Baroreceptor reflex sensitivity (BRS) is an important prognostic factor because a reduced BRS has been associated with an adverse cardiovascular outcome. The threshold for a ‘reduced’ BRS was established by the ATRAMI study at BRS <3 ms/mmHg in patients with a previous myocardial infarction, and has been shown to improve risk assessment in many other cardiac dysfunctions. The successful application of this cut-off to other populations suggests that it may reflect an inherent property of baroreflex functioning, so our goal is to investigate whether it represents a ‘natural’ partition of BRS values. As reduced baroreflex responsiveness is also associated with ageing, we investigated whether a BRS estimate <3 ms/mmHg could be the result of a process of physiological senescence as well as a sign of BRS dysfunction. This study involved 228 chronic heart failure patients and 60 age-matched controls. Our novel method combined transfer function BRS estimation and automatic clustering of BRS probability distributions, to define indicative levels of different BRS activities. The analysis produced a fit clustering (cophenetic correlation coefficient 0.9 out of 1) and identified one group of homogeneous patients (well separated from the others by 3 ms/mmHg) with an increased BRS-based mortality risk [hazard ratio (HR): 3.19 (1.73, 5.89), P<0.001]. The age-dependent BRS cut-off, estimated by 5% quantile regression of log (BRS) with age (considering the age-matched controls), provides a similar mortality value [HR: 2.44 (1.37, 4.43), P=0.003]. In conclusion, the 3 ms/mmHg cut-off identifies two large clusters of homogeneous heart failure (HF) patients, thus supporting the hypothesis of a natural cut-off in the HF population. Furthermore, age was found to have no statistical impact on risk assessment, suggesting that there is no need to establish age-based cut-offs because 3 ms/mmHg optimally identifies patients at high mortality risk.

CLINICAL PERSPECTIVES

  • Previous studies have shown that a BRS <3 ms/mmHg identifies cardiac patients at higher mortality risk. The present study aimed to assess whether this threshold represents a natural partition of the data and whether its risk stratification capability depends on age.

  • We found, in a large cohort of HF patients, that 3 ms/mmHg is the best identifier of two groups with homogeneous BRS, and those with a BRS <3 ms/mmHg show increased cardiac mortality risk, independent of age.

  • Novel interventional and device-based therapies aimed at modulating the autonomic nervous system are under investigation in cardiac conditions associated with sympathetic overactivity and autonomic imbalance. The individual characterization of the arterial baroreflex regulation, and the recognition of a clinically meaningful dysregulation of the arterial baroreflex control, offer a number of opportunities for identifying patients who may benefit more from treatment and for tracking the effects of the applied therapies over time.

INTRODUCTION

It is well established that the joint analysis of beat-to-beat fluctuations in systolic blood pressure and the R–R interval series provides a useful indicator of the arterial–cardiac baroreceptor reflex sensitivity (BRS), expressed in milliseconds per millimetre of mercury (ms/mmHg) [1]. The BRS has been extensively studied in cardiac patients [2], particularly in those surviving an acute myocardial infarction (MI), but also in those with heart failure (HF) or left ventricular dysfunction. Most studies have shown that lower BRS values are associated with higher cardiovascular disease-related mortality within a few years of the event or the diagnosis [35]. However, there is no objective criterion to distinguish between different levels of BRS function, so the definition of depressed BRS has been commonly taken from the Autonomic Tone and Reflexes After Myocardial Infarction (ATRAMI) study [3], which was the first study to provide cut-off values for risk stratification according to BRS values. The ATRAMI was a large-scale prospective study, including 1284 participants with a recent MI and aimed at the identification of significant risk factors for cardiac mortality, including the phenylephrine BRS and the traditional parameter left ventricular ejection fraction (LVEF). The results showed that reduced BRS values increase risk of cardiac-related mortality independently of LVEF, with risk classes defined according to the sample BRS distribution: high risk below the 15th percentile (BRS <3.0 ms/mmHg), medium risk between the 15th and 50th percentiles (3.0≤BRS≤6.1 ms/mmHg) and low risk above the 50th percentile (BRS >6.1 ms/mmHg). The 3.0 ms/mmHg cut-off was also shown to be suitable for predicting 5-year mortality in MI patients with preserved LVEF or age >65 years [3].

Even though the ATRAMI cut-offs were determined in post-MI patients, such cut-offs (in particular 3.0 ms/mmHg) have also been used for patients with non-ischaemic dilated cardiomyopathy [6] and HF patients, including those under β-blocker treatment [7], with BRS assessed using the phenylephrine technique. Surprisingly, when BRS is estimated through bivariate spectral analysis of spontaneous oscillations of systolic arterial pressure and R–R intervals [8], the cut-off value for BRS, maximizing the hazard ratio (HR) for the 2-year outcome, was found to be 3.1 ms/mmHg. This value, obtained for transfer function BRS assessment, is remarkably close to the one obtained for the phenylephrine test. Indeed, transfer function (TF) and phenylephrine methods provide estimates with poor agreement, although with similar predictive value when considering the 3.0-ms/mmHg cut-off in HF patients [9].

Recently, Huikuri and Stein [10] pointed out the need to define an algorithm for determining the ideal cut-offs for different measures of heart rate variability. With regard to the BRS, the finding that the 3.0-ms/mmHg cut-off has been successfully applied in most studies using different methodologies suggests that this threshold might actually have a physiological meaning. The need to determine an ‘ideal’ cut-off value of autonomic variables is particularly relevant in the setting of HF patients in whom a device-based approach to autonomic nervous system modulation is under investigation for treatment of the disease. Although clinical studies obviously rely on clinical outcomes, to define whether these new interventions are biologically active at their physiological end-point (i.e. the autonomic balance) might help in identifying those patients who might have more benefit from further attention.

The first goal of the present study is to investigate whether the 3.0-ms/mmHg cut-off represents a natural partition of the data into clusters of homogeneous individuals. The present study is based on automatic agglomerative clustering of individual BRS-estimated probability density functions, so it takes into consideration the BRS estimate as well as the corresponding intraindividual BRS variability. Therefore, this methodology is independent of the patient outcome and based solely on the most probable BRS values for each individual.

The second goal is related to age effects. It is known that BRS decreases with age [1113], and age-specific reference BRS values have been reported for both a healthy working population (<60 years) and older individuals (>50 years) [11,12]. Thus, we investigated the possibility that the optimal cut-offs may vary as a function of the individual's age, a possibility that so far has not been considered in the literature.

EXPERIMENTAL

Study patients

This is a retrospective analysis conducted on recordings from a group of 228 HF patients collected within a previous study [8]. In short, 84% of the patients were male and the median age of the group was 54 years (ranging from 26 years to 70 years). The distribution of participants per New York Heart Association (NYHA) functional class (I, II, III, IV) was 10%, 51%, 37% and 2%, respectively, and per cause of chronic HF (CHF; ischaemic, idiopathic, other) it was 48%, 40% and 12%. The patients were all clinically stable, had not had an MI or cardiac surgery within the last 6 months, were in sinus rhythm and had a low ectopic event rate (<5%). We also considered a group of 60 healthy individuals in the same age range (26–66 years), selected from those enrolled in previous investigations at the same laboratory. None of them was on medication or had a chronic or acute disease.

Experimental protocol

The participants were studied in the morning in the supine position. The experimental protocol was carried out in the laboratory for autonomic evaluations in Montescano, Italy and comprised: (a) instrumentation, patient familiarization with paced breathing and signal stabilization; and (b) 8-min recording of an electrocardiogram (ECG), lung volume (Respitrace Plus, Non-Invasive Monitoring Systems), and non-invasive arterial blood pressure (Finapres 2300, Ohmeda) at a 250-Hz sampling frequency and during paced breathing at 15 breaths/min (0.25 Hz). Beat-by-beat R–R interval values (resolution 1 ms) were obtained from the ECG signals using a software package developed in house [14]. The R–R interval time series were then resampled at 2 Hz by cubic spline interpolation. The local ethical committee had approved the study. All patients provided written consent for the scientific treatment of their data in an anonymous form at the time of hospitalization. Individuals who were healthy also provided an informed written consent at the time of enrolment.

Patients followed recorded instructions to breathe in and out at a frequency of 0.25 Hz, with the inspiratory duty cycle [inspiratory time (TI)/total breath time (TTOT)] set at 0.4 [15]. A short familiarization session was carried out before starting the recording session. The reason for adopting the 0.25-Hz paced breathing protocol for BRS evaluation was 2-fold: first, the voluntary control of respiration at 0.25 Hz avoids the quantification of respiratory effects (mechanically on systolic arterial pressure and centrally on heart rate) in the low-frequency (LF) band [15]. In the condition of HF, about 50% of patients develop a periodic breathing pattern during supine laboratory recordings [16], which causes profound entrainment of cardiovascular rhythms [17]. Thus, voluntary control of respiration at frequencies outside the LF range is used as a means of avoiding this problem. Second, it is easier for a patient to follow a regular breathing pattern rather than keep the breathing rate within given limits. Finally, note that, despite slight hyperventilation, BRS evaluated under 0.25-Hz paced breathing showed no significant differences from BRS measured during spontaneous HF breathing (0.15–0.40 Hz) for both healthy individuals and HF patients [15,18].

METHODS

BRS estimate and probability distribution

BRS was evaluated by bivariate spectral analysis between systolic arterial pressure and R–R time series, and averaging the estimated gain of the TF in the 0.04- to 0.15-Hz range [19]. Spectral estimates were obtained via the Blackman–Tukey algorithm using a Parzen window [20] with a 0.03-Hz bandwidth [8,21]. In addition to the point BRS estimate, we estimated the BRS probability density function for each participant, in order to define an individualized range of probable BRS values. The probability density function is based on approximate 100(1 − a)% confidence intervals (CIs) for the TF gain per frequency, using the weighted covariance estimator [22], which has been found to be sufficiently accurate for use in practical BRS analysis [21]. For each frequency value, the 100(1 − a)% CI of the TF gain was quantified by varying 1 − a from 0.01 to 1 (stepwise 0.01), where the case of a=1 corresponds to the point TF estimate. Then, the probability density function percentile limits 100(1 − a)% were obtained by averaging the 100(1 − a)% CI limits over all frequencies, so that case a=1 corresponds to the BRS estimate for each participant. Using this procedure, the probability density function for the BRS of each participant was evaluated at 99 points uniformly spaced over the probability range (0,1).

Clustering methodology

The starting point of the clustering procedure is the panel of BRS probability density function estimates for each participant. This procedure builds the hierarchy from the individual elements by progressively merging clusters, using an appropriate dissimilarity measure and a group linkage criterion [23]. In the present study, the dissimilarity matrix D has entries Dij that correspond to the pairwise dissimilarity between participants i and j, and quantified by the weighted L2–Wasserstein distance. This distance corresponds to a weighted sum of squared differences between quantiles of distributions i and j and, therefore, the dissimilarity between two participants takes into account the differences between the BRS-estimated probability distributions and the differences between the mean/median BRS behaviour of the participants (i.e. the point BRS estimate).

A dendrogram was then obtained by applying the group linkage criterion to the matrix I–D. The dendrograms obtained by the different linkage procedures (single, complete and average) were compared using the cophenetic correlation coefficient, which is a measure of how faithfully the linkage preserves the pairwise distances between the original data (see, for example, Everitt et al. [23, p. 91]). Values close to 1 indicate that the linkage/dendrogram accurately reflects the data.

BRS distribution and prognostic studies as a function of age

To determine the impact of using an age-dependent BRS cut-off in prognostic studies rather than a constant threshold, we addressed two aspects. First, we investigated whether values of BRS below the ATRAMI cut-off value (3.0 ms/mmHg) could be considered either the result of a process of physiological senescence or a sign of BRS dysfunction. This was accomplished by comparing BRS as a function of age in CHF patients and age-matched controls. Second, we assessed the performance of age-dependent cut-offs in the outcome prognostic of CHF patients, compared with the constant 3.0 ms/mmHg value. The statistical analysis was based on logarithmically transformed BRS values due to the substantial positive skewedness of the BRS distribution per group (and lack of normality, as assessed by Kolmogorov–Smirnov testing), as well as to the enhancement of linearity between BRS and age [24]. The age effect on BRS in the HF and control groups was compared using analysis of covariance (ANCOVA), with age acting as a co-variate. Furthermore, the distribution of log (BRS) per age in the HF group was analysed with respect to the control group as follows: quantile regression was used to specify the 5% and 95% conditional quantile functions of log (BRS), given the age in controls [25], and to determine the interval in which 90% of the controls were included. Then, the proportion of HF patients and controls in each region was compared by independent-sample Student's t-test with Bonferroni's correction on bilateral hypothesis testing. A comprehensive summary of the theoretical formulation and estimation of parameters in quantile regression can be found in Gouveia et al. [26].

The predictive value of BRS depending on age was first assessed using Cox's analysis, with age as a continuous adjusting factor and the estimation of HRs and corresponding 95% CIs. Unless stated otherwise, all results are presented as the value [95% CI]. The Kaplan–Meier analysis was used to estimate survival curves for the age-specific threshold, based on the 5% quantile line from the control group, R5%, and compared with the constant log (3.0) threshold. Survival curves from different subgroups were compared by log-rank test (Mantel–Cox) and multiple comparisons with Bonferroni's correction. P<0.05 was considered statistically significant and all tests were two sided.

RESULTS

The dendrogram in Figure 1(a) shows the hierarchy from the individual elements by progressively merging clusters. It shows how participants are linked; the lower the linkage level, the higher the similarity between participants. The pairs of participants exhibiting the lowest distances are represented at the first level nodes of the dendrogram. At higher-level nodes of a dendrogram, the distance between clusters of distributions is higher, reflecting the larger variety of BRS distributions for the participants included in that cluster. The grouping criteria considered to build the dendrogram in Figure 1(a) were the average linkage procedure because it exhibits the highest cophenetic correlation coefficient (0.90). As this value is quite close to 1, the produced clustering is a close fit, thus properly reproducing the original matrix distance, D, between participants. The average linkage method is also shown to be less dependent on extreme values, producing clusters with small within-cluster variation and approximately equal variances.

Dendrogram and estimated BRS probability

Figure 1
Dendrogram and estimated BRS probability

(a) Dendrogram produced by an average link method according to the L2–Wasserstein distance on the BRS distributions. (b) Estimated BRS probability density for each participant, where the dashed line positions the empirical cut-off of 3.0 ms/mmHg. The participant group coloured green identified by cluster analysis (a) corresponds to the set of green coloured BRS estimated probability density functions, with lowest BRS values and lowest variability (b).

Figure 1
Dendrogram and estimated BRS probability

(a) Dendrogram produced by an average link method according to the L2–Wasserstein distance on the BRS distributions. (b) Estimated BRS probability density for each participant, where the dashed line positions the empirical cut-off of 3.0 ms/mmHg. The participant group coloured green identified by cluster analysis (a) corresponds to the set of green coloured BRS estimated probability density functions, with lowest BRS values and lowest variability (b).

The clustering procedure allows the identification of groups of homogeneous participants. In particular, Figure 1(a) highlights in green a group of participants with BRS distributions that show a high similarity within the group and a low similarity with the BRS distributions of the remaining participants. As displayed in Figure 1(b), the coloured group corresponds to the participants with the lowest mean BRS estimate and lowest variability in BRS distribution. Moreover, the empirical cut-off of 3.0 ms/mmHg separates the participants of the coloured group from the remaining ones as well.

BRS distribution is age dependent

The log (BRS) evaluated in the HF group was significantly lower than that of the control group (mean±S.D., 1.18 ± 0.96 vs 1.91 ± 0.66, P<0.0001; Figure 2a). Due to the fairly symmetrical shape of the log-transformed distributions (Figure 2a), the exponential of the log (BRS) average (i.e. the geometric mean of the BRS values) is a good estimate of the median of BRS values in ms/mmHg (3.25 [2.86, 3.68] vs 7.78 [5.73, 8.04]). As illustrated in Figure 2(b), log (BRS) correlated inversely with age in both groups, although the correlation is weaker in the HF than in the control group (Table 1). However, the larger sample size in the HF group makes possible an almost equivalent estimation precision in both groups and, consequently, regression confidence bands per group are of a similar width (Figure 2b). Statistical ANCOVA provided no significant interaction between age and group (P=0.46), indicating that HF and control slopes are not significantly different, so log (BRS) and age relate similarly in both groups. The joint regression analysis for HF and control groups indicates that the coefficients associated with age, intercept and group are significant in the regression model (Table 1). Therefore, HF patients were found to exhibit a lower log (BRS) value compared with controls, differing by a constant factor of 0.562 [0.44, 0.75] after adjusting for age.

Table 1
Results comparing HF and control groups
VariableHFControlP
No. of participants 228 60  
Log (BRS) 1.18±0.96 1.91±0.66 <0.0001 
BRS (ms/mmHg) geometric mean [95% CI] 3.25 [2.86, 3.68] 7.78 [5.73, 8.04]  
Log (BRS) and age regression Log (BRS)=−0.017×age+2.10,(r=−0.17, F=5.77, P<0.02) Log (BRS)=−0.027×age+3.13,(r=0.53, F=22.70, P<0.0001)  
Log (BRS) and age regression Log (BRS)=−0.022×age+2.844 − 0.562×(1 − group), P<0.0001 
 (group=0 for HF and group=1 for controls) (F=16.99) 
R5% – Log (BRS)=−0.026×age+2.19  
R95% – Log (BRS)=−0.014×age+3.34  
Percentage participants <R5% 76/228 4/60 <0.001 
Percentage participants >R95% 6/228 4/60  0.13 
Percentage participants <3.0 98/228 7/60 <0.001 
VariableHFControlP
No. of participants 228 60  
Log (BRS) 1.18±0.96 1.91±0.66 <0.0001 
BRS (ms/mmHg) geometric mean [95% CI] 3.25 [2.86, 3.68] 7.78 [5.73, 8.04]  
Log (BRS) and age regression Log (BRS)=−0.017×age+2.10,(r=−0.17, F=5.77, P<0.02) Log (BRS)=−0.027×age+3.13,(r=0.53, F=22.70, P<0.0001)  
Log (BRS) and age regression Log (BRS)=−0.022×age+2.844 − 0.562×(1 − group), P<0.0001 
 (group=0 for HF and group=1 for controls) (F=16.99) 
R5% – Log (BRS)=−0.026×age+2.19  
R95% – Log (BRS)=−0.014×age+3.34  
Percentage participants <R5% 76/228 4/60 <0.001 
Percentage participants >R95% 6/228 4/60  0.13 
Percentage participants <3.0 98/228 7/60 <0.001 

BRS in HF and control groups

Figure 2
BRS in HF and control groups

(a) Boxplot of log (BRS) per group and (b) dispersion diagram of age vs log (BRS) for the HF group. In (a), the 95% CI for the group median is represented by the boxplot notch around the median. In (b), the linear regression line with 95% CI is shown for HF patients and the shaded area defines the 95% CI for controls.

Figure 2
BRS in HF and control groups

(a) Boxplot of log (BRS) per group and (b) dispersion diagram of age vs log (BRS) for the HF group. In (a), the 95% CI for the group median is represented by the boxplot notch around the median. In (b), the linear regression line with 95% CI is shown for HF patients and the shaded area defines the 95% CI for controls.

The log (BRS) distribution per age was also compared across the different groups. Figure 3(a) displays the 5% and the 95% conditional quantile lines obtained for the control group (R5% and R95% with equations in Table 1), dividing the controls into two sets (below/above the line), where approximately 5% and 95% of the participants are below the line, respectively. Therefore, the region between R5% and R95% contains approximately 90% of the controls. As illustrated in Figure 3(b), the proportion of HF patients below R5% is larger than that observed for controls (76/228 vs 4/60, P<0.001; Table 1), although there are no significant differences between those proportions evaluated above the R95% line (6/228 vs 4/60, P=0.13; Table 1). Thus, the lower log (BRS) average in HF compared with controls is a consequence of the lower minimum and similar maximum log (BRS) values per age in HF patients when compared with controls. Finally, the proportion of HF patients with BRS <3.0 ms/mmHg is larger than for the controls (98/228 vs 7/60, P<0.001; Table 1) and larger than that evaluated for HF patients below R5% (98/228 vs 76/228, P=0.017; Table 1).

Log (BRS) versus age in controls and HF patients

Figure 3
Log (BRS) versus age in controls and HF patients

(a) Controls and (b) HF patients. The colours in (b) follow the clustering displayed in Figure 1. Full lines localize the empirical cut-offs 3.0 and 6.1 ms/mmHg and dotted lines localize R5% and R95% lines obtained from controls. Top and bottom statistics represent the percentage of participants, <R5% and >R95%, respectively, for each group.

Figure 3
Log (BRS) versus age in controls and HF patients

(a) Controls and (b) HF patients. The colours in (b) follow the clustering displayed in Figure 1. Full lines localize the empirical cut-offs 3.0 and 6.1 ms/mmHg and dotted lines localize R5% and R95% lines obtained from controls. Top and bottom statistics represent the percentage of participants, <R5% and >R95%, respectively, for each group.

BRS prognostic value is not age dependent

In the HF group, Cox's regression analysis shows that log (BRS) is significantly associated with the outcome [HR: 0.61 (0.47, 0.79), P<0.001], thus indicating that lower BRS values are associated with cardiac mortality. With age as the adjusting factor in a multivariate Cox's model, log (BRS) maintains a highly significant association with the outcome [HR: 0.61 (0.47, 0.80), P<0.001], although there was no evidence of age impact on the outcome [HR: 1.01 (0.98, 1.05), P=0.48]. The log (BRS) values dichotomized in two regions are also associated with outcome when considering either the log (3.0) threshold [HR: 3.19 (1.73, 5.89), P<0.001] or the age-dependent threshold [HR: 2.44 (1.37, 4.43), P=0.003]; in both cases, patients with log (BRS) values below the threshold show increased cardiac mortality risk. Moreover, the mortality prognoses in HF patients based on either the constant or the age-dependent threshold were found to be similar (Figure 4). Notice that such a result is expected because the value of log (3.0) intersects the age-based threshold R5% roughly in the middle of the age range (around 45 years, see Figure 3), and the slope of log (BRS) over the age regression line is not too steep. Therefore, the patients falling in the region below the thresholds [either log (3.0) or R5% defining the high-risk region] are approximately the same (see Figure 3b).

Kaplan–Meier survival curves

Figure 4
Kaplan–Meier survival curves

Kaplan–Meier survival curves showing BRS dichotomization according to (a) log (BRS) < log (3.0) criterion and (b) log (BRS) < R5% criterion.

Figure 4
Kaplan–Meier survival curves

Kaplan–Meier survival curves showing BRS dichotomization according to (a) log (BRS) < log (3.0) criterion and (b) log (BRS) < R5% criterion.

Survival functions were also obtained when considering four distinct regions (Figure 5a), to investigate mortality risk for the patients with different risk at log (3.0) and R5% criteria. Patients at high risk for the log (3.0) criterion and at low risk for the R5% criterion are identified in blue (region 3). Grey identifies patients in region 2, i.e. a region with the same size as region 3 which includes patients of the same age [patients with log (BRS) in between log (3.0) and the line obtained by reflection of R5% at log (3.0), i.e. log (BRS)=−0.026×age − 2.19+2×log (3.0)]. Finally, regions 1 and 4 correspond to regions of low and high risk of mortality respectively, considering both criteria. As observed in Figure 5b, the survival in region 3 is similar to that in region 4 (P=0.90) whereas it is significantly different from that of regions 2 and 1 (P=0.038 and 0.031, respectively). Finally, the estimated survival rate is higher in region 2 than in region 1, although not exhibiting significant differences (P=0.55).

Log (BRS) vs age and Kaplan–Meier survival curves

Figure 5
Log (BRS) vs age and Kaplan–Meier survival curves

(a) Log (BRS) vs age in HF group. Lines represent R5%, log (3.0) and R5% line mirrored by reflection at log (3.0) (dashed line). (b) Kaplan–Meier survival curves according to the regions coloured in (a). Pairwise comparisons indicate significant differences between all regions, except for the pairs of regions (1, 2) and (3, 4) at the 5% significance level.

Figure 5
Log (BRS) vs age and Kaplan–Meier survival curves

(a) Log (BRS) vs age in HF group. Lines represent R5%, log (3.0) and R5% line mirrored by reflection at log (3.0) (dashed line). (b) Kaplan–Meier survival curves according to the regions coloured in (a). Pairwise comparisons indicate significant differences between all regions, except for the pairs of regions (1, 2) and (3, 4) at the 5% significance level.

DISCUSSION

There are two main points to highlight. First, the results of this study suggest that the 3.0 cut-off on BRS estimates represents a natural partition of HF patients at risk. Second, the 3.0 cut-off is rather independent of age in patients with HF.

The results of the present study are relevant in the current clinical scenario, in which a number of different approaches, including baroreceptor activation therapy, have been developed to modulate those autonomic abnormalities that characterize HF. Although it is assumed that all these procedures would act on a deranged autonomic balance to improve it, there was no systematic evaluation of the effects of such therapeutic options on the autonomic profile of the treated patients. The appropriate identification of patients who can benefit from these procedures is still an unsettled issue needing evaluation.

Clinical value of the 3.0 cut-off

Patients with a BRS estimate <3.0 were shown to exhibit higher mortality risk in several literature studies [1,3,57]. Such a threshold seems to be constant through different methodologies. In ATRAMI and also in other studies, BRS has been assessed using the drug infusion technique with phenylephrine. The 3.0 cut-off value of the phenylephrine method is close to the 3.1 cut-off value obtained by a very distinct method, in which the TF method was used for BRS evaluation [8].

The results in the present study support the concept that a cut-off value of around 3 ms/mmHg can be viewed as a biological threshold for the functioning of the baroreflex. In fact, in this sample of HF patients, the 3.0-ms/mmHg cut-off value represents the intrinsic grouping of BRS distributions. This is quite clear from the visualization of the results in the dendrogram and corresponding estimated probability density function distributions (see Figure 1), where it is possible to observe that the group with highly similar probability density functions (see Figure 1a) constitutes the group of patients exhibiting a punctual BRS estimate lower than 3.0 ms/mmHg (see Figure 1b).

The natural partition of BRS data around 3 ms/mmHg also suggests that, below this level, a proper functioning of the baroreceptor reflex is no longer present and the R–R interval changes (if any) are no longer linearly related to blood pressure oscillations. This is also demonstrated by the observation that a coherence value >0.5, generally taken to satisfy the assumption of linearity, is often not found for the lower estimates of BRS.

Studies with ganglionic blockade also support the concept that, in the presence of functional denervation, the TF gain is substantially decreased but not completely abolished. Zhang et al. [27] calculated the mean value of the TF gain in the LF band in resting humans after complete ganglionic blockade. This is equivalent to measuring BRS according to the method used in the present study after functionally opening the baroreflex arc. These authors found that BRS was 0.9±0.1 ms/mmHg (mean±S.E.M.) in a sample of 10 participants. Assuming that ganglionic blockade was fully effective in all participants, 1.52 ms/mmHg represents an estimate of an upper boundary of biological/methodological noise, if a 95% range of variation in the BRS values of the sample was considered. Thus, a BRS value <3.0 ms/mmHg would be indicative of complete impairment or marked attenuation of the baroreceptor heart rate reflex.

BRS and age

It is largely acknowledged that increasing age significantly reduces baroreflex control of the heart rate. The decline of BRS with age was first described by Gribbin et al. [28] and was subsequently confirmed by several studies that consistently showed the relationship with age being diminished or lost beyond the age of 60, which suggests that most of the reduction seen in BRS with age has already occurred by the fifth decade [2932]. Recent studies showed that decreasing baroreflex sensitivity with ageing is not due to efferent autonomic dysfunction [33].

As expected in the present study, log (BRS) is significantly lower for HF than for control groups, being inversely correlated with age in both groups. To our knowledge no literature is available aimed at exploring the linear association between BRS estimates and age for HF patients, although it is expected that the BRS would decrease with age. In the present study, there was no significant interaction between log (BRS) and group, thus indicating that the BRS decay/slope with age is similar in HF patients and controls and, consequently, differences between the groups are constant across all ages. Different results were obtained in a recent study when male patients who had had a recent MI were compared with age-matched controls [5]. Although the BRS slope was found to be significantly lower in the MI patients, the decrease in BRS associated with age was significantly steeper in controls than in MI patients. Unfortunately, we do not have a tenable explanation for this disparity. It is possible, however, that the limited number of control participants in the study of Hartikainen et al. [5] and the inclusion of younger participants (range 30–69 years) might have resulted in an overestimation of the BRS values in younger participants (due to the increased inter- and intra-individual variability at higher BRS values).

The results with regard to the correlation between BRS and age in controls (n=60) agree with those of other studies using larger groups (e.g. n=1026 [11]), which support their use as a reference group in the present study. Figure 3a represents the BRS estimates for normal participants as a function of age and demonstrates the tendency for BRS to decrease significantly with increasing age (see Table 1). In particular, Kardos et al. [11] obtained a log (BRS) age-related regression line with similar parameters [log (BRS)=−0.027×age+3.24, n=1026, r2=0.23, P<0.0001] in a much larger group of healthy working individuals aged between 18 and 60 years. The slightly higher value at the origin provided by Kardos et al. [11], compared with the present study (3.24 vs 3.13, with no significant differences), may be attributed to the differences in BRS estimation approaches and breathing protocol. Specifically, Kardos et al. [11] quantified the BRS through the sequences technique and spontaneous breathing, which are factors that have been shown to provide higher BRS estimates than those obtained by spectral analysis and controlled breathing at 15/min [12,15,34,35]. Moreover, the age-dependent thresholds obtained in the present study (R5% and R95%) are lower than those depicted in Kardos et al. [11], showing similar variability in the log (BRS) values around the age-dependent regression line (r2=0.28 in the present study and 0.23 in Kardos et al. [11]). These arguments also support the idea that the differences between the studies reflect the different approaches to estimating BRS.

BRS cut-off value and prognosis

As reduced baroreflex responsiveness is also associated with ageing, age was introduced as a co-variate to investigate whether cut-off values vary as a function of age. First, as assessed by Cox's regression, log (BRS) was significantly associated with the outcome. When age was considered in the regression, log (BRS) maintained a significant association with outcome, whereas age had no significant impact. Second, log (BRS) values dichotomized in two regions [either with the log (3.0) or the age-based R5% criterion] had significant association with outcome, because both cut-offs clearly identified the same patients at high risk. To further investigate differences between cut-offs, the individuals at risk for R5% criteria and not for log (3.0) were considered separately (mainly the older participants) and found to exhibit a lower mortality risk, similar to that evaluated for the patients with a log (BRS) higher than log (3.0) and similar age. Therefore, our results do not indicate the need to establish an age-based cut-off. Rather, they corroborate the hypothesis that log (3.0) is a natural threshold in the HF population that can optimally identify patients at high mortality risk.

CONCLUSIONS

This study provides mathematical and statistical support for the age-independent biological value of 3 ms/mmHg, which has been used for defining a depressed BRS. The analysis in the present study shows that this threshold value represents a natural partition in the HF population, through BRS probability functions (including punctual BRS estimates and intraindividual variability) obtained from the analysis of spontaneous oscillations in heart rate and blood pressure. Thus, these data reinforce the clinical applicability of this measure, not only for risk stratification purposes but also as a means of tracking effects of therapeutic interventions targeted at improving autonomic modulation and baroreceptor activity.

AUTHOR CONTRIBUTION

All authors contributed extensively to the work presented in this paper. Sónia Gouveia led the study conception/design, computational algorithm implementation, statistical analysis and interpretation of data and wrote the paper. Roberto Maestri was involved in data management and computational implementation. All co-authors provided participation in the study design, discussion of the results and writing/reviewing of the paper, contributing with critical revisions and important intellectual content. All authors provided final approval of this paper.

FUNDING

This work was supported by Portuguese Funds through the Portuguese Foundation for Science and Technology (‘FCT–Fundação para a Ciência e a Tecnologia’), in the context of the projects UID/CEC/00127/2013 and Incentivo/EEI/UI0127/2014 (IEETA/UA, Instituto de Engenharia Electrónica e Informática de Aveiro, “http://www.ieeta.pt”) and UID/MAT/04106/2013 (CIDMA/UA, Centro de I&D em Matemática e Aplicações, “http://www.cidma.mat.ua.pt). S. Gouveia acknowledges the postdoctoral grant by FCT [SFRH/BPD/87037/2012], financed through POPH–QREN programme (European Social Fund and National funds). This study was funded exclusively by institutional resources.

Abbreviations

     
  • ANCOVA

    analysis of covariance

  •  
  • BRS

    baroreceptor reflex sensitivity

  •  
  • CHF

    chronic heart failure

  •  
  • CI

    confidence interval

  •  
  • ECG

    electrocardiogram

  •  
  • HF

    heart failure

  •  
  • HR

    hazard ratio

  •  
  • LVEF

    left ventricular ejection fraction

  •  
  • MI

    myocardial infarction

  •  
  • TF

    transfer function

References

References
1
La Rovere
 
M.T.
Pinna
 
G.D.
Raczak
 
G.
 
Baroreflex sensitivity: measurement and clinical implications
Ann. Noninvasive Electrocardiol.
2008
, vol. 
13
 (pg. 
191
-
207
)
[PubMed]
2
Pinna
 
G.D
Maestri
 
M.
La Rovere
 
M.T.
 
Assessment of baroreflex sensitivity from spontaneous oscillations of blood pressure and heart rate: proven clinical value?
Physiol. Meas.
2015
, vol. 
36
 (pg. 
741
-
753
)
[PubMed]
3
La Rovere
 
M.T.
Bigger
 
J.T.
Marcus
 
F.I.
Mortara
 
A.
Schwartz
 
P.J.
 
for the ATRAMI investigators
Baroreflex sensitivity and heart-rate variability in prediction of total cardiac mortality after myocardial infarction
Lancet
1998
, vol. 
351
 (pg. 
478
-
84
)
[PubMed]
4
De Ferrari
 
G.M.
Sanzo
 
A.
Bertoletti
 
A.
Specchia
 
G.
Vanoli
 
E.
Schwartz
 
P.J.
 
Baroreflex sensitivity predicts long-term cardiovascular mortality after myocardial infarction even in patients with preserved left ventricular function
J. Am. Coll. Cardiol.
2007
, vol. 
50
 (pg. 
2285
-
2290
)
[PubMed]
5
Hartikainen
 
J.
Mäntysaari
 
M.
Mussalo
 
H.
Tahvanainen
 
K.
Länsimies
 
E.
Pyörälä
 
K.
 
Baroreflex sensitivity in men with recent myocardial infarction: impact of age
Eur. Heart J.
1994
, vol. 
15
 (pg. 
1512
-
1519
)
[PubMed]
6
Klingenheben
 
T.
Ptaszynski
 
P.
Hohnloser
 
S.H.
 
Heart rate turbulence and other autonomic risk markers for arrhythmia risk stratification in dilated cardiomyopathy
J. Electrocardiol.
2008
, vol. 
41
 (pg. 
306
-
311
)
[PubMed]
7
La Rovere
 
M.T.
Pinna
 
G.D.
Maestri
 
R.
Robbi
 
E.
Caporotondi
 
A.
Guazzotti
 
G.
Febo
 
O.
 
Prognostic implications of baroreflex sensitivity in heart failure patients in the beta-blocking era
J. Am. Coll. Cardiol.
2009
, vol. 
53
 (pg. 
193
-
199
)
[PubMed]
8
Pinna
 
G.D.
Maestri
 
R.
Capomolla
 
S.
Febo
 
O.
Robbi
 
E.
Cobelli
 
F.
La Rovere
 
M.T.
 
Applicability and clinical relevance of the transfer function method in the assessment of baroreflex sensitivity in heart failure patients
J. Am. Coll. Cardiol.
2005
, vol. 
46
 (pg. 
1314
-
1321
)
[PubMed]
9
La Rovere
 
M.T.
Maestri
 
R.
Robbi
 
E.
Caporotondi
 
A.
Guazzotti
 
G.
Febo
 
O.
Pinna
 
G.D.
 
Comparison of the prognostic values of invasive and noninvasive assessments of baroreflex sensitivity in heart failure
J. Hypertens.
2011
, vol. 
29
 (pg. 
1546
-
1552
)
[PubMed]
10
Huikuri
 
H.V.
Stein
 
P.K.
 
Heart rate variability in risk stratification of cardiac patients
Prog. Cardiovasc. Dis.
2013
, vol. 
56
 (pg. 
153
-
159
)
[PubMed]
11
Kardos
 
A.
Watterich
 
G.
de Menezes
 
R.
Csanády
 
M.
Casadei
 
B.
Rudas
 
L.
 
Determinants of spontaneous baroreflex sensitivity in a healthy working population
Hypertension
2001
, vol. 
37
 (pg. 
911
-
916
)
12
Gerritsen
 
J.
Voorde
 
B.J.T.
Dekker
 
J.M.
Kostense
 
P.J.
Bouter
 
L.M.
Heethaar
 
R.M.
 
Baroreflex sensitivity in the elderly: influence of age, breathing and spectral methods
Clin. Sci.
2000
, vol. 
99
 (pg. 
371
-
381
)
[PubMed]
13
Fauvel
 
J.P.
Cerutti
 
C.
Mpio
 
I.
Ducher
 
M.
 
Aging process on spectrally determined spontaneous baroreflex sensitivity: a 5-year prospective study
Hypertension
2007
, vol. 
50
 (pg. 
543
-
546
)
[PubMed]
14
Maestri
 
R.
Pinna
 
G.D.
 
POLYAN: a computer program for polyparametric analysis of cardio-respiratory variability signals
Comput. Meth. Prog. Bio.
1998
, vol. 
56
 (pg. 
37
-
48
)
15
Pinna
 
G.D.
Maestri
 
R.
La Rovere
 
M.T.
Gobbi
 
E.
Fanfulla
 
F.
 
Effect of paced breathing on ventilatory and cardiovascular variability parameters during short-term investigations of autonomic function, Am
J. Physiol. Heart Circ. Physiol.
2006
, vol. 
290
 (pg. 
H424
-
H433
)
16
La Rovere
 
M.T.
Pinna
 
G.D.
Maestri
 
R.
Robbi
 
E.
Mortara
 
A.
Fanfulla
 
F.
Febo
 
O.
Sleight
 
P.
 
Clinical relevance of short-term day-time breathing disorders in chronic heart failure patients
Eur. J. Heart Fail.
2007
, vol. 
9
 (pg. 
949
-
954
)
[PubMed]
17
Mortara
 
A.
Sleight
 
P.
Pinna
 
G.D.
Maestri
 
R.
Prpa
 
A.
La Rovere
 
M.T.
Cobelli
 
F.
Tavazzi
 
L.
 
Abnormal awake respiratory patterns are common in chronic heart failure and may prevent evaluation of autonomic tone by measures of heart rate variability
Circulation
1997
, vol. 
96
 (pg. 
246
-
252
)
[PubMed]
18
Pinna
 
G.D.
Maestri
 
R.
Robbi
 
E.
Gnemmi
 
M.
La Rovere
 
M.T.
 
Effect of controlled breathing on short-term cardiovascular variability: an investigation in chronic heart failure patients
Proc. Comput. Cardiol.
2001
, vol. 
28
 (pg. 
217
-
219
)
19
Pinna
 
G.D.
Maestri
 
R.
Raczak
 
G.
La Rovere
 
M.T.
 
Measuring baroreflex sensitivity from the gain function between arterial pressure and heart period
Clin. Sci.
2002
, vol. 
103
 (pg. 
81
-
88
)
[PubMed]
20
Priestley
 
M.B.
 
Spectral Analysis and Time Series
1981
London
Academic Press
21
Pinna
 
G.D.
Maestri
 
R.
 
Reliability of transfer function estimates in cardiovascular variability analysis
Med. Biol. Eng. Comput.
2001
, vol. 
39
 (pg. 
338
-
347
)
[PubMed]
22
Koopmans
 
L.H.
 
The Spectral Analysis of Time Series
1974
New York
Academic Press
23
Everitt
 
B.S.
Landau
 
S.
Leese
 
M.
Stahl
 
D.
 
Cluster Analysis
2011
Chichester
Wiley
24
Dawson
 
S.L.
Robinson
 
T.G.
Youde
 
J.H.
Martin
 
A.
James
 
M.A.
Weston
 
P.J.
Panerai
 
R.B.
Potter
 
J.F.
 
Older subjects show no age-related decrease in cardiac baroreceptor sensitivity
Age Ageing
1999
, vol. 
28
 (pg. 
347
-
53
)
[PubMed]
25
Koenker
 
R.
 
Quantile Regression
2005
Cambridge
Cambridge University Press
26
Gouveia
 
S.
Rocha
 
C.
Rocha
 
A.P.
Silva
 
M.E.
 
Time domain BRS estimation: least squares versus quantile regression
Proc. Comp. Cardiol.
2010
, vol. 
37
 (pg. 
931
-
934
)
27
Zhang
 
R.
Iwasaki
 
K.
Zuckerman
 
J.H.
Behbehani
 
K.
Crandall
 
C.G.
Levine
 
B.D.
 
Mechanisms of blood pressure and R–R variability: insight from ganglionic blockade in humans
J. Physiol.
2002
, vol. 
543
 (pg. 
337
-
48
)
[PubMed]
28
Gribbin
 
B.
Pickering
 
T.G.
Sleight
 
P.
Peto
 
R.
 
Effect of age and high blood pressure on baroreflex sensitivity in man
Circ. Res.
1971
, vol. 
29
 (pg. 
424
-
31
)
[PubMed]
29
Laitinen
 
T.
Hartikainen
 
J.
Vanninen
 
E.
Niskanen
 
L.
Geelen
 
G.
Länsimies
 
E.
 
Age and gender dependence of baroreflex sensitivity in healthy subjects
J. Appl. Physiol.
1998
, vol. 
84
 (pg. 
535
-
40
)
30
James
 
M.A.
Robinson
 
T.G.
Panerai
 
R.B.
Potter
 
J.F.
 
Arterial baroreflex–cardiac reflex sensitivity in the elderly
Hypertension
1996
, vol. 
28
 (pg. 
953
-
60
)
[PubMed]
31
Ebert
 
T.J.
Morgan
 
B.J.
Barney
 
J.A.
Denahan
 
T.
Smith
 
J.J.
 
Effects of aging on baroreflex regulation of sympathetic activity in humans
Am. J. Physiol.
1992
, vol. 
263
 (pg. 
798
-
803
)
32
Piccirillo
 
G.
Cacciafesta
 
M.
Viola
 
E.
Santagada
 
E.
Nocco
 
M.
Lionetti
 
M.
Bucca
 
C.
Moisè
 
A.
Tarantini
 
S.
Marigliano
 
V.
 
Influence of aging on cardiac baroreflex sensitivity determined non-invasively by power spectral analysis
Clin. Sci.
2001
, vol. 
100
 (pg. 
267
-
274
)
[PubMed]
33
O'Mahony
 
D.
Bennett
 
C.
Green
 
A.
Sinclair
 
A.J.
 
Reduced baroreflex sensitivity in elderly humans is not due to efferent autonomic dysfunction
Clin. Sci.
2000
, vol. 
98
 (pg. 
103
-
110
)
[PubMed]
34
Persson
 
P.B.
DiRienzo
 
M.
Castiglioni
 
P.
Cerutti
 
C.
Pagani
 
M.
Honzikova
 
N.
Akselrod
 
S.
Parati
 
G.
 
Time versus frequency domain techniques for assessing baroreflex sensitivity
J. Hyperten.
2001
, vol. 
19
 (pg. 
1699
-
1705
)
35
Bernardi
 
L.
Porta
 
C.
Spicuzza
 
L.
Bellwon
 
J.
Spadacini
 
G.
Frey
 
A.W.
Yeung
 
L.Y.
Sanderson
 
J.E.
Pedretti
 
R.
Tramarin
 
R.
 
Slow breathing increases arterial baroreflex sensitivity in patients with chronic heart
Circulation
2002
, vol. 
105
 (pg. 
143
-
145
)
[PubMed]