Bedside autonomic risk stratification after myocardial infarction by means of short-term deceleration capacity of heart rate

Abstract

Aims

Twenty-four-hour deceleration capacity (DC_24h) of heart rate is a strong predictor of mortality after myocardial infarction (MI). Assessment of DC from short-term recordings (DC_st) would be of practical use in everyday clinical practice but its predictive value is unknown. Here, we test the usefulness of DC_st for autonomic bedside risk stratification after MI.

Methods and results

We included 908 patients after acute MI enrolled in Munich and 478 patients with acute (n = 232) and chronic MI (n = 246) enrolled in Tuebingen, both in Germany. We assessed DC_st from high-resolution resting electrocardiogram (ECG) recordings (<30 min) performed under standardized conditions in supine position. In the Munich cohort, we also assessed DC_24h from 24-h Holter recordings. Deceleration capacity was dichotomized at the established cut-off value of ≤ 2.5 ms. Primary endpoint was 3-year mortality. Secondary endpoint was 3-year cardiovascular mortality. In addition to DC, multivariable analyses included the Global Registry of Acute Coronary Events score >140 and left ventricular ejection fraction ≤ 35%. During follow-up, 48 (5.3%) and 48 (10.0%) patients died in the Munich and Tuebingen cohorts, respectively. On multivariable analyses, DC_st ≤ 2.5 ms was the strongest predictor of mortality, yielding hazard ratios of 5.04 (2.68–9.49; P < 0.001) and 3.19 (1.70–6.02; P < 0.001) in the Munich and Tuebingen cohorts, respectively. Deceleration capacity assessed from short-term recordings ≤ 2.5 ms was also an independent predictor of cardiovascular mortality in both cohorts. Implementation of DC_st ≤ 2.5 ms into the multivariable models led to a significant increase of C-statistics and integrated discrimination improvement score.

Conclusion

Deceleration capacity assessed from short-term recordings is a strong and independent predictor of mortality and cardiovascular mortality after MI, which is complementary to existing risk stratification strategies.

Deceleration capacity of heart rate , Heart rate variability , Myocardial infarction , Risk stratification , Short-term electrocardiogram recordings

What’s new?

• Twenty-four-hour deceleration capacity (DC_24h) of heart rate is a strong predictor of mortality after myocardial infarction. However, the need of 24-h recordings is considered as major drawback. Here, we tested the usefulness of deceleration capacity assessed from short-term recordings (DC_st) in two large cohorts.

• In both cohorts, DC_st was the strongest predictor of 3-year mortality.

• Deceleration capacity assessed from short-term recordings was independent from established risk predictors including the Global Registry of Acute Coronary Events score and left ventricular ejection fraction (LVEF).

• The predictive value of DC assessed from short-term recordings was comparable to that of DC assessed from 24-h recordings.

• Deceleration capacity assessed from short-term recordings is a useful tool for bedside risk stratification that complements LVEF and clinical factors.

Introduction

Risk stratification after myocardial infarction (MI) is of crucial importance to guide preventive strategies. Reduced left ventricular ejection fraction (LVEF ≤ 35%) remains the current gold standard risk predictor.¹ However, it lacks both sensitivity and specificity.² In fact, the majority of patients who die after MI have an LVEF of 36% or greater,^2–4 highlighting the need for refined risk stratification strategies.

Deceleration capacity (DC) of heart rate is an advanced marker of heart rate variability (HRV) that yields strong and independent prognostic information in post-infarction patients.⁵ Deceleration capacity is an integral measure of all deceleration-related oscillations of heart rate over 24 h, including regulations in the ultra-low, very low, low, and high frequency bands. In previous studies, the prognostic value of impaired DC in predicting late mortality after MI exceeded that of abnormal standard measures of HRV and even reduced LVEF.⁵

However, in the light of increasingly shorter hospital stays the need of a full 24-h Holter recording for DC assessment is considered as a major drawback. For practical reasons, bedside autonomic risk stratification that could be performed in a comparable time frame as the echocardiographic assessment of left ventricular function would be greatly appreciated and would help implementing autonomic risk stratification in everyday clinical practice. In addition, compared to 24-h recordings, short-term ECGs can be performed under standardized conditions eliminating noise, artifacts, and non-stationarities.

We therefore undertook this study to test the usefulness of DC obtained from short-term ECG-recordings (DC_st) in predicting death after MI.

Methods

Study design and study populations

The prognostic value of DC_st was evaluated in two post-infarction cohorts, the Munich and Tuebingen cohort.

The study design of the Munich cohort has been described elsewhere⁶ and is illustrated in Figure 1A. The study included 908 survivors of acute MI (left column of Table 1). Patients were enrolled between May 2000 and March 2005 at two university centres in Munich, Germany, the German Heart Centre and the Klinikum Rechts der Isar. Eligible patients had survived an acute MI (≤40 days), were aged ≤80 years, had sinus rhythm, and did not meet the criteria for prophylactic ICD implantation prior hospital discharge.

Table 1

Patients’, clinical, and treatment characteristics in the Munich and Tuebingen cohorts

Characteristic	Munich cohort	Tuebingen cohort
Study characteristics
Number of patients, n	908	478
Median follow-up (IQR), months	36.0 (0)	27.3 (11.3)
Total deaths, n	48	48
Cardiovascular deaths, n	27	24
Patients’ and clinical characteristics
Age (IQR), years	61 (17)	66 (17)
Acute MI, n (%)	908 (100)	232 (48.6)
Acute MI localization (AW, LW, PW) (%)	42/11/47	46/19/35
Time from index MI, days (IQR)	7 (4)	154 (3071)
Females, n (%)	174 (19.2)	109 (22.8)
Diabetes mellitus, n (%)	179 (19.7)	210 (43.9)
LVEF, % (IQR)	53 (15)	50 (15)
GRACE, Score (IQR)	110 (32)	134 (43)
eGFR, mL/min/1.73 m2 (IQR)	69 (25)	79 (32)
SP on admission, mmHg (IQR)	130 (35)	130 (27)
Cardiogenic shock on admission, n (%)	10 (1.1)	11 (2.3)
Previous MI, n (%)	86 (9.5)	277 (58)
Multivessel CAD, n (%)	563 (62.0)	387 (81)
MHR, bpm (IQR)	62 (13)	65 (16)
DC24h, ms (IQR)	5.2 (3.5)	n/a
DCst, ms (IQR)	4.7 (4.0)	3.9 (4.0)
Treatment
PCI, n (%)	848 (93.4)	439 (92)
CABG surgery, n (%)	32 (3.5)	48 (10)
Beta-blockers, n (%)	864 (95.1)	440 (92.1)
Statins, n (%)	843 (92.8)	432 (90.4)
ACE Inhibitor	850 (93.6)	366 (76.6)

ACE, angiotensin converting enzyme; AW, anterior wall; CAD, coronary artery disease; DC_st, deceleration capacity assessed from short-term recordings; DC_24h, deceleration capacity assessed from 24-h Holter electrocardiograms; CABG, coronary artery bypass graft; eGFR, estimated glomerular filtration rate; GRACE, Global Registry of Acute Coronary Events; IQR, interquartile range; LVEF, left-ventricular ejection fraction; LW, lateral wall; MHR, mean heart rate from short-term recordings; MI, myocardial infarction; PCI, percutaneous coronary intervention; PW, posterior wall; SP, systolic pressure.

Figure 1

Consort flow-diagrams for the training and validation cohorts.

Patients of the Tuebingen cohort were prospectively enrolled between September 2010 and February 2014 at the university hospital of Tuebingen, Germany (Figure 1B). Inclusion criteria consisted of history of previous MI, sinus rhythm and age ≤80 years [n = 478, median age 66 (IQR 17), 109 females (22.8%); right column of Table 1]. Compared with the Munich cohort, the study was not limited to patients with acute MI. In 246 of the patients (51.5%), the index MI was older than 40 days. The ethics committees of Munich and Tuebingen approved both studies.

Procedures

In both cohorts, high-resolution digital ECGs (Munich cohort: TMS, Porti System version 1; 1600 Hz; 30 min; Tuebingen cohort: TMS, Porti System version 2; 2048 Hz; 20 min) were performed in Frank leads configuration. The recordings were done in the morning hours under standardized conditions in supine and resting position. Patients were spontaneously breathing. In the Munich cohort, the median time from the index MI to the ECG recording was 7 (IQR 4) days. In Tuebingen cohort, which also included patients with chronic MI, the median time from the index MI to the ECG recording was 154 (IQR 3071) days.

In the Munich cohort, additionally, 24-h Holter recordings (Oxford Excel Holter system, Oxford instruments; Pathfinder700, Reynolds Medical; and Mortara Holter system, Mortara Instrument) were performed in 861 patients within the second week after MI.

Assessment of deceleration capacity

Deceleration capacity assessed from short-term recordings and DC_24h were calculated using the same algorithms.^5,7 The exact technology of DC assessment has been described elsewhere. Very briefly, computation of DC is based on the transformation of the sequence of RR intervals into a new time series by phase-rectified signal averaging (PRSA). In a first step, RR intervals that are longer than their respective preceding RR-intervals are identified (so-called anchors). In a second step, segments around anchors are averaged to obtain the so-called PRSA-signal. The PRSA-signal can be considered as a condensed version of the original RR-interval time series, including all periodic components of HRV related to decelerations. The central part of the PRSA-signal is quantified by wavelet-analysis to obtain the numerical measure of DC (Figure 2). Thus, DC is an integral measure of all deceleration-related oscillations that take place during the observational period. As previously described, patients with DC ≤ 2.5 ms were classified as high-risk patients.

Figure 2

Typical phase-rectified signals from (A) a 57-year-old post-MI woman who survived the 3-year follow-up period and (B) a 48-year-old man who suddenly died 9 weeks after the index MI. In the surviving patient the amplitude of the phase-rectified signal is significantly higher compared to the non-surviving patient.

Assessment of other risk predictors

Left ventricular ejection fraction was assessed by echocardiography or angiography. The Global Registry of Acute Coronary Events (GRACE) score was calculated as previously described⁸ and included following variables: age, heart rate, systolic blood pressure, creatinine level, Killip class, cardiac arrest at admission, ST-segment deviation, and elevated cardiac enzyme levels. The GRACE score was dichotomized at the established cut-off value of >140. Estimated glomerular filtration rate (eGFR) was estimated using the modified diet in renal disease formula and dichotomized at the cut-off value of <60 mL/min/1.73 m². To compare DC_st with standard measures of HRV,⁹ all measures in time and frequency domain as well as non-linear indices were calculated according to the recommendations provided by the task force.⁹ For each domain, we identified the measure that showed the strongest association with the primary endpoint in the Munich cohort. As no established cut-off values for short-term HRV measures exist, we used maximally selected rank statistics to determine the optimum cut-off values for all markers, which were then also used in the Tuebingen cohort.

Study endpoints and follow-up

In both cohorts, primary endpoint was 3-year all-cause mortality. Secondary endpoint was 3-year cardiovascular mortality. Patients were followed-up at regular intervals, either in the outpatient clinic or by telephone calls, with median follow-up periods of 36 and 27 months, respectively.

Statistical analysis

Continuous variables are presented as medians with IQRs and were compared using the Wilcoxon rank-sum test. Categorical variables are expressed as percentages and were analysed using the χ² and Fisher’s exact tests. We quantified receiver-operator characteristic (ROC) curves by the integrals of the curves (AUC, area under the curve), plotting the dependency of specificity on sensitivity. To test the difference between two ROC curves, we used bootstrapping, based on the creation of pseudo-replicate datasets by random resampling of the dataset N times for error estimation (N = 2000 in this study). We estimated survival curves using the Kaplan–Meier method and compared them by means of the log-rank test. Sensitivities and specificities were extracted from the survival curves. Multivariable analyses were implemented by the adaptation of Cox regression models. Subgroup analyses were performed by means of the regression technique. Results are presented as hazard ratios with 95% confidence interval (CI). To test the incremental prognostic value of DC_st on top of the GRACE score and LVEF, we implemented C-statistics, integrated discrimination improvement (IDI) score, and continuous net reclassification improvement analysis (NRI). To test the difference between C-statistics bootstrapping was employed. Agreement between DC_st with DC_24h was assessed by the method described by Bland and Altman. Differences were considered statistically significant when the two-sided P-value was less than 0.05. All statistical analyses were performed using CRAN R, version 3.2.3.

Results

Table 1 shows baseline, clinical, follow-up, and treatment characteristics in the Munich (n = 908) and Tuebingen cohorts (n = 478). In the Munich cohort, 48 patients died (27 cardiovascular deaths) during a median follow-up time of 36.0 months. In the Tuebingen cohort, 48 patients died (24 cardiovascular deaths) over a median interval of 27.3 months. Supplementary material online, Table S1 shows association of clinical parameters with DC_st ≤ 2.5 ms.

Figure 2 shows representative PRSA-signals from short-term recordings in two post-MI patients of the Munich cohort. Panel A shows a patient who survived the 3-year follow-up period. Panel B shows a patient who suddenly died 9 weeks after the index MI. In the surviving patient, the amplitude of the PRSA-signal is significantly higher as compared to the non-surviving patient.

Table 2 shows the association of risk factors with 3-year all-cause mortality. In both cohorts, non-surviving patients were older, had a lower LVEF and had higher GRACE scores. Non-surviving patients in the Munich cohort but not in the Tuebingen cohort had higher incidence of diabetes mellitus. In both cohorts, non-surviving patients had substantially lower DC_st compared to surviving patients (2.0 ± 3.6 vs. 4.8 ± 3.9 ms in the Munich cohort and 1.4 ± 3.5 vs. 4.1 ± 4.0 ms in the Tuebingen cohort, P < 0.001 for both).

Table 2

Statistical association of clinical and HRV-markers with 3-year mortality in the Munich and Tuebingen cohorts

Characteristic	Munich cohort			Tuebingen cohort
	Survivors	Non-survivors	P-value	Survivors	Non-survivors	P-value
Number of patients, n	860	48		430	48
Age, years (IQR)	60 (17)	70 (9)	<0.001	65 (18)	72 (7)	<0.001
Females, n (%)	164 (19.1)	10 (20.1)	0.910	98 (22.8)	11 (22.9)	0.999
Acute MI, n (%)	860 (100)	48 (100)	n/a	209 (48.6)	23 (47.9)	0.999
Diabetes mellitus, n (%)	156 (18.1)	23 (47.9)	<0.001	184 (42.8)	26 (54.2)	0.176
Median LVEF, % (IQR)	53 (16)	44 (23)	<0.001	50 (15)	40 (20)	<0.001
GRACE, Score (IQR)	109 (32)	132 (24)	<0.001	132 (42)	163 (46)	<0.001
MHR, bpm (IQR)	62 (13)	69 (19)	0.033	65 (15)	70 (18)	0.003
DC24h, ms (IQR)	5.3 (3.4)	2.8 (2.1)	<0.001	n/a	n/a	n/a
DCst, ms (IQR)	4.8 (3.9)	2.0 (3.6)	<0.001	4.1 (4.0)	1.4 (3.5)	<0.001

DC_st, deceleration capacity assessed from short-term recordings; DC_24h, deceleration capacity assessed from 24-h Holter electrocardiograms; HRV, heart rate variability; GRACE, Global Registry of Acute Coronary Events; IQR, interquartile range; LVEF, left-ventricular ejection fraction; MHR, mean heart rate from short-term recordings; MI, myocardial infarction n/a, not available.

Figure 3A and B depict cumulative mortality rates of patients stratified by DC_st ≤ 2.5 ms in the Munich cohort and Tuebingen cohort, respectively. In the Munich cohort, the 182 patients with DC_st ≤ 2.5 ms had a 3-year mortality rate of 16.5% compared to 3-year mortality rate of 2.5% in the 726 patients with DC_st > 2.5 ms (Figure 3A), corresponding to a sensitivity of 63% and specificity of 82%. Also in the Tuebingen cohort, risk stratification by DC_st ≤ 2.5 ms led to highly significant separation of low and high-risk patients. The 152 patients with DC_st ≤ 2.5 ms have a 3-year mortality rate of 27.5% compared to a 3-year mortality rate of 6.3% in the 326 patients with DC_st > 2.5 ms (Figure 3B), corresponding to a sensitivity of 67% and specificity of 73%.

Figure 3

Cumulative 3-year mortality rates in patients stratified by DC_st ≤ 2.5 ms in the Munich (A) and Tuebingen (B) cohorts, respectively.

In both cohorts, DC_st was the strongest predictor of mortality in uni- and multivariable Cox regression analyses (Table 3 and Supplementary material online, Table S2). Its predictive value was independent from and incremental to that of the GRACE score and LVEF, as well as that of HRV-index (HRVI) ≤ 26 units, the normalized power in the low frequency range (LFn) ≤ 0.42, and the multiscale sample entropy index ≤ 4.47, which were the strongest HRV measures in the Munich cohort (Table 3 and Supplementary material online, Table S2). Implementing DC_st into the multivariable model of GRACE score and LVEF led to a significant increase of C-statistics from 68.23% (95% CI 61.16–75.71%) to 76.80% (69.11–83.86%; P = 0.004 for difference) in the Munich cohort and from 70.16% (62.16–77.55%) to 76.83% (69.98–82.98%; P = 0.017 for difference) in the Tuebingen cohort. Integrated discrimination improvement increased by 0.045 (0.018–0.092, P < 0.001) in the Munich cohort and by 0.037 (0.006–0.110, P = 0.007) in the Tuebingen cohort, whereas continuous NRI improved by 0.449 (0.314–0.577, P < 0.001) in the Munich cohort and by 0.389 (0.214–0.558, P < 0.001) in the Tuebingen cohort. Deceleration capacity assessed from short-term recordings was also a strong and independent predictor of cardiovascular mortality in both cohorts (Table 3). Adjusting for other parameters, including treatment with beta-blockers and statins did not have a significant impact on the predictive power of DC_st.

Table 3

Univariable and multivariable cox regression analyses of the association of risk variables with 3-year total and cardiovascular mortality in the Munich and Tuebingen cohorts

Munich cohort
All-cause mortality
Risk variable	Univariable cox regression		Multivariable cox regression
	Hazard ratio (95% CI)	P-value	Hazard ratio (95% CI)	P-value
GRACE score > 140	4.35 (2.36–8.01)	<0.001	1.91 (0.98–3.73)	0.059
LVEF ≤ 35%	5.27 (2.89–9.61)	<0.001	2.75 (1.43–5.27)	0.002
DCst ≤ 2.5 ms	7.18 (4.00–12.87)	<0.001	5.04 (2.68–9.49)	<0.001
Cardiovascular mortality
	Hazard ratio (95% CI)	P-value	Hazard ratio (95% CI)	P-value
GRACE score > 140	4.00 (1.75–9.14)	0.001	1.72 (0.69–4.27)	0.245
LVEF ≤ 35%	6.13 (2.81–13.38)	<0.001	3.47 (1.48–8.12)	0.004
DCst ≤ 2.5 ms	6.21 (2.88–13.37)	<0.001	4.18 (1.81–9.68)	<0.001
Tuebingen cohort
All-cause mortality
	Hazard ratio (95% CI)	P-value	Hazard ratio (95% CI)	P-value
GRACE score >140	3.48 (1.89–6.42)	<0.001	2.29 (1.20–4.37)	0.012
LVEF ≤ 35%	4.16 (2.35–7.37)	<0.001	2.07 (1.11–3.86)	0.023
DCst ≤ 2.5 ms	4.57 (2.51–8.33)	<0.001	3.19 (1.70–6.02)	<0.001
Cardiovascular mortality
	Hazard ratio (95% CI)	P-value	Hazard ratio (95% CI)	P-value
GRACE score > 140	3.79 (1.57–9.16)	0.003	2.49 (0.98–6.35)	0.056
LVEF ≤ 35%	4.44 (1.99–9.93)	<0.001	2.31 (0.95–5.60)	0.065
DCst ≤ 2.5 ms	3.79 (1.66–8.67)	0.002	2.51 (1.04–6.03)	0.040

DC_st, deceleration capacity assessed from short-term recordings; GRACE, Global Registry of Acute Coronary Events; LVEF, left-ventricular ejection fraction.

In the Munich cohort, we also compared DC from short-term and 24-h recordings. DC_24h was slightly higher than DC_st (mean difference 0.62 ± 5.97, P = 0.002, Supplementary material online, Figure S2). The Pearson’s correlation coefficient between the two measurements was 0.75 (0.72–0.78, P < 0.001). Univariable analysis revealed that both measurements yielded comparable AUC (73.37%, [64.28–81.67%] for DC_st and 76.24%, [67.75–83.85%] for DC_24h; P = 0.386 for the difference; Supplementary material online, Figure S1). However, in multivariable analysis including both DC_st and DC_24h, only DC_st was significantly associated with prediction of 3-year mortality (see Supplementary material online, Table S3).

Subgroup analyses revealed that DC_st was stronger predictor of mortality among younger patients (HR 10.89 [4.62–25.70], P < 0.001 for patients < 70 years and 2.78 [1.25–6.19], P = 0.012 among patients ≥ 70 years; P = 0.023 for the difference between the two subgroups) in the Munich cohort but not in the Tuebingen cohort (Figure 4). Although in the Tuebingen cohort there was a trend for better predictive power of DC_st among non-diabetics and patients with eGFR ≥ 60 mL/min/1.73 m², this did not reach statistical significance (Figure 4). Moreover, DC_st ≤ 2.5 ms was a strong predictor of mortality in both acute and chronic MI [HR 5.36 (2.27–12.65), P < 0.001 for acute MI vs. 4.06, (1.75–9.40), P = 0.001 for chronic MI; P = 0.211 for the difference between the two subgroups].

Figure 4

Hazard ratios of abnormal DC_st ≤ 2.5 ms in different subgroups of patients in the Munich and Tuebingen cohorts. GRACE, Global Registry of Acute Coronary Events; eGFR, estimated glomerular filtration rate; LVEF, left-ventricular ejection fraction.

Discussion

In the present study, we assessed DC from short-term ECG recordings under standardized conditions and tested its predictive value in two large cohorts of post-MI patients. Our findings demonstrate that DC_st is a strong predictor of 3-year mortality, which is independent from and incremental to that of established risk markers. The findings of our study therefore indicate that DC_st is a useful tool for bedside risk stratification that complements LVEF and clinical factors.

Heart rate variability refers to the variation over time of consecutive heart beat intervals and predominantly reflects the overall state of the autonomic nervous system. Heart rate variability has been conventionally analysed by means of time- and frequency- domain methods, as well as non-linear methods such as fractal scaling exponents. Reduced HRV has been proven to be a strong and independent predictor of adverse outcome in the general population¹⁰ and in patients with heart disease. The strong prognostic value of reduced HRV in post-MI patients has been documented in many studies. The multicenter post-infarction project (MPIP) showed that reduced HRV quantified by means of 24-h standard deviation of normal-to-normal RR intervals (SDNN) was a strong and independent predictor of 4-year mortality after acute MI.¹¹ These results have been validated in several prospective cohorts.^12,13 Other studies, including a reanalysis of MPIP have shown that reduced spectral measures of HRV in the very low- and low-frequency domain are also associated with increased risk of death.¹⁴ More recently some new indices describing non-linear heart rate dynamics, such as fractal scaling exponents¹⁵ and DC⁵ of heart rate have been emerged. These markers are more robust to non-stationarities and artifacts and have been proven to yield stronger prognostic information than the traditional measures of HRV among post-MI patients. Heart rate turbulence is a method that in comparison to other HRV measures quantifies the physiological short-term oscillation of cardiac cycle length that follows spontaneous ventricular premature complexes. Heart rate turbulence alone¹⁶ and in combination with other HRV parameters, such as DC³ has been shown to be a strong predictor of mortality and SCD after acute MI. Besides MI, the prognostic meaning of HRV has been evaluated in various other cardiovascular diseases, including patients with chronic heart failure,¹⁷ patients with severe aortic stenosis¹⁸ and patients presenting to the emergency department.¹⁹

One major problem of current autonomic risk markers is that their assessment is either too complicated or too time-consuming. In most clinical studies HRV was assessed from 24-h Holter recordings. Only very few studies have analysed the predictive value of HRV measures assessed from short-term recordings. Fei et al.¹² evaluated the predictive power of both short- and long-term HRV in 700 post-MI patients. Although the statistical association with cardiac mortality could be demonstrated for both, short-term SDNN and long-term HRVI, the predictive power of short-term SDNN was significantly lower than that of the 24-h HRVI. One reason for the lower predictive value of standard short-term HRV might be that spectral measures of short-term HRV are more prone to artifacts and noise, leading to substantial variations of normal values.²⁰ Assessment of short-term HRV by means of DC provides significant advantages over standard measures. Due to its underlying signal processing algorithm, DC is more robust to artifacts, noise, and non-stationarities. In addition, DC specifically captures deceleration-related oscillations of heart rate,^5,19 irrespective of their frequency, which are more closely related to vagal modulations. In our study, predictive power of DC_st was superior to other short-term HRV measures, including other non-linear indices of heart rate dynamics.

We also compared DC assessed from short-term and 24-h recordings. The two measurements were significantly correlated and their predictive value as estimated by ROC analysis was comparable. In multivariable analysis including both DC_st and DC_24h, DC_st was a significant and independent predictor of 3-year mortality, whereas DC_24h was not. One reason for the higher predictive value of DC_st in multivariable analysis might be that 30-min ECGs recordings are performed in standardized conditions compared to 24-h Holter recordings, which also include phases of physical activity, ECG lead displacement and non-stationarities created by variations in heart rate.

Clinical implications

An important finding of our study is that DC_st provides information of mortality risk on top of the information obtained by LVEF and GRACE score. In particular, in both Munich and Tuebingen cohorts the addition of DC_st led to a significant increase of C-statistics, IDI, and NRI scores. Assessment of DC from short-term recordings allows standardized, bedside risk assessment of post-MI patients, which makes the technology more suitable for clinical practice. Deceleration capacity-based risk assessment from short-term recordings could also open new perspectives in other fields of cardiac risk stratification, such as the integration of this technology in implantable devices.

Limitations

Our study has several limitations: first, autonomic function by means of DC can only be assessed in patients with sinus rhythm; second, in both cohorts, age was restricted to 80 years. The results cannot, therefore, be extrapolated to older patients; third, although we adjusted for various important risk predictors there are many other markers, such as heart rate recovery and brain natriuretic peptide, which were not routinely measured in this study; fourth, the use of total mortality as the primary endpoint has both advantages and disadvantages. Although the definition of death is without any ambiguity, the potential association with arrhythmic death might be lost. However, the incidence of unambiguous arrhythmic death in both cohorts was too low to power further statistical analyses.

Conclusion

In conclusion, DC_st is a strong and independent predictor of mortality and cardiovascular mortality after MI. The prognostic value of short-term DC is incremental to that of established risk markers, including LVEF and GRACE Score.

Supplementary material

Supplementary material is available at Europace online.

Conflict of interest: none declared.

Funding

The study was supported in part by grants from the program ‘Angewandte klinische Forschung’ (AKF) of the University of Tübingen 252-1-0 to A. Bauer. No additional external funding was received for this study. The Autonomic Regulation Trial (training cohort) was supported by Bundesministerium für Bildung, Wissenschaft, Forschung, und Technologie (13N/7073/7), the Kommission für Klinische Forschung and the Deutsche Forschungsgemeinschaft (SFB 386).

Author notes

Konstantinos D. Rizas and Christian Eick authors contributed equally to this work.