Abstract

Aims

Acceleration and deceleration capacity (AC and DC) for beat-to-beat short-term heart rate dynamics are powerful predictors of mortality after acute myocardial infarction (AMI). We examined if AC and DC for minute-order long-term heart rate dynamics also have independent predictive value.

Methods and results

We studied 24-hr Holter electrcardiograms in 708 post-AMI patients who were followed up for up to 30 months thereafter. Acceleration capacity and DC was calculated with the time scales of T (window size defining heart rate) and s (wavelet scale) from 1 to 500 s and compared their prognostic values with conventional measures (ACconv and DCconv) that were calculated with (T,s) = [1,2 (beat)]. During the follow-up, 47 patients died. Both increased ACconv and decreased DCconv predicted mortality (C statistic, 0.792 and 0.797). Concordantly, sharp peaks of C statistics were observed at (T,s) = [2,7 (sec)] for both increased AC and decreased DC (0.762 and 0.768), but there were larger peaks of C statistics at around [30,60 (sec)] for both (0.783 and 0.796). The C statistic was greater for DC than AC at (30,60) (P = 0.0012). Deceleration capacity at (30,60) was a significant predictor even after adjusted for ACconv (P = 0.020) and DCconv (P = 0.028), but the predictive power of AC at (30,60) was no longer significant.

Conclusion

A decrease in DC for minute-order long-term heart rate dynamics is a strong predictor for post-AMI mortality and the predictive power is independent of ACconv and DCconv for beat-to-beat short-term heart rate dynamics.

What's new?

• While acceleration and deceleration capacity (AC and DC) for beat-to-beat short-term heart rate dynamics have been reported as powerful predictors of mortality after acute myocardial infarction (AMI), this study demonstrates that DC for minute-order long-term heart rate dynamics also have strong predictive power for post-AMI mortality.

• For minute-order heart rate dynamics, DC has greater predictive power than AC, indicating asymmetric predictive value between the different directions of heart rate regulation.

• DC but not AC for minute-order heart rate dynamics predicts post-AMI mortality independent of conventional beat-to-beat AC and DC.

Introduction

Phase-rectified signal averaging (PRSA) is a novel method for the analysis of heart rate variability. This method estimates individual capacity for increasing and decreasing heart rate from spontaneous heart rate fluctuations as acceleration and deceleration capacity (AC and DC).1,2 Phase-rectified signal averaging is gaining increasing attention since reduction in DC is a powerful predictor for adverse outcomes in various clinical settings, particularly in patients after acute myocardial infarction (AMI).3–5

In PRSA, increase and decrease in heart rate are detected over a given set of time scales; time scale T for determining the window size for comparing heart rates and scale s for wavelet scale. Phase-rectified signal averaging provides the averaged waveform of R-R intervals surrounding the anchor points from which heart rate accelerates or decelerates in a given time scale defined as T.2 Phase-rectified signal averaging waveform shows quasi-periodicity, decaying to both sides on the time axis. The waveforms are analysed with wavelet transforms and the magnitudes at the centre portion determined by wavelet scale s are measured as AC and DC (see Appendix and Figure 1).

Figure 1

Phase-rectified signal averaging waveforms for heart rate acceleration (A) and deceleration (B) with time scale T = 130 s in a patient. Acceleration and deceleration excursions (AE and DE) are measured as the excursion ranges of the waveform. Acceleration capacity (AC[T,s]) and deceleration capacity (DC[T,s)]) are calculated by Haar wavelet transformation with wavelet scale s from 1 to 500 s (C and D). L represents relative temporal position from anchor point (L = 0) of PRSA. 

Figure 1

Phase-rectified signal averaging waveforms for heart rate acceleration (A) and deceleration (B) with time scale T = 130 s in a patient. Acceleration and deceleration excursions (AE and DE) are measured as the excursion ranges of the waveform. Acceleration capacity (AC[T,s]) and deceleration capacity (DC[T,s)]) are calculated by Haar wavelet transformation with wavelet scale s from 1 to 500 s (C and D). L represents relative temporal position from anchor point (L = 0) of PRSA. 

Although earlier clinical studies3,4 have reported that decreased DC obtained with T = 1 beat and s = 2 beats is a powerful predictor for post-AMI mortality, Kantelhardt et al.2 have shown that AC and DC for T from 1 to 45 s and s from 1 to 12 s are also relevant to post-AMI mortality risk. These suggest that the predictive power of PRSA may not localize to beat-to-beat short-term heart rate dynamics but may instead distribute over minute-order long-term heart rate dynamics. Indeed, the predictive power of post-AMI mortality has been reported for impaired heart rate regulations that appear at longer-time scales, such as chronotropic incompetence,6,7 delayed heart rate recovery in the first minute after exercise,8,9 and altered power law behaviour at very low frequency (VLF; 0.0033–0.04 Hz) or lower.10,11 However, predictive values of PRSA for a wide range of time scales have not been investigated systematically.

In the present study, we performed PRSA of 24-hr heart rate in post-AMI patients with known clinical outcomes. We computed PRSA waveforms for T from 1 to 500 s and analysed AC and DC for s from 1 to 500 s. The hypotheses we tested were (i) PRSA for minute-order long-term heart rate dynamics has prognostic value for post-AMI mortality independently of that for beat-to-beat short-term heart rate dynamics, (ii) the predictive value of PRSA for long-term dynamics, if any, is asymmetric between AC and DC, depending on the direction of heart rate regulations.

Methods

Subjects

We used an archival database of the Enhancing Recovery in Coronary Heart Disease (ENRICHD) clinical trial.12–14 The database consisted of 24-hr Holter electrcardiograms (ECG) and clinical data at baseline and follow-up outcomes in 716 post-AMI patients. The details of this database have been reported elsewhere.13,14 Briefly, out of 716 patients, 333 were depressed and 383 were non-depressed at baseline. The depressed patients participated in the ENRICHD clinical trial and were randomly allocated to receive usual care or usual care plus cognitive behaviour therapy. However, there was no significant difference in event-free or overall survival curves between the groups.12 The non-depressed patients were otherwise eligible for ENRICHD study but had no prior episodes of major depression or current depression symptoms (Beck Depression Inventory score ≥10).

The Holter ECG was recorded within 28 days after an index AMI. The data were excluded if patients: (i) had other life-threatening illnesses; (ii) had analysable Holter ECG data <22 h or sinus rhythm <80% of total recorded beats; (iii) had atrial fibrillation, atrial flutter, or an implanted pacemaker or defibrillator; or (iv) declined to provide informed written consent. The patients were subsequently followed up for up to 30 months (median, 25 months) with the composite endpoint of all-cause mortality or recurrent non-fatal AMI, although only the mortality endpoint was used in the present study. The use of the archival dataset of the ENRICHD clinical trial in the present study was approved by the Institutional Review Board of the Washington University School of Medicine (number, 96-0849).

Data analysis

Holter recordings were scanned at the Heart Rate Variability Core laboratory at Washington University School of Medicine on a Marquette SXP Laser scanner with software version 5.8 (Marquette Electronics) using standard procedures. QRS complexes were considered as sinus rhythm only when (i) they had a narrow morphology, (ii) R-R intervals were between 300 and 2000 ms and differed ≤20% from the average of 5 preceding sinus rhythm R-R intervals, and (iii) differences in consecutive R-R intervals were ≤20% and also ≤200 m. Holter analyses were reviewed and any errors in R-wave detection and in QRS labelling were edited manually. The labelled beat file was exported to a personal computer for PRSA and other heart rate variability analysis.

Multi-scale phase-rectified signal averaging

Phase-rectified signal averaging was performed according to a previously published method.2 The details of the analysis are reported in the Appendix. Briefly, R-R interval time series were interpolated using only normal-to-normal (NN) intervals and resampled at 2 Hz. We computed acceleration and deceleration PRSA waveforms for T = 1–500 s. Wavelet transform was performed for s = 1–500 s to calculate AC(T,s) and DC(T,s) using Haar wavelet. In addition to AC and DC measures, we also measured the range of excursion of PRSA waveforms for each T value as acceleration and deceleration excursions [AE(T) and DE(T)] (Figures 1A and 1B).

For comparison with earlier observations,3,4 we also calculated AC and DC using the conventional method;2 i.e. (i) NN intervals that differed >5% of the previous NN interval were excluded from the anchors, (ii) T = 1 beat and s = 2 beats were used as time scales, and (iii) AC and DC were quantified by Haar wavelet as forumla, where forumla is the average of R-R intervals at anchors and forumla is the average of R-R intervals i beats from the anchors. The AC[1,2 beat)] and DC[1,2 (beat)] thus obtained are represented here as ACconv and DCconv.

Statistical analyses

The endpoint of this study was all-cause death. Differences between survivors and non-survivors were evaluated by the Student's t-test for quantitative data and by the χ2 test for categorical data. Differences in PRSA measures at different T values were evaluated by repeated measures analysis of variance and, when significant interaction with T was detected, the range of T values causing a significant difference were detected by generating single-degree-of-freedom contrasts with Helmert transformation. Predictive power of PRSA and other indices for all-cause mortality was evaluated by logistic regression analysis with C statistic. The difference in C statistic was evaluated by the method of DeLong et al.15 We evaluated the independence of new predictors by logistic regression models adjusted for ACconv or DCconv. Receiver-operating characteristic curve analysis was used for determining cut-off values for PRSA measures, which were used for calculating sensitivities and specificities to predict mortality. We defined a P value <0.05 as statistically significant. We used SAS version 9.3 (SAS Institute) for statistical analyses.

Results

Among 716 patients with ECG data, 8 were excluded due to <22 h of analysable ECG or <80% of sinus rhythm. Thus, the analysis was performed on ECG data in 708 patients. As reported previously,13,14 47 deaths occurred during a median follow-up period of 25 months; all of these patients had provided analysable data. Table 1 shows the characteristics of 661 survivors and 47 non-survivors.

Table 1

Patients' characteristics

Characteristics Survivor (n = 661) Non-survivor (n = 47) P 
Age (years) 58.9 ± 11.5 64.9 ± 10.7 0.0006 
Women 262 (40%) 20 (43%) 0.69 
Non-white 157 (24%) 14 (30%) 0.35 
Body mass index (kg/m229.00 ± 5.95 29.82 ± 6.36 0.39 
Systolic blood pressure (mmHg) 122.4 ± 18.6 128.8 ± 25.0 0.11 
Hypertension 126 (19%) 12 (26%) 0.28 
Diabetes mellitus 162 (25%) 33 (70%) <0.0001 
Current smoker 221 (33%) 7 (15%) 0.008 
Depression 296 (45%) 33 (70%) 0.0007 
LVEF (%) 47.2 ± 12.0 31.2 ± 13.5 <0.0001 
Killip class III–IV 26 (3.9%) 8 (17%) <0.0001 
Anterior wall AMI 201 (30%) 14 (30%) 0.95 
Creatinine (mg/dL) 1.08 ± 0.67 1.80 ± 1.51 0.002 
Prior AMI 127 (19%) 17 (36%) 0.005 
Prior coronary bypass 58 (8.8%) 14 (30%) <0.0001 
β-Blocker 522 (84%) 38 (81%) 0.96 
ACE inhibitor 305 (46%) 28 (60%) 0.074 
Coronary angioplasty <24 h 420 (64%) 15 (32%) <0.0001 
Characteristics Survivor (n = 661) Non-survivor (n = 47) P 
Age (years) 58.9 ± 11.5 64.9 ± 10.7 0.0006 
Women 262 (40%) 20 (43%) 0.69 
Non-white 157 (24%) 14 (30%) 0.35 
Body mass index (kg/m229.00 ± 5.95 29.82 ± 6.36 0.39 
Systolic blood pressure (mmHg) 122.4 ± 18.6 128.8 ± 25.0 0.11 
Hypertension 126 (19%) 12 (26%) 0.28 
Diabetes mellitus 162 (25%) 33 (70%) <0.0001 
Current smoker 221 (33%) 7 (15%) 0.008 
Depression 296 (45%) 33 (70%) 0.0007 
LVEF (%) 47.2 ± 12.0 31.2 ± 13.5 <0.0001 
Killip class III–IV 26 (3.9%) 8 (17%) <0.0001 
Anterior wall AMI 201 (30%) 14 (30%) 0.95 
Creatinine (mg/dL) 1.08 ± 0.67 1.80 ± 1.51 0.002 
Prior AMI 127 (19%) 17 (36%) 0.005 
Prior coronary bypass 58 (8.8%) 14 (30%) <0.0001 
β-Blocker 522 (84%) 38 (81%) 0.96 
ACE inhibitor 305 (46%) 28 (60%) 0.074 
Coronary angioplasty <24 h 420 (64%) 15 (32%) <0.0001 

ACE, angiotensin-converting enzyme; LVEF, left ventricular ejection fraction.

The predictive power of AC(T,s) and DC(T,s), expressed as the C statistic, is presented in Figure 2. A sharp peak was observed in short-time scale (second-order) AC and DC at (T,s) = (2,7) s (C statistic, 0.762 and 0.768). This was concordant with the strong predictive power that was observed for ACconv and DCconv calculated with (T,s) = (1,2) beat (C statistic, 0.792 and 0.797). As shown in Figure 2, a larger peak was also observed in long-time scale (minute-order) AC and DC at (T,s) = (30,60) s (C statistic, 0.784 and 0.797). Although ACconv, AC(2,7), and AC(30,60) were greater and DCconv, DC(2,7), and DC(30,60) were smaller in non-survivors than in survivors (Table 2), DC(30,60) showed a greater C statistic than AC(30,60) (P = 0.0012), indicating asymmetric predictive value between these PRSA measures for minute-order long-term heart rate dynamics. No significant difference in C statistic was observed between AC(2,7) and DC(2,7) (P = 0.072) or between ACconv and DCconv (P = 0.34).

Table 2

Phase-rectified signal averaging measures in survivors and non-survivors

Characteristics Survivor (n = 661) Non-survivor (n = 47) P 
ACconv (ms) −5.7 ± 2.4 −3.4 ± 1.6 <0.0001 
AC(2,7) (ms) −29 ± 13 −18 ± 12 <0.0001 
AC(30,60) (ms) −37 ± 16 −23 ± 11 <0.0001 
DCconv (ms) 5.4 ± 2.1 3.3 ± 1.6 <0.0001 
DC(2,7) (ms) 31 ± 14 18 ± 12 <0.0001 
DC(30,60) (ms) 37 ± 16 22 ± 11 <0.0001 
Characteristics Survivor (n = 661) Non-survivor (n = 47) P 
ACconv (ms) −5.7 ± 2.4 −3.4 ± 1.6 <0.0001 
AC(2,7) (ms) −29 ± 13 −18 ± 12 <0.0001 
AC(30,60) (ms) −37 ± 16 −23 ± 11 <0.0001 
DCconv (ms) 5.4 ± 2.1 3.3 ± 1.6 <0.0001 
DC(2,7) (ms) 31 ± 14 18 ± 12 <0.0001 
DC(30,60) (ms) 37 ± 16 22 ± 11 <0.0001 

AC, acceleration capacity; conv, conventional; DC, deceleration capacity.

Figure 2

Three-dimentional plots of mortality predictive power of AC(T,s) (A) and DC(T,s) (B) on Ts matrix plains. Colour scales from violet (lower) to red (higher) correspond to the levels of C statistic.

Figure 2

Three-dimentional plots of mortality predictive power of AC(T,s) (A) and DC(T,s) (B) on Ts matrix plains. Colour scales from violet (lower) to red (higher) correspond to the levels of C statistic.

As shown in Figure 3, AE and DE were smaller in non-survivors than in survivors (P < 0.0001) and there was no significant interaction by T value. For both AE and DE, the C statistic showed a small peak at T = 6 s and a large peak at T = 50 s (Figure 4). While the C statistic did not differ between DE(6) and AE(6) (0.757 vs. 0.750), it was greater in DE(50) than in AE(50) (0.794 vs. 0.773) (P = 0.0001), also indicating asymmetric predictive value.

Figure 3

Acceleration excursion (A) and DE (B) in survivors (closed circles) and non-survivors (open circles) for T values from 1 to 500 s. Data are presented for every 10th of T for T > 10 s.

Figure 3

Acceleration excursion (A) and DE (B) in survivors (closed circles) and non-survivors (open circles) for T values from 1 to 500 s. Data are presented for every 10th of T for T > 10 s.

Figure 4

Mortality predictive powers of AE and DE at T values from 1 to 500 s. C statistics had the maximum value at 50 s for both AE and DE (vertical line) and the C statistics at this point are significantly different between AE and DE (P = 0.0001).

Figure 4

Mortality predictive powers of AE and DE at T values from 1 to 500 s. C statistics had the maximum value at 50 s for both AE and DE (vertical line) and the C statistics at this point are significantly different between AE and DE (P = 0.0001).

Table 3 shows the independence of new PRSA measures from ACconv or DCconv in predicting post-AMI mortality. As expected, AC(2,7) and DC(2,7) had no significant independent predictive power after being adjusted for ACconv or DCconv. After the adjustment, AC(30,60) and AE(50) also has no significant predictive power. In contrast, the predictive power of DC(30,60) and DE(50) remained significant even after being adjusted for ACconv and DCconv.

Table 3

Unadjusted and adjusted odds ratios estimated by logistic regression analysis for new phase-rectified signal averaging measures

  Unadjusted
 
Adjusted for ACconv
 
Adjusted for DCconv
 
OR (95% Cl) P OR (95% Cl) P OR (95% Cl) P 
AC(2,7) 3.00 (2.02–4.48) <0.0001 1.13 (0.58–2.21) 0.717 1.12 (0.60–2.10) 0.729 
AC(30,60) 3.76 (2.42–5.84) <0.0001 1.88 (0.96–3.67) 0.065 1.78 (0.94–3.36) 0.076 
AE(50) 3.42 (2.23–5.22) <0.0001 1.52 (0.79–2.91) 0.210 1.47 (0.80–2.70) 0.221 
DC(2,7) 3.14 (2.09–4.72) <0.0001 1.23 (0.61–2.46) 0.566 1.21 (0.63–2.31) 0.563 
DC(30,60) 4.10 (2.61–6.45) <0.0001 2.29 (1.14–4.61) 0.020 2.13 (1.09–4.16) 0.028 
DE(50) 3.97 (2.54–6.22) <0.0001 2.13 (1.05–4.31) 0.036 1.98 (1.01–3.88) 0.047 
  Unadjusted
 
Adjusted for ACconv
 
Adjusted for DCconv
 
OR (95% Cl) P OR (95% Cl) P OR (95% Cl) P 
AC(2,7) 3.00 (2.02–4.48) <0.0001 1.13 (0.58–2.21) 0.717 1.12 (0.60–2.10) 0.729 
AC(30,60) 3.76 (2.42–5.84) <0.0001 1.88 (0.96–3.67) 0.065 1.78 (0.94–3.36) 0.076 
AE(50) 3.42 (2.23–5.22) <0.0001 1.52 (0.79–2.91) 0.210 1.47 (0.80–2.70) 0.221 
DC(2,7) 3.14 (2.09–4.72) <0.0001 1.23 (0.61–2.46) 0.566 1.21 (0.63–2.31) 0.563 
DC(30,60) 4.10 (2.61–6.45) <0.0001 2.29 (1.14–4.61) 0.020 2.13 (1.09–4.16) 0.028 
DE(50) 3.97 (2.54–6.22) <0.0001 2.13 (1.05–4.31) 0.036 1.98 (1.01–3.88) 0.047 

AC, acceleration capacity; AE, acceleration excursion; DC, deceleration capacity; DE, deceleration excursion; OR, odds ratio; CI, confidence interval.

Table 4 shows the cut-off values determined by ROC curve analysis for PRSA measures and their sensitivity and specificity to predict post-AMI mortality. The classification performance of decreased PRSA measures for minute-order long-term heart rate dynamics was comparable to that of decreased ACconv and DCconv for beat-to-beat short-term heart rate dynamics.

Table 4

Sensitivity and specificity of phase-rectified signal averaging measures to predict mortality

 Cut-off (ms)a Sensitivity (95% Cl) Specificity (95% Cl) 
ACconv >−4.3 87 (74–95) 68 (65–72) 
AC(2,7) >−22 81 (67–91) 73 (70–77) 
AC(30,60) >−32 85 (72–94) 63 (59–67) 
AE(50) <25 81 (67–91) 67 (63–71) 
DCconv <4.4 89 (77–97) 63 (59–67) 
DC(2,7) <22 81 (67–91) 74 (71–78) 
DC(30,60) <28 83 (69–92) 70 (66–74) 
DE(50) <23 85 (72–94) 70 (66–73) 
 Cut-off (ms)a Sensitivity (95% Cl) Specificity (95% Cl) 
ACconv >−4.3 87 (74–95) 68 (65–72) 
AC(2,7) >−22 81 (67–91) 73 (70–77) 
AC(30,60) >−32 85 (72–94) 63 (59–67) 
AE(50) <25 81 (67–91) 67 (63–71) 
DCconv <4.4 89 (77–97) 63 (59–67) 
DC(2,7) <22 81 (67–91) 74 (71–78) 
DC(30,60) <28 83 (69–92) 70 (66–74) 
DE(50) <23 85 (72–94) 70 (66–73) 

AC, acceleration capacity; AE, acceleration excursion; DC, deceleration capacity; DE, deceleration excursion; OR, odds ratio; CI, confidence interval.

aCut-off values determined by receiver-operating characteristic curve analysis as that maximized Youden index.16

Discussion

This is the first study to report the prognostic value of PRSA of heart rate dynamics for a wide range of time scales from 1 to 500 s. In a cohort of 708 post-AMI patients, we found that PRSA not only of beat-to-beat short-term but also of minute-order long-term heart rate dynamics has strong predictive power. We also found that the predictive powers of long-term PRSA measures are independent of that of short-term measures. In addition, for long-term heart rate dynamics, decrease in DC has greater predictive power than increase in AC, indicating asymmetric predictive value with the directions of heart rate regulation.

In an earlier study of post-AMI patients, Kantelhardt et al.2 have reported the prognostic relevance of AC and DC calculated over T from 1 to 45 s and s from 1 to 12 s. We also observed significant predictive power for AC and DC in these ranges of T and s (Figure 2). In the present study, however, we found the presence of two distinct peaks of C statistics, a sharp peak at (T,s) = (2,7) and a large broad peak around (T,s) = (30,60); the latter exists in the area of longer time scales beyond the region evaluated by Kantelhardt et al. To confirm this finding, we introduced new measures (AE and DE) that simply reflect the excursion range of the PRSA waveform defined by a single time scale of T. Acceleration excursion and DE also showed two peaks in the C statistic at T = 6 and 50 s (Figure 4). These suggest the presence of two distinct temporal regions, short-term (several seconds) and long-term (around 1 min), of heart rate regulations that are associated with post-AMI mortality.

Although the exact mechanisms for the presence of two distinct peaks are not clear, the time scale of measures seems an important clue. The studies of transfer function analysis of the autonomic neural regulations have indicated that beat-to-beat rapid heart rate fluctuation >0.15 Hz (9 cpm) is not mediated by the sympathetic nerves and is mediated solely by the vagus.17,18 Thus, the sharp peak in the area of second-order short time scales may reflect mortality risk associated with cardiac vagal dysfunction. Whereas, for the large broad peak in minute-order long time scales, we need to consider numerous mechanisms, which include both neural and humoral heart rate regulations. A study in conscious unanaesthetized dogs, Akselrod et al.19 demonstrated that angiotensin-converting enzyme inhibitor causes 2- to 4.5-fold increase in the power of heart rate variability at around 0.04 Hz (corresponding period of 25 s), suggesting that the activated renin–angiotensin system may be a possible cause of decreased long-term heart rate fluctuation and the mediator of its adverse outcome. Because the present data were obtained by ambulatory ECG recordings, heart rate response characteristics to daily activities may be reflected in long-term PRSA measures. Delayed heart rate recovery in the first minute after exercise has been known to predict mortality in cardiac patients8,9 including those after AMI20 independently of myocardial ischaemia or exercise capacity.6,7 Decrease in long-term DC and DE may be another representation of this phenomenon. Finally, altered power-law behaviour (increased spectral exponent β) of heart rate fluctuation at VLF or lower frequencies has been known as increased risk for post-AMI mortality.10,11 Although the long-term PRSA measures showed only modest correlations with β (data not shown), they may provide a useful probe into the pathophysiological meanings of heart rate variability at VLF or lower frequencies.

The ability to analyse the capacities separately for accelerating and decelerating heart rate is a unique feature of PRSA. In the present study, we found asymmetric predictive value between deceleration and acceleration measures for long-time scales with greater value for deceleration. No such difference was found for short time scale measures, including ACconv or DCconv. Earlier studies have reported that the heart rate recovery is mediated by parasympathetic reactivation during post-exercise period21 and that its predictive power is independent of an inability to increase heart rate during exercise (chronotropic incompetence).7 Decreased acceleration and deceleration measures for long time scales may also reflect deleterious heart rate regulation with different mechanisms.

In the clinical point of view, the most important finding of the present study is that long-term PRSA measures have predictive power not only comparable with but also independent of the predictive power of conventional short-term PRSA measures. We observed that the sensitivity and specificity of long-term PRSA measures to predict mortality were comparable to those of ACconv and DCconv and that their predictive power was significant even after being adjusted for ACconv and DCconv. These indicate that combined use of short-term and long-term PRSA measures may improve risk stratification for post-AMI mortality.

Limitations

This study has several limitations. First, in the ENRICHD archival database, 329 (46.5%) out of 709 patients were at least mildly depressed and about half (n = 146) of the depressed patients had cognitive behavioural therapy in addition to usual AMI care. However, the frequency of patients with depression is comparable to the reported prevalence of depression in general post-AMI populations (45–47%).22,23 Moreover, we observed that DC(30, 60) and DE(50) were independent predictors of mortality even when analyzed after excluding depressed patients (data not shown). Also, because the cognitive behavioral therapy had no significant effect on mortality in the ENRICHD clinical trial,12 we did not include the effect of treatment in the present study. Second, the data of ENRICHD database were collected between October 1997 and January 2000.12 The frequency of primary percutaneous coronary intervention has been increasing. Consequently, the frequency of post-AMI patients with preserved LVEF has increased and most of post-AMI deaths arise in those patients recently.24 Thus, the present observations may not be applicable directly to the current post-AMI populations. Third, we did not consider sudden cardiac death as a separate endpoint, because the causes of cardiac death were not identified in the ENRICHD study database. Furthermore, the number of deaths, particularly of non-cardiac deaths, was small (n = 13), so we could not detect a possible association between PRSA measures and causes of death.

Conclusion

We conclude that decreases in DC and DE for minute-order long-term heart rate dynamics are strong predictors for post-AMI mortality and the predictive power is independent of AC and DC for beat-to-beat short-term heart rate dynamics.

Funding

This work was supported by the grant of the Knowledge Hub of Aichi, Japan [the Priority Research Project, P3-G1-S1-2b (J. H.)]; the Japan Society for the Promotion of Science, Japan [Grant-in-Aid for Scientific Research (C) 20590832, 23591055 (J.H.)]; the Ministry of Health, Labor and Welfare, Japan [Research Grant for Nervous and Mental Disorders 20B-7, 23-2 (J.H.)]; and the National Heart, Lung, and Blood Institute, National Institutes of Health [Grant no. 2 RO1 HL058946].

Conflict of interest: none declared.

Appendix

Multi-scale phase-rectified signal averaging

The analysis was performed in 24 h R-R interval time series after labelling all QRS complexes for normal (sinus rhythm), ectopic beat, and noise/artefact. To obtain functions of true time, we used sec for the unit of time scales. R-R interval time series were interpolated using only normal-to-normal (NN) intervals with a step function and resampled at 2 Hz.

The anchors for computing acceleration capacity (AC) at time scale of T were defined as all forumla satisfying 

formula

where forumla is the interpolated R-R interval at time t. Similarly, the anchors for decelerating capacity (DC) were defined as all forumla satisfying: 

formula

Then, segments of interpolated NN intervals from L s before to L− 1 s after the all anchor points were extracted with allowing overlapping between segments. Anchor points close to the beginning or end of the time series, where no full surrounding of length 2L were available, were excluded from the analysis. All segments thus obtained were aligned at the anchors and averaged over them to obtain the PRSA waveform as 

formula

where M is the number of obtained segments, km is the position of anchor point in the mth segment (m = 1, … , M), and L is the half length of segment (in this study L = 512 s). The wavelet transform of forumla is 

formula

where s is the wavelet scale, p is time position in PRSA waveform, and w(t) is the mother wavelet. According to the earlier study by Kantelhardt et al.2 we employed Haar wavelet: 

formula

as w(t), and AC(T,s) and DC(T,s) were calculated as forumla that reflects the wave magnitude of forumla at the centre portion (l = –s, … , 0, … , s).

We performed these phase-rectified signal averaging analyses on a Microsoft Windows-based personal computer (Intel Core™ i7 870 processor, 2.93 GHz) with a FORTRAN program developed for this purpose. The computation took 0.78 s per one time scale T per one 24 h electrocardiogram recording.

References

1
Bauer
A
Kantelhardt
JW
Bunde
A
Barthel
P
Schneider
R
Malik
M
, et al.  . 
Phase-rectified signal averaging detects quasi-periodicities in non-stationary data
Phys A
 , 
2006
, vol. 
364
 (pg. 
423
-
34
)
2
Kantelhardt
JW
Bauer
A
Schumann
AY
Barthel
P
Schneider
R
Malik
M
, et al.  . 
Phase-rectified signal averaging for the detection of quasi-periodicities and the prediction of cardiovascular risk
CHAOS
 , 
2007
, vol. 
17
 pg. 
015112
 
3
Bauer
A
Kantelhardt
JW
Barthel
P
Schneider
R
Makikallio
T
Ulm
K
, et al.  . 
Deceleration capacity of heart rate as a predictor of mortality after myocardial infarction: cohort study
Lancet
 , 
2006
, vol. 
367
 (pg. 
1674
-
81
)
4
Bauer
A
Barthel
P
Schneider
R
Ulm
K
Muller
A
Joeinig
A
, et al.  . 
Improved Stratification of Autonomic Regulation for risk prediction in post-infarction patients with preserved left ventricular function (ISAR-Risk)
Eur Heart J
 , 
2009
, vol. 
30
 (pg. 
576
-
83
)
5
Barthel
P
Bauer
A
Muller
A
Junk
N
Huster
KM
Ulm
K
, et al.  . 
Reflex and tonic autonomic markers for risk stratification in patients with type 2 diabetes surviving acute myocardial infarction
Diabetes Care
 , 
2011
, vol. 
34
 (pg. 
1833
-
7
)
6
Shetler
K
Marcus
R
Froelicher
VF
Vora
S
Kalisetti
D
Prakash
M
, et al.  . 
Heart rate recovery: validation and methodologic issues
J Am Coll Cardiol
 , 
2001
, vol. 
38
 (pg. 
1980
-
7
)
7
Vivekananthan
DP
Blackstone
EH
Pothier
CE
Lauer
MS
Heart rate recovery after exercise is a predictor of mortality, independent of the angiographic severity of coronary disease
J Am Coll Cardiol
 , 
2003
, vol. 
42
 (pg. 
831
-
8
)
8
Cole
CR
Blackstone
EH
Pashkow
FJ
Snader
CE
Lauer
MS
Heart-rate recovery immediately after exercise as a predictor of mortality
N Eng J Med
 , 
1999
, vol. 
341
 (pg. 
1351
-
7
)
9
Tang
YD
Dewland
TA
Wencker
D
Katz
SD
Post-exercise heart rate recovery independently predicts mortality risk in patients with chronic heart failure
J Card Fail
 , 
2009
, vol. 
15
 (pg. 
850
-
5
)
10
Bigger
JT
Jr
Steinman
RC
Rolnitzky
LM
Fleiss
JL
Albrecht
P
Cohen
RJ
Power law behavior of RR-interval variability in healthy middle-aged persons, patients with recent acute myocardial infarction, and patients with heart transplants
Circulation
 , 
1996
, vol. 
93
 (pg. 
2142
-
51
)
11
Huikuri
HV
Mäkikallio
TH
Airaksinen
KEJ
Seppänen
T
Puukka
P
Räihä
IJ
, et al.  . 
Power-law relationship of heart rate variability as a predictor of mortality in the elderly
Circulation
 , 
1998
, vol. 
97
 (pg. 
2031
-
6
)
12
Berkman
LF
Blumenthal
J
Burg
M
Carney
RM
Catellier
D
Cowan
MJ
, et al.  . 
Effects of treating depression and low perceived social support on clinical events after myocardial infarction: the Enhancing Recovery in Coronary Heart Disease Patients (ENRICHD) Randomized Trial
JAMA
 , 
2003
, vol. 
289
 (pg. 
3106
-
16
)
13
Carney
RM
Blumenthal
JA
Catellier
D
Freedland
KE
Berkman
LF
Watkins
LL
, et al.  . 
Depression as a risk factor for mortality after acute myocardial infarction
Am J Cardiol
 , 
2003
, vol. 
92
 (pg. 
1277
-
81
)
14
Carney
RM
Steinmeyer
B
Freedland
KE
Blumenthal
JA
Stein
PK
Steinhoff
WA
, et al.  . 
Nighttime heart rate and survival in depressed patients post acute myocardial infarction
Psychosom Med
 , 
2008
, vol. 
70
 (pg. 
757
-
63
)
15
DeLong
ER
DeLong
DM
Clarke-Pearson
DL
Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach
Biometrics
 , 
1988
, vol. 
44
 (pg. 
837
-
45
)
16
Youden
WJ
Index for rating diagnostic tests
Cancer
 , 
1950
, vol. 
3
 (pg. 
32
-
5
)
17
Pomeranz
B
Macaulay
RJB
Caudill
MA
Kutz
I
Adam
D
Gordon
D
, et al.  . 
Assessment of autonomic function in humans by heart rate spectral analysis
Am J Physiol
 , 
1985
, vol. 
248
 (pg. 
H151
-
3
)
18
Berger
RD
Saul
JP
Cohen
RJ
Transfer function analysis of autonomic regulation. I: canine atrial rate response
Am J Physiol
 , 
1989
, vol. 
256
 (pg. 
H142
-
52
)
19
Akselrod
S
Gordon
D
Ubel
FA
Shannon
DC
Barger
AC
Cohen
RJ
Power spectrum analysis of heart rate fluctuation: a quantitative probe of beat-to-beat cardiovascular control
Science
 , 
1981
, vol. 
213
 (pg. 
220
-
2
)
20
Nissinen
SI
Makikallio
TH
Seppanen
T
Tapanainen
JM
Salo
M
Tulppo
MP
, et al.  . 
Heart rate recovery after exercise as a predictor of mortality among survivors of acute myocardial infarction
Am J Cardiol
 , 
2003
, vol. 
91
 (pg. 
711
-
4
)
21
Goldberger
JJ
Le
FK
Lahiri
M
Kannankeril
PJ
Ng
J
Kadish
AH
Assessment of parasympathetic reactivation after exercise
Am J Physiol Heart Circ Physiol
 , 
2006
, vol. 
290
 (pg. 
H2446
-
2452
)
22
Schleifer
SJ
Macari-Hinson
MM
Coyle
DA
Slater
WR
Kahn
M
Gorlin
R
, et al.  . 
The nature and course of depression following myocardial infarction
Arch Intern Med
 , 
1989
, vol. 
149
 (pg. 
1785
-
9
)
23
Steeds
RP
Bickerton
D
Smith
MJ
Muthusamy
R
Assessment of depression following acute myocardial infarction using the Beck depression inventory
Heart
 , 
2004
, vol. 
90
 (pg. 
217
-
8
)
24
Barthel
P
Schneider
R
Bauer
A
Ulm
K
Schmitt
C
Schomig
A
, et al.  . 
Risk stratification after acute myocardial infarction by heart rate turbulence
Circulation
 , 
2003
, vol. 
108
 (pg. 
1221
-
6
)