetic neural activity. In many clinical conditions (e.g., myocardial infarction [MI], heart failure), this autonomic pattern has been associated with a poor prognosis (4).
Various parameters have been proposed for RSA quantification and its use as a predictor for subsequent mortality in cardiac patients (5,6). Commonly, these markers quantify the HRV within entire respiratory cycles. We present a method that addresses heart rate changes mediated by the expiration-driven increase of vagal activity. For this purpose, we used bivariate phase-rectified signal averaging (7,8) technology to quantify fluctuations of heartbeat intervals triggered by expiration. We term such fluctuations expiration-triggered sinus arrhythmia (ETA). We also investigated the prognostic power of this parameter for risk stratification of survivors of acute MI.
From May 2000 to March 2005, 941 consecutive survivors of acute MI were enrolled at 2 university hospitals (Deutsches Herzzentrum München and Klinikum Rechts der Isar, both in Munich, Germany). The study protocol was approved by the local ethics committee, and all participants gave written informed consent.
Patients were eligible for the study if they were ≤80 years of age, survived the acute MI phase, presented in sinus rhythm, and did not meet criteria for implantable cardioverter-defibrillator implantation for secondary prevention before hospital discharge. Mean follow-up was 4.9 years, with the last follow-up performed in August 2010. The primary endpoint of the study was all-cause mortality within 5 years after the index MI. Patients were followed up with clinical appointments every 6 months. If patients did not attend, they were contacted by letter, by telephone, or through their general practitioner. If this was not successful, the local population registry office was contacted to trace their new address or to identify those who died.
Within 2 weeks of their index MI and during the initial hospitalization, patients underwent 30-min recordings of high-resolution electrocardiograms (ECGs) (1.6 kHz in orthogonal XYZ leads, TMS International, Enschede, the Netherlands) and respiratory activity (Pro-Tech piezoelectric thoracic sensor, signal acquired at 1.6 kHz, Porti system, TMS International). To avoid disturbing spontaneous respiration, patients were not reminded that a sensor was attached to monitor respiration (although they initially consented to their ECG, arterial pressure, and respiration being monitored). Recordings were made in the morning after administration of routine medications, in a supine resting position in a quiet surrounding. In analogy to the standards of Holter ECG analysis (9), raw signals were reviewed, artefacts were eliminated where needed, and QRS classifications were carefully reviewed and manually corrected when appropriate. This post-processing, which typically took <5 min, was performed by an experienced technician unaware of the clinical outcome data.
Assessment of RSA
A mathematical representation of the respiratory signal (analytical respiratory signal) was constructed, ranging from −1 (beginning of inspiration) to +1 (end of expiration) (Figure 1). The expiratory phase was defined as the time interval between 2 threshold values: ϕ1 and ϕ2. For computation of ETA, heartbeat intervals starting between ϕ1 and ϕ2 were identified as anchors. About 700 of 2,000 RR intervals become anchors in a typical 30-min recording. RR interval segments around the anchors were selected and averaged. The average change in RR intervals was quantified by Haar wavelet analysis with a scale of 2 (8).
The numerical values of ϕ1 and ϕ2 (0.6 and −0.1, respectively) were optimized with respect to prediction of 5-year all-cause mortality, as assessed by receiver-operating characteristic (ROC) analysis (details are presented in the Online Appendix).
Figure 2A shows a typical phase-rectified and averaged RR interval tachogram centered on expiratory RR intervals in a long-term MI survivor. A positive central deflection is clearly visible, indicating that, on average, expiration leads to RR interval prolongation (i.e., heart rate decrease) (Figure 2A). By contrast, Figure 2B shows an averaged tachogram of a patient who died during follow-up. In nonsurvivors, the central deflection was typically less pronounced or even negative.
RSA was also assessed by 2 standard methods: a spectral analytic technique, that is, the high-frequency component of HRV—RSAHF, and a peak-to-trough algorithm that estimates RSA from the variability of RR intervals in individual respiratory cycles—RSAPT. For calculation of RSAHF, the high-frequency (0.15 to 0.40 Hz) component of HRV was obtained by fast Fourier transformation in 5-min segments of the RR interval time series, and RSAHF was calculated as the average of all segments (5). For the calculation of RSAPT, inspiratory and expiratory phases of each respiratory cycle were analyzed separately, and the minimum RR interval during inspiration was subtracted from the maximum RR interval during expiration. RSAPT was calculated as the average of these differences over all respiratory cycles (6).
Definitions and clinical variables
For the initial MI diagnosis, at least 2 of the following 3 criteria were required: typical chest pain for at least 20-min duration, creatine kinase above twice the upper normal limit, and ST-segment elevation ≥0.1 mV in at least 2 limb leads or ≥0.2 mV in at least 2 contiguous precordial leads at admission.
Diabetes mellitus was diagnosed if a patient was already receiving treatment (diet, tablets, or insulin) or if fasting blood glucose concentration repeatedly exceeded 11 mmol/l. GRACE score (a clinical score combining patient age, serum creatinine, past MI, past heart failure, in-hospital percutaneous coronary intervention, heart rate, systolic blood pressure, ST-segment deviation, and positive enzymes) was assessed as previously proposed for the prediction of long-term prognosis (10).
Left ventricular ejection fraction (LVEF) was assessed by angiography (n = 445; 47.3%) or biplane echocardiography according to Simpson’s method (n = 496; 52.7%; Sonos 5500, Hewlett Packard, Palo Alto, California) within the first 2 weeks (median 7 days, interquartile range [IQR]: 5 to 9 days) after the index MI.
Evaluation of risk predictors
In the primary analysis, ETA, respiration rate, LVEF, and GRACE score were treated as continuous variables. In the secondary analysis, ETA was dichotomized at the optimum cutoff according to Youden’s index J (i.e., maximizing the sum of sensitivity and specificity) (11). Other variables were dichotomized prospectively: respiratory frequency at 18.6 breaths/min (12,13), LVEF at 35% (14), and the GRACE score at 120 points (15).
Continuous variables are presented as median and IQR. Categorical data are expressed as absolute frequencies and percentages. Cox proportional hazards models were used with all variables entered simultaneously to assess the independence and prognostic value of mortality predictors. Survival curves were estimated by the Kaplan-Meier method and compared using the log-rank test. To assess the intradata reproducibility of the findings, Kaplan-Meier survival curves and areas under the ROC curves were calculated together with their confidence intervals (CIs) obtained by bootstrap with 10,000 repetitions. Differences were considered statistically significant if p < 0.05 (IBM SPSS Statistics 20.0, SPSS, Armonk, New York).
Table 1 shows clinical and demographic characteristics of the enrolled patients. Five years after the index MI, 72 patients (7.7%) had died.
ETA as a continuous variable
Median ETA in the cohort was 0.54 ms (IQR: −0.25 to 1.66 ms). In patients who died during follow-up, ETA was significantly smaller compared with that of the survivors: −0.35 (−1.33 to 0.30) ms versus 0.64 (−0.16 to 1.75) ms (p < 0.0001). The area under the ROC curve for all-cause mortality prediction by ETA was 0.72 (95% CI: 0.67 to 0.78).
In univariable Cox analysis, ETA was a significant predictor of outcome. An increase of ETA by 1 ms was associated with a hazard ratio of 0.94 (95% CI: 0.90 to 0.97; p = 0.001) (Table 2). Respiratory rate, LVEF, presence of diabetes mellitus, and GRACE score also were significant predictors of all-cause mortality (Table 2). In multivariable Cox analysis considering all the aforementioned risk predictors, the association of ETA with mortality remained statistically significant, with a hazard ratio of 0.95 (95% CI: 0.91 to 0.99; p = 0.035) (Table 2).
Performance of the standard methods for RSA assessment was much weaker, with areas under the ROC curve of 0.57 (95% CI: 0.49 to 0.64) and 0.62 (95% CI: 0.55 to 0.69) for RSAHF and RSAPT, respectively; neither parameter reached statistical significance in univariable Cox analysis.
ETA as a dichotomous variable
According to Youden’s index J, the optimum dichotomy for ETA was 0.19 ms. In univariable Cox analysis, the hazard ratio for ETA ≤0.19 ms versus >0.19 ms was 5.12 (95% CI: 3.00 to 8.73; p < 0.001). Kaplan-Meier survival probability curves of patients with ETA values above and below this threshold are shown in the Central Illustration together with their CIs. Five-year mortality rates in the high- and low-risk groups were 15% and 3%, respectively. In multivariable Cox analysis, ETA ≤0.19 ms remained a strong risk predictor (with a hazard ratio of 3.43; 95% CI: 1.10 to 5.89; p < 0.0001) (Table 3). Upon addition of ETA to a model comprising respiratory rate, LVEF, presence of diabetes mellitus, and GRACE score, the area under the ROC curve increased significantly from 0.79 to 0.82 (p = 0.02) (Online Figure 1).
Figure 3 shows Kaplan-Meier survival probability curves for combinations of dichotomized ETA with dichotomized LVEF, respiratory rate, GRACE score, and presence or absence of diabetes mellitus. In all cases, the addition of ETA led to strong separation of survival curves. Within patient subgroups defined by abnormal LVEF, respiratory rate, GRACE score, or presence of diabetes mellitus, ETA >0.19 ms was consistently associated with a considerably improved prognosis, and vice versa for ETA ≤0.19 ms (Table 4, Central Illustration).
This study demonstrates that ETA is a potent risk marker in post-infarction patients. ETA is particularly strong in patients with reduced LVEF (Figure 3A). This observation is of clinical importance, because risk stratification in patients with reduced LVEF has so far been very challenging. ETA allows for further risk stratification, not only in patients pre-stratified by LVEF, but also by other accepted risk predictors such as GRACE score, presence of diabetes mellitus, or respiratory rate, with hazard ratios between 3 and 5 (Table 4). Moreover, ETA is independent of a number of established risk markers, including late potentials in signal-averaged ECGs (16), heart rate turbulence (17), deceleration capacity (18), spontaneous baroreflex sensitivity (15,19), post-extrasystolic blood pressure potentiation (20), and periodic repolarization dynamics (21) (Online Appendix), and standard methods of RSA assessment, that is, RSAHF and RSAPT, were by far not as strong as ETA. The predictive power of ETA was independent from β-blocker use, as well as from infarct localization (Online Appendix).
We understand that ETA extracts the effects of the expiration-triggered vagal surge on the sinus node discharge rate. Because attenuated vagal function is apparent already at early stages of left ventricular dysfunction (22), ETA seems particularly suitable for risk stratification purposes. Also, the phase-rectified signal averaging method makes ETA assessment robust and practically independent of noise and nonstationarities. ETA is ideally assessed during spontaneous breathing, because controlled breathing may change the sympathovagal balance (23).
Inspiration-triggered sinus arrhythmia derived by an ETA-analogous parameter that uses inspiration segments for the selection of anchor RR intervals was also a significant mortality predictor, albeit at a lower statistical significance level as compared with ETA (data not shown).
This study only included post-infarction patients age 80 years and younger. Our results might thus not translate to other patient populations.
The respiration signal was obtained using a chest belt, that is, by a method not regularly used in clinical routine. However, respiratory cycles and respiration phases can be identified in high-resolution ECG signals alone (24). Broad applications of ETA are thus technically feasible.
Although the ETA assessment method was optimized to achieve large areas under the ROC curve of mortality prediction, the dependency of the areas under the ROC curve on the values of ϕ1 and ϕ2 was rather shallow (Online Figure 2).
The primary endpoint of the study was all-cause mortality. With respect to cardiac mortality, we obtained similar results as with all-cause mortality, albeit at a lower statistical significance level. The number of sudden cardiac deaths in our study was too small to perform multivariable analyses considering multiple risk factors in addition to ETA (25,26). However, in pairwise multivariable analyses, ETA appeared to also predict sudden cardiac death independently of other risk predictors (Online Appendix, Online Figure 3).
ETA appears to be a potent risk marker in post-infarction patients. The association of ETA and mortality is independent of other established risk factors, indicating that inclusion of ETA in risk stratification models will significantly improve the predictive power of these models.
COMPETENCY IN MEDICAL KNOWLEDGE: ETA, determined by signal-averaged ECG-derived RR interval tachograms gated to the respiratory cycle, reflects the influence of vagal tone on heart rate. Attenuation of ETA, reflecting less respiratory variation in heart rate, is a strong, independent predictor of mortality in survivors of MI.
TRANSLATIONAL OUTLOOK: Future investigations should address the mechanisms relating ETA to clinical outcomes among MI survivors and explore interventions that might reduce mortality in those with impaired reflex activity.
This work was supported by the Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie (13N/7073/7), by the Deutsche Forschungsgemeinschaft (Si 1747/1-1), and by the Else Kröner-Fresenius-Stiftung. The sponsors had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The authors have reported that they have no relationships relevant to the contents of this paper to disclose. Drs. Sinnecker and Dommasch contributed equally to this work.
- Abbreviations and Acronyms
- confidence interval
- expiration-triggered sinus arrhythmia
- heart rate variability
- interquartile range
- left ventricular ejection fraction
- myocardial infarction
- receiver-operating characteristic
- respiratory sinus arrhythmia
- Received February 3, 2016.
- Revision received March 2, 2016.
- Accepted March 7, 2016.
- 2016 American College of Cardiology Foundation
- Ludwig C.
- Schumann A.Y.,
- Kantelhardt J.W.,
- Bauer A.,
- et al.
- Crawford M.H.,
- Bernstein S.J.,
- Deedwania P.C.,
- et al.
- Dommasch M.,
- Sinnecker D.,
- Barthel P.,
- et al.
- Barthel P.,
- Bauer A.,
- Müller A.,
- et al.
- Sinnecker D.,
- Dirschinger R.,
- Barthel P.,
- et al.
- Kinugawa T.,
- Dibner-Dunlap M.E.