- Email: [email protected]
Refining the deceleration capacity index in phase-rectified signal averaging to assess physical conditioning level
Available online at www.sciencedirect.com
ScienceDirect Journal of Electrocardiology 47 (2014) 306 – 310 www.jecgonline.com
Refining the deceleration capacity index in phase-rectified signal averaging to assess physical conditioning level☆ Olivassé Nasario-Junior, DSc, a Paulo Roberto Benchimol-Barbosa, MD, DSc, b,⁎ Jurandir Nadal, DSc a a
Programa de Engenharia Biomédica, COPPE, Universidade Federal do Rio de Janeiro, Rio de Janeiro, RJ, Brazil b Hospital Universitário Pedro Ernesto, Universidade do Estado do Rio de Janeiro, Rio de Janeiro, RJ, Brazil
Abstract
Background: Deceleration capacity (DC) of heart rate is a measure of cardiac vagal modulation. This study introduced a DC adaptation (Modified Index) that measured the velocity of change in the phase-rectified signal averaging curve, and assessed its ability to discriminate athletes from controls. Materials and methods: The Modified Index was compared to Standard DC approach in a prospective case–control study. Subjects were classified according to maximal metabolic equivalents as the control group (CG) and athlete group (AG). The Modified Index was compared to Standard DC and classical approaches (RMSSD and HF) by the area under receiver operating characteristic curve (AUC) using 10,000 bootstraps. Results: In Standard DC and Modified Index bootstrap median values were (ms), respectively, 11.80 and 17.94 (p b 0.01) in CG, and 25.98 and 45.62 in AG (p b 0.01). AUC (mean ± SD) was 0.70 ± 0.12 for Standard DC and 0.96 ± 0.04 for Modified Index (p b 0.01). Conclusions: Modified Index appropriately discriminates athletes from healthy sedentary subjects. © 2014 Elsevier Inc. All rights reserved.
Keywords:
HR variability; Phase-rectified signal averaged; Cardiac vagal modulation; Deceleration capacity index; Physical conditioning
Introduction Regular aerobic exercise provides beneficial changes on the cardiovascular system, characterized by mechanical, autonomic and electrophysiological remodeling [1]. Autonomic remodeling is evidenced by both resting heart rate (HR) reduction and cardiac vagal modulation increase [2,3]. HR variability (HRV) is a non-invasive diagnostic tool employed to assess autonomic nervous system (ANS) activity. HRV is modulated by both sympathetic and vagal efferent and afferents neural pathways over sinus and AV nodes [4,5]. Recently, a novel index termed deceleration capacity (DC) of HR has been proposed to measure cardiac vagal modulation. By detecting anchor points, which directs a phase-rectified signal averaging (PRSA) procedure, this
☆ Each author has contributed to read and approved the manuscript; and none of the authors have any conflict of interest. The manuscript is original, and any part of it has not been and will not be submitted elsewhere for publication. ⁎ Corresponding author at: Boulevard Vinte e Oito de Setembro, 77 Board of Directors Suite, Rio de Janeiro 20551-030, Brazil. E-mail address: [email protected]
0022-0736/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jelectrocard.2013.12.006
technique eliminates noises that permeate RR interval series and yields a quasi-periodic oscillation from a non-stationary signal [6,7]. The averaging of RR interval data segments is synchronized by anchor points, which allows internal quasiperiodic components enhancement, and non-stationary parts abated. However, some anchor points may lack adequate synchronization because of a failure of the classical approach to analyze slope variations of successive RR intervals [8,9]. It was hypothesized that depending on vagal stimulus intensity, the rate of ascend of the RR interval series would change accordingly, determining slope variation. Thus, a strongest vagal stimulus would determine a steepest slope and vice versa, potentially affecting the synchronization output. Although, PRSA has been originally developed to risk stratify subjects post-myocardial infarction [6], its application in assessing physiological conditions strongly related to vagal activity modulation is attractive. Thus, the present study proposes a modified DC index (algorithm adaptations) that (i) selects anchor point of PRSA in deceleration phase with the highest derivative and (ii) measures the velocity of change in the PRSA curve, then tests both classical and proposed methods to distinguish athletes from sedentary healthy volunteers.
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
Methods Study population The analyzed signals were extracted from an existing high-resolution ECG database as described previously [1]. The study protocol was approved by a local ethics committee and informed consent was obtained from each volunteer. Ten elite runners ([mean ± SD] 8.9 ± 3.2 years of training; 6 to 8 training sessions/wk; 90 to 120 min/session; 90 to 110 km/ wk) were enrolled (athlete group). A group of 10 healthy sedentary volunteers were included as control (control group). Inclusion criteria, physical assessment procedures and maximal oxygen consumption (VO2MAX) estimation protocol have been published elsewhere [1]. VO2MAX was divided by the constant 3.5 mL kg − 1 min − 1 to be converted into metabolic equivalents (METs). The control and athlete groups were separated according to estimated VO2MAX, by arbitrarily defining as less than 11.5 METs for sedentary controls and more than 16.0 METs for athletes. Both groups were matched by gender and anthropometric data (Table 1). The aim of anthropometric data intergroup matching was to minimize intergroup physiological and anthropometric variability, thus reducing potential effect of thoracic geometry on surface ECG signals. ECG was acquired shortly after application of the questionnaire and physical examination. Before a 15-min continuous signal acquisition, subjects remained in the supine position for 5 min for stabilization of autonomic modulation after the change from the orthostatic position [10]. Detailed description of the processing and equipment was previously described [11]. The detection of the RR intervals, PRSA algorithm and analysis of the results were performed using a custom-made program in Matlab language. PRSA methods (Standard DC) The PRSA technique has been described elsewhere [6–9]. In the RR interval series, the first step was to identify each of the data points as deceleration anchor. If a particular RR interval increased relative to the previous one, it is identified as a deceleration anchor interval (represented by circle or lozenge symbols in Fig. 1). Any RR intervals that exhibited more than a 20% change from the previous RR interval were
307
excluded, as they are likely to be related to measurement noise or ectopic beats [12]. After determining the deceleration anchors, a window surrounding each anchor is created. The window is defined by the two intervals immediately preceding and following the anchor interval. All of the deceleration windows are then aligned at the anchor point (phase rectified). Once aligned, the respective intervals are averaged together. Once averaged, the DC is calculated by: DC ¼ ½RRð0Þ þ RRð1Þ−RRð−1Þ−RRð−2Þ=4
ð1Þ
where RR(0) is defined as the anchor point (Fig. 2a). Data segments synchronized by anchor interval were averaged so that the internal quasi-periodic components were enhanced and non-stationary parts of the signals were filtered off. However, according to Pan et al. [9], generally, the anchor points could be categorized into two types, as illustrated in Fig. 1. The “type I” anchor points indicated a steeper slope, while the “type II” anchor points are uncertain to be on increasing or decreasing trend. Hence, it remained a doubt whether slope variations of successive RR intervals anchor points may lack adequate synchronization, potentially damaging the PRSA method. PRSA procedure by synchronizing faster RR intervals (Anchor Selection) A couple of studies showed significant improvement in cardiovascular risk stratification for PRSA variables, through alternative selection criteria of anchors intervals [8,9]. To further optimize selection criteria, this part of the study proposes a “physiological” adjustment, which only takes into account the steepest anchor interval of each ascent phase (see Fig. 1 (∙∙∙∙○), faster anchor interval). It has been hypothesized that the rate of ascend of the RR interval series would change according to vagal stimulus intensity, determining slope variations. Thus, a strongest vagal stimulus would determine a fastest anchor interval and vice versa, potentially affecting the synchronization output. Additionally, by selecting only the fastest anchor point in each ascend
Table 1 Anthropometric and demographic characteristics (mean ± SD) of the subjects who participated in the study.
Age (years) BMI (kg/m2) BSA (m2) APTD (cm) LLTD (cm) METs
Controls
Athletes
29.4 23.6 1.8 21.1 27.7 8.1
23.7 21.0 1.8 21.1 28.4 19.4
± ± ± ± ± ±
5.3 3.9 0.2 2.2 3.3 2.0
± ± ± ± ± ±
7.1 1.9 0.2 1.3 1.6 1.6⁎
Data are reported as means ± SD for 10 subjects in each group. BMI = body mass index; BSA = body surface area; APTD = anteroposterior thoracic diameter; LLTD = laterolateral thoracic diameter; METs = metabolic equivalents. ⁎ p = 0.001.
Fig. 1. RR interval samples derived from an ECG recording. Deceleration anchor intervals are represented as (—◇) slower or (∙∙∙∙○) faster. Dotted line identified the steepest slope, during respective RR interval ascending phase.
308
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
The studied algorithms include two modifications with respect to standard PRSA method, both procedures contribution were addressed separately: (1) standard PRSA method (Standard DC); (2) PRSA procedure by selecting anchor point in deceleration phase with the highest derivative (Anchor Selection); (3) modified DC index to measure the velocity of change in the PRSA transient state (Modified Index); and (4) combination of Anchor Selection and Modified Index (called as DC-faster). The DC values have been compared to classical cardiac vagal modulation indexes—(5) RMSSD = root mean square values of the successive RR interval differences and (6) HF = high-frequency power estimation in a power spectrum band ranging from 0.15 to 0.4 Hz [4,5]—and, additionally, had their Pearson's coefficient of correlation (r) values calculated.
Results Fig. 2. Example of an athlete phase-rectified signal-averaged windows. Deceleration capacity index was calculated using all anchor (a) and only faster anchor (b) intervals. Note that all intervals were averaged (black line) and then employed to calculate the following: (a) Standard DC index (DC = [RR(0) + RR(1) − RR(− 1) − RR(− 2)]/4); (b) Modified Index (arrow slope: θ = Δy/Δx). See text for details.
phase the jittering of the averaging process is minimized [14], improving segment alignment. Modified DC index measurement (Modified Index) The second step was to test a modified DC index over the final PRSA curve, which measures the velocity of change in the transient state (synchronized anchor interval). In this case, the index was calculated by the maximum derivative during transient state (transient analysis) (Fig. 2b) and has neglected its quasi-periodicity, whereas standard DC index is determined by the difference between initial and final states (Fig. 2a). Statistical analysis It was hypothesized that the DC index, which expresses vagal modulation of the heart, was directly related to aerobic conditioning. A threshold for optimal group separation was defined and the area under the receiver operating characteristic curve (AUC) was calculated for all methods. The hypotheses of normality of the variables were rejected after skewness coefficient analysis. Thus, nonparametric Mann– Whitney tests compared groups. Within each group, the comparison was carried out by Wilcoxon test. The AUC values were recalculated 10,000 times, using a bootstrap method to estimate the overall statistical distribution so that an average could be calculated. Each resampling was carried out randomly with replacement and uniform distribution. The bootstrap was executed while controlling the minimum number of subjects in each group (10 subjects). The comparison of methods was carried out by AUC (Cstatistic). Continuous variables were reported as median (median absolute deviation), except for AUC bootstrap, where data were expressed as mean ± standard deviation.
Performance of DC The comparisons of the different methods in the control and athletes groups were summarized in Table 2. The median values of Standard DC and Modified Index for the control group were 11.80 (2.08) and 25.98 (14.18) ms, respectively; while the corresponding values for the athletes group were 17.94 (6.14) and 45.62 (33.82) ms, respectively. Comparing to Standard DC, only the values of Anchor Selection, for both groups, were not significantly different. In both groups, DC-faster was the highest, and Standard DC is the lowest among the DC indexes (p b 0.05). Concerning the AUC values for all methods, the Cstatistic of the standard DC (0.70 ± 0.12) was significantly different (p b 0.01) from both Modified Index (0.96 ± 0.04) and DC-faster (0.91 ± 0.07) (Fig. 3). Comparison between DC indexes and classical methods The comparison showed no significant difference between Modified Index, DC-faster and the classical cardiac modulation methods (Fig. 3). Table 3 shows the Pearson's coefficient of correlation (r) values for all DC indexes vs. classical RMSSD and HF parameters (Table 3). Table 2 Deceleration capacities (DC) indexes and classical cardiac vagal modulation methods (RMSSD and HF) after 10,000 bootstraps.
1. 2. 3. 4. 5. 6.
Standard DC (ms) Anchor Selection (ms) Modified Index (ms) DC-faster (2 + 3) (ms) RMSSD (ms) HF (ms2)
Controls
Athletes
11.80 (2.08) 13.89 (4.73) 25.98 (14.18)⁎ 35.69 (23.90)⁎ 27.60 (15.81)⁎ 42.92 (31.13)⁎
17.94 (6.14) 19.49 (8.62) 45.62 (33.82)⁎,⁎⁎ 59.93 (48.13)⁎,⁎⁎ 51.25 (39.45)⁎,⁎⁎ 62.32 (50.52)⁎,⁎⁎
Data as median (median absolute deviation). (2 + 3) = combination of Anchor Section (2) and Modified Index (3); RMSSD = root mean square values of the successive RR interval differences (time domain); HF = high-frequency band ranging from 0.15 to 0.4 Hz in a power spectrum (frequency domain). ⁎ p b 0.05, comparison with standard DC. ⁎⁎ p b 0.05, controls vs. athletes.
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
309
subjects without cardiac diseases (reported standard DC [mean ± SD]: lower fitness = 7.9 ± 3.8 ms vs. higher fitness = 9.7 ± 3.4 ms, p b 0.05). Nonetheless, a positive relationship is postulated for standard HRV metrics and physical conditioning [1,2]. Effect of anchor point selection algorithm (Anchor Selection)
Fig. 3. Area under the ROC curve (AUC) of all methods after 10,000 bootstraps, and two-by-two comparison. Data presented as mean ± SD. DCfaster (2 + 3) = combination of Anchor Selection (2) and Modified Index (3) (see text for details).
Discussion This study showed that an alternative PRSA algorithm adaptation with a modified derivative index to measure the velocity of change in the transient state (synchronized anchor interval) improved athletes and sedentary controls classification when compared to standard procedure. The advantages of Modified Index measurement over standard method are its simpler algorithm and its ability to assess the physiological response of the heart to autonomic nervous system input. Thus, the proposed derivative index is focused over the transient state, neglecting quasi-periodic oscillations prior to and after anchor points. State of the art Bauer et al. [6] have shown that DC index was reduced in high-risk post-infarction subjects and that the reduction in DC in 24-h ECG recordings was the best ECG-based noninvasive predictor of mortality. Recently, Guzik et al. [15] have shown that DC was reduced also in resting 10-min ECGs recordings of type 1 diabetes patients. Additionally, DC showed high correlation with standard HRV variables, and has been reported to decline with age in both health [3,13,16] and disease subjects [8,9,17]. According to McNarry and Lewis [16], the decline in HRV has been strongly related to age advancement itself, and aerobic fitness has apparently a small contribution to determining the level of HRV. Their study has been the first to relate PRSAderived variables to aerobic fitness in a group of health Table 3 Pearson's coefficient of correlation table of deceleration capacities (DC) indexes vs. classical cardiac vagal modulation methods (RMSSD and HF).
1. Standard DC 2. Anchor Selection 3. Modified Index 4. DC-faster (2 + 3)
RMSSD
HF
0.39 0.28 0.92 0.88
0.56 0.46 0.85 0.86
DC-faster (2 + 3) = combination of Anchor Selection (2) and Modified Index (3); RMSSD = root mean square values of the successive RR interval differences (time domain); HF = high-frequency band ranging from 0.15 to 0.4 Hz in the power spectrum of 15-min normal RR interval series (frequency domain).
By applying the PRSA method in a population with cardiovascular risk, two studies showed significant improvement in risk stratification, through different selection criteria of anchors intervals. Kantelhardt et al. [8], defined the anchor points by comparing sums of “n” values of the time RR series before and after the anchor point candidate (RR(0)), and Pan et al. [9] employed a sinusoidal signal analysis to address the problem of anchor point selection criteria on PRSA processing. Thus, both studies investigated the effect of steeper anchor interval selection in a diseased population, and suggested that removing pseudo-anchor points brings about a better outcome for PRSA average. The Anchor Selection algorithm, which selects anchor point of PRSA in deceleration phase (RR series) with the highest derivative showed a limited effect on discriminating athletes from controls. This may have occurred because of the large number of discarded intervals combined to the short-term segment recordings. As a consequence of RR intervals analysis protocol, the percentage of anchor points remaining for analysis in 15-min ECG segments reduced, on average, from 48.2% using all anchor points to 25.1% using only faster anchor points (number of RR intervals: before analysis protocols: 930 ± 154; all DC anchor points: 452 ± 100; only DC-faster anchor points: 232 ± 47). Performance of Modified Index In the present study, Modified Index algorithm adaptation was able to improve classification of healthy controls from athletes as compared to standard DC. In addition to showing the highest AUC values among all methods tested, the Modified Index also showed (Fig. 3) the lowest standard deviation, thus featuring best reproducibility. Table 3 present coefficient of correlation (r) values for all DC indexes vs. classical cardiac vagal modulation variables (RMSSD and HF). Both DC-faster and Modified Index, which measured the velocity of change in the PRSA curve, showed the highest correlation with classic vagal-mediated variables. This finding indicates that the vagal contribution to HRV as expressed in DC method is bound to the slope of ascent of the PRSA. To the best of our knowledge, this is the first report of this correlation. Limitation of the study Study limitations include the following: (i) small sample size, (ii) two physiologically well-defined groups, and (iii) VO2MAX was estimated indirectly according to the 12-min Cooper test [1]. Although a high correlation with spirometrically assessed VO2MAX has been demonstrated, a direct oxygen consumption measurement may be necessary to validate the present results [18].
310
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
The original DC index was developed for 24-h Holter recordings and, to suppress errors due to artifacts; RR interval prolongations of more than 5% are excluded [6]. In this study, less restrictive criteria of exclusion (RR interval prolongations N 20%) were adopted. Notwithstanding, this restriction criteria have already been validated [12], and applied [13,15] to compensate the utilization of short-term recordings. It has already been demonstrated that a significant correlation exists between the physical conditioning and DC index [16]. In the present study, the proposed index was applied to evaluate aerobic fitness. Nevertheless, it should be emphasized that the present method has not been tested as a risk stratification tool for clinical conditions, particularly myocardial infarction. Further studies are required to demonstrate its potential clinical utility. The advantages of this approach in assessing physical fitness are its non-invasive nature, lower cost, and an intrinsic lower risk for unexpected arrhythmic events. However, the ability of the method in assisting conventional exercise stress tests needs further demonstration.
Conclusion The modified deceleration capacity index (Modified index), which focuses on the transient state, neglecting quasi-periodic oscillations prior to and after anchor points, appropriately classifies athletes and healthy sedentary subjects and improves the AUC diagnostic accuracy of deceleration capacity index to assess physical conditioning. Acknowledgment This work was partially supported by the Brazilian Research Council (CNPq/MCTI) and PROEX Grant (CAPES/MEC). References [1] Marocolo M, Nadal J, Benchimol-Barbosa PR. The effect of an aerobic training program on the electrical remodeling of heart high-frequency components of the signal-averaged electrocardiogram is a predictor of the maximal aerobic power. Braz J Med Biol Res 2007;40:199–208.
[2] Melanson EL, Freedson PS. The effect of endurance training on resting HR variability in sedentary adult males. Eur J Appl Physiol 2001; 85:442–9. [3] Trevizani GA, Benchimol-Barbosa PR, Nadal J. Effects of age and aerobic fitness on heart rate recovery in adult men. Arq Bras Cardiol 2012;99:802–10. [4] Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Heart rate variability. Standards of measurement, physiological interpretation, and clinical use. Eur Heart J 1996;17:354–81. [5] Rajendra AU, Paul JK, Kannathal N, et al. Heart rate variability: a review. Med Biol Eng Comput 2006;44:1031–51. [6] Bauer A, Kantelhardt JW, Barthel P, et al. Deceleration capacity of heart rate as a predictor of mortality after myocardial infarction: cohort study. Lancet 2006;367:1674–81. [7] Bauer A, Kantelhardt JW, Bunde A, et al. Phase-rectified signal averaging detects quasi-periodicities in nonstationary data. Physica A 2006;364:423–34. [8] Kantelhardt JW, Bauer A, Schumann AY, et al. Phase-rectified signal averaging for the detection of quasi-periodicities detection of quasiperiodicities. Chaos 2007;17:015112. [9] Pan Q, Gong Y, Gong S, et al. Enhancing the deceleration capacity index of heart rate by modified-phase-rectified signal averaging. Med Biol Eng Comput 2010;48:399–405. [10] Lipsitz LA, Mietus J, Moody GB, Goldberger AL. Spectral characteristics of HR variability before and during postural tilt. Relations to aging and risk of syncope. Circulation 1990;81:1803–10. [11] Barbosa PR, Barbosa-Filho J, De Sá CA, et al. Reduction of electromyographic noise in the signal-averaged electrocardiogram by spectral decomposition. IEEE Trans Biomed Eng 2003;50:114–7. [12] Clifford GD, McSharry PE, Tarassenko L. Characterizing artifact in the normal human 24-hour RR time series to aid identification and artificial replication of circadian variations in human beat to beat heart rate using a simple threshold. Comput Cardiol 2002;29:129–32. [13] Campana LM, Owens RL, Clifford GD, et al. Phase-rectified signal averaging as a sensitive index of autonomic changes with aging. J Appl Physiol 2010;108:1668–73. [14] Uijen GJH, Weerd JPC, Vendrik AJH. Accuracy of QRS detection in relation to the analysis of high-frequency components in the electrocardiogram. Med Biol Eng Comput 1979;17:492–502. [15] Guzik P, Piskorski J, Contreras P, Migliaro ER. Asymmetrical properties of heart rate variability in type 1 diabetes. Clin Auton Res 2010;20:255–7. [16] McNarry MA, Lewis MJ. Interaction between age and aerobic fitness in determining HR dynamics. Physiol Meas 2012;33:901–14. [17] Ricca-Mallada R, Migliaro ER, Piskorski J, Guzik P. Exercise training slows down HR and improves deceleration and acceleration capacity in patients with heart failure. J Electrocardiol 2012;45:214–9. [18] Cooper KH. A means of assessing maximal oxygen intake. Correlation between field and treadmill testing. JAMA 1968;203:201–4.
ScienceDirect Journal of Electrocardiology 47 (2014) 306 – 310 www.jecgonline.com
Refining the deceleration capacity index in phase-rectified signal averaging to assess physical conditioning level☆ Olivassé Nasario-Junior, DSc, a Paulo Roberto Benchimol-Barbosa, MD, DSc, b,⁎ Jurandir Nadal, DSc a a
Programa de Engenharia Biomédica, COPPE, Universidade Federal do Rio de Janeiro, Rio de Janeiro, RJ, Brazil b Hospital Universitário Pedro Ernesto, Universidade do Estado do Rio de Janeiro, Rio de Janeiro, RJ, Brazil
Abstract
Background: Deceleration capacity (DC) of heart rate is a measure of cardiac vagal modulation. This study introduced a DC adaptation (Modified Index) that measured the velocity of change in the phase-rectified signal averaging curve, and assessed its ability to discriminate athletes from controls. Materials and methods: The Modified Index was compared to Standard DC approach in a prospective case–control study. Subjects were classified according to maximal metabolic equivalents as the control group (CG) and athlete group (AG). The Modified Index was compared to Standard DC and classical approaches (RMSSD and HF) by the area under receiver operating characteristic curve (AUC) using 10,000 bootstraps. Results: In Standard DC and Modified Index bootstrap median values were (ms), respectively, 11.80 and 17.94 (p b 0.01) in CG, and 25.98 and 45.62 in AG (p b 0.01). AUC (mean ± SD) was 0.70 ± 0.12 for Standard DC and 0.96 ± 0.04 for Modified Index (p b 0.01). Conclusions: Modified Index appropriately discriminates athletes from healthy sedentary subjects. © 2014 Elsevier Inc. All rights reserved.
Keywords:
HR variability; Phase-rectified signal averaged; Cardiac vagal modulation; Deceleration capacity index; Physical conditioning
Introduction Regular aerobic exercise provides beneficial changes on the cardiovascular system, characterized by mechanical, autonomic and electrophysiological remodeling [1]. Autonomic remodeling is evidenced by both resting heart rate (HR) reduction and cardiac vagal modulation increase [2,3]. HR variability (HRV) is a non-invasive diagnostic tool employed to assess autonomic nervous system (ANS) activity. HRV is modulated by both sympathetic and vagal efferent and afferents neural pathways over sinus and AV nodes [4,5]. Recently, a novel index termed deceleration capacity (DC) of HR has been proposed to measure cardiac vagal modulation. By detecting anchor points, which directs a phase-rectified signal averaging (PRSA) procedure, this
☆ Each author has contributed to read and approved the manuscript; and none of the authors have any conflict of interest. The manuscript is original, and any part of it has not been and will not be submitted elsewhere for publication. ⁎ Corresponding author at: Boulevard Vinte e Oito de Setembro, 77 Board of Directors Suite, Rio de Janeiro 20551-030, Brazil. E-mail address: [email protected]
0022-0736/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jelectrocard.2013.12.006
technique eliminates noises that permeate RR interval series and yields a quasi-periodic oscillation from a non-stationary signal [6,7]. The averaging of RR interval data segments is synchronized by anchor points, which allows internal quasiperiodic components enhancement, and non-stationary parts abated. However, some anchor points may lack adequate synchronization because of a failure of the classical approach to analyze slope variations of successive RR intervals [8,9]. It was hypothesized that depending on vagal stimulus intensity, the rate of ascend of the RR interval series would change accordingly, determining slope variation. Thus, a strongest vagal stimulus would determine a steepest slope and vice versa, potentially affecting the synchronization output. Although, PRSA has been originally developed to risk stratify subjects post-myocardial infarction [6], its application in assessing physiological conditions strongly related to vagal activity modulation is attractive. Thus, the present study proposes a modified DC index (algorithm adaptations) that (i) selects anchor point of PRSA in deceleration phase with the highest derivative and (ii) measures the velocity of change in the PRSA curve, then tests both classical and proposed methods to distinguish athletes from sedentary healthy volunteers.
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
Methods Study population The analyzed signals were extracted from an existing high-resolution ECG database as described previously [1]. The study protocol was approved by a local ethics committee and informed consent was obtained from each volunteer. Ten elite runners ([mean ± SD] 8.9 ± 3.2 years of training; 6 to 8 training sessions/wk; 90 to 120 min/session; 90 to 110 km/ wk) were enrolled (athlete group). A group of 10 healthy sedentary volunteers were included as control (control group). Inclusion criteria, physical assessment procedures and maximal oxygen consumption (VO2MAX) estimation protocol have been published elsewhere [1]. VO2MAX was divided by the constant 3.5 mL kg − 1 min − 1 to be converted into metabolic equivalents (METs). The control and athlete groups were separated according to estimated VO2MAX, by arbitrarily defining as less than 11.5 METs for sedentary controls and more than 16.0 METs for athletes. Both groups were matched by gender and anthropometric data (Table 1). The aim of anthropometric data intergroup matching was to minimize intergroup physiological and anthropometric variability, thus reducing potential effect of thoracic geometry on surface ECG signals. ECG was acquired shortly after application of the questionnaire and physical examination. Before a 15-min continuous signal acquisition, subjects remained in the supine position for 5 min for stabilization of autonomic modulation after the change from the orthostatic position [10]. Detailed description of the processing and equipment was previously described [11]. The detection of the RR intervals, PRSA algorithm and analysis of the results were performed using a custom-made program in Matlab language. PRSA methods (Standard DC) The PRSA technique has been described elsewhere [6–9]. In the RR interval series, the first step was to identify each of the data points as deceleration anchor. If a particular RR interval increased relative to the previous one, it is identified as a deceleration anchor interval (represented by circle or lozenge symbols in Fig. 1). Any RR intervals that exhibited more than a 20% change from the previous RR interval were
307
excluded, as they are likely to be related to measurement noise or ectopic beats [12]. After determining the deceleration anchors, a window surrounding each anchor is created. The window is defined by the two intervals immediately preceding and following the anchor interval. All of the deceleration windows are then aligned at the anchor point (phase rectified). Once aligned, the respective intervals are averaged together. Once averaged, the DC is calculated by: DC ¼ ½RRð0Þ þ RRð1Þ−RRð−1Þ−RRð−2Þ=4
ð1Þ
where RR(0) is defined as the anchor point (Fig. 2a). Data segments synchronized by anchor interval were averaged so that the internal quasi-periodic components were enhanced and non-stationary parts of the signals were filtered off. However, according to Pan et al. [9], generally, the anchor points could be categorized into two types, as illustrated in Fig. 1. The “type I” anchor points indicated a steeper slope, while the “type II” anchor points are uncertain to be on increasing or decreasing trend. Hence, it remained a doubt whether slope variations of successive RR intervals anchor points may lack adequate synchronization, potentially damaging the PRSA method. PRSA procedure by synchronizing faster RR intervals (Anchor Selection) A couple of studies showed significant improvement in cardiovascular risk stratification for PRSA variables, through alternative selection criteria of anchors intervals [8,9]. To further optimize selection criteria, this part of the study proposes a “physiological” adjustment, which only takes into account the steepest anchor interval of each ascent phase (see Fig. 1 (∙∙∙∙○), faster anchor interval). It has been hypothesized that the rate of ascend of the RR interval series would change according to vagal stimulus intensity, determining slope variations. Thus, a strongest vagal stimulus would determine a fastest anchor interval and vice versa, potentially affecting the synchronization output. Additionally, by selecting only the fastest anchor point in each ascend
Table 1 Anthropometric and demographic characteristics (mean ± SD) of the subjects who participated in the study.
Age (years) BMI (kg/m2) BSA (m2) APTD (cm) LLTD (cm) METs
Controls
Athletes
29.4 23.6 1.8 21.1 27.7 8.1
23.7 21.0 1.8 21.1 28.4 19.4
± ± ± ± ± ±
5.3 3.9 0.2 2.2 3.3 2.0
± ± ± ± ± ±
7.1 1.9 0.2 1.3 1.6 1.6⁎
Data are reported as means ± SD for 10 subjects in each group. BMI = body mass index; BSA = body surface area; APTD = anteroposterior thoracic diameter; LLTD = laterolateral thoracic diameter; METs = metabolic equivalents. ⁎ p = 0.001.
Fig. 1. RR interval samples derived from an ECG recording. Deceleration anchor intervals are represented as (—◇) slower or (∙∙∙∙○) faster. Dotted line identified the steepest slope, during respective RR interval ascending phase.
308
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
The studied algorithms include two modifications with respect to standard PRSA method, both procedures contribution were addressed separately: (1) standard PRSA method (Standard DC); (2) PRSA procedure by selecting anchor point in deceleration phase with the highest derivative (Anchor Selection); (3) modified DC index to measure the velocity of change in the PRSA transient state (Modified Index); and (4) combination of Anchor Selection and Modified Index (called as DC-faster). The DC values have been compared to classical cardiac vagal modulation indexes—(5) RMSSD = root mean square values of the successive RR interval differences and (6) HF = high-frequency power estimation in a power spectrum band ranging from 0.15 to 0.4 Hz [4,5]—and, additionally, had their Pearson's coefficient of correlation (r) values calculated.
Results Fig. 2. Example of an athlete phase-rectified signal-averaged windows. Deceleration capacity index was calculated using all anchor (a) and only faster anchor (b) intervals. Note that all intervals were averaged (black line) and then employed to calculate the following: (a) Standard DC index (DC = [RR(0) + RR(1) − RR(− 1) − RR(− 2)]/4); (b) Modified Index (arrow slope: θ = Δy/Δx). See text for details.
phase the jittering of the averaging process is minimized [14], improving segment alignment. Modified DC index measurement (Modified Index) The second step was to test a modified DC index over the final PRSA curve, which measures the velocity of change in the transient state (synchronized anchor interval). In this case, the index was calculated by the maximum derivative during transient state (transient analysis) (Fig. 2b) and has neglected its quasi-periodicity, whereas standard DC index is determined by the difference between initial and final states (Fig. 2a). Statistical analysis It was hypothesized that the DC index, which expresses vagal modulation of the heart, was directly related to aerobic conditioning. A threshold for optimal group separation was defined and the area under the receiver operating characteristic curve (AUC) was calculated for all methods. The hypotheses of normality of the variables were rejected after skewness coefficient analysis. Thus, nonparametric Mann– Whitney tests compared groups. Within each group, the comparison was carried out by Wilcoxon test. The AUC values were recalculated 10,000 times, using a bootstrap method to estimate the overall statistical distribution so that an average could be calculated. Each resampling was carried out randomly with replacement and uniform distribution. The bootstrap was executed while controlling the minimum number of subjects in each group (10 subjects). The comparison of methods was carried out by AUC (Cstatistic). Continuous variables were reported as median (median absolute deviation), except for AUC bootstrap, where data were expressed as mean ± standard deviation.
Performance of DC The comparisons of the different methods in the control and athletes groups were summarized in Table 2. The median values of Standard DC and Modified Index for the control group were 11.80 (2.08) and 25.98 (14.18) ms, respectively; while the corresponding values for the athletes group were 17.94 (6.14) and 45.62 (33.82) ms, respectively. Comparing to Standard DC, only the values of Anchor Selection, for both groups, were not significantly different. In both groups, DC-faster was the highest, and Standard DC is the lowest among the DC indexes (p b 0.05). Concerning the AUC values for all methods, the Cstatistic of the standard DC (0.70 ± 0.12) was significantly different (p b 0.01) from both Modified Index (0.96 ± 0.04) and DC-faster (0.91 ± 0.07) (Fig. 3). Comparison between DC indexes and classical methods The comparison showed no significant difference between Modified Index, DC-faster and the classical cardiac modulation methods (Fig. 3). Table 3 shows the Pearson's coefficient of correlation (r) values for all DC indexes vs. classical RMSSD and HF parameters (Table 3). Table 2 Deceleration capacities (DC) indexes and classical cardiac vagal modulation methods (RMSSD and HF) after 10,000 bootstraps.
1. 2. 3. 4. 5. 6.
Standard DC (ms) Anchor Selection (ms) Modified Index (ms) DC-faster (2 + 3) (ms) RMSSD (ms) HF (ms2)
Controls
Athletes
11.80 (2.08) 13.89 (4.73) 25.98 (14.18)⁎ 35.69 (23.90)⁎ 27.60 (15.81)⁎ 42.92 (31.13)⁎
17.94 (6.14) 19.49 (8.62) 45.62 (33.82)⁎,⁎⁎ 59.93 (48.13)⁎,⁎⁎ 51.25 (39.45)⁎,⁎⁎ 62.32 (50.52)⁎,⁎⁎
Data as median (median absolute deviation). (2 + 3) = combination of Anchor Section (2) and Modified Index (3); RMSSD = root mean square values of the successive RR interval differences (time domain); HF = high-frequency band ranging from 0.15 to 0.4 Hz in a power spectrum (frequency domain). ⁎ p b 0.05, comparison with standard DC. ⁎⁎ p b 0.05, controls vs. athletes.
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
309
subjects without cardiac diseases (reported standard DC [mean ± SD]: lower fitness = 7.9 ± 3.8 ms vs. higher fitness = 9.7 ± 3.4 ms, p b 0.05). Nonetheless, a positive relationship is postulated for standard HRV metrics and physical conditioning [1,2]. Effect of anchor point selection algorithm (Anchor Selection)
Fig. 3. Area under the ROC curve (AUC) of all methods after 10,000 bootstraps, and two-by-two comparison. Data presented as mean ± SD. DCfaster (2 + 3) = combination of Anchor Selection (2) and Modified Index (3) (see text for details).
Discussion This study showed that an alternative PRSA algorithm adaptation with a modified derivative index to measure the velocity of change in the transient state (synchronized anchor interval) improved athletes and sedentary controls classification when compared to standard procedure. The advantages of Modified Index measurement over standard method are its simpler algorithm and its ability to assess the physiological response of the heart to autonomic nervous system input. Thus, the proposed derivative index is focused over the transient state, neglecting quasi-periodic oscillations prior to and after anchor points. State of the art Bauer et al. [6] have shown that DC index was reduced in high-risk post-infarction subjects and that the reduction in DC in 24-h ECG recordings was the best ECG-based noninvasive predictor of mortality. Recently, Guzik et al. [15] have shown that DC was reduced also in resting 10-min ECGs recordings of type 1 diabetes patients. Additionally, DC showed high correlation with standard HRV variables, and has been reported to decline with age in both health [3,13,16] and disease subjects [8,9,17]. According to McNarry and Lewis [16], the decline in HRV has been strongly related to age advancement itself, and aerobic fitness has apparently a small contribution to determining the level of HRV. Their study has been the first to relate PRSAderived variables to aerobic fitness in a group of health Table 3 Pearson's coefficient of correlation table of deceleration capacities (DC) indexes vs. classical cardiac vagal modulation methods (RMSSD and HF).
1. Standard DC 2. Anchor Selection 3. Modified Index 4. DC-faster (2 + 3)
RMSSD
HF
0.39 0.28 0.92 0.88
0.56 0.46 0.85 0.86
DC-faster (2 + 3) = combination of Anchor Selection (2) and Modified Index (3); RMSSD = root mean square values of the successive RR interval differences (time domain); HF = high-frequency band ranging from 0.15 to 0.4 Hz in the power spectrum of 15-min normal RR interval series (frequency domain).
By applying the PRSA method in a population with cardiovascular risk, two studies showed significant improvement in risk stratification, through different selection criteria of anchors intervals. Kantelhardt et al. [8], defined the anchor points by comparing sums of “n” values of the time RR series before and after the anchor point candidate (RR(0)), and Pan et al. [9] employed a sinusoidal signal analysis to address the problem of anchor point selection criteria on PRSA processing. Thus, both studies investigated the effect of steeper anchor interval selection in a diseased population, and suggested that removing pseudo-anchor points brings about a better outcome for PRSA average. The Anchor Selection algorithm, which selects anchor point of PRSA in deceleration phase (RR series) with the highest derivative showed a limited effect on discriminating athletes from controls. This may have occurred because of the large number of discarded intervals combined to the short-term segment recordings. As a consequence of RR intervals analysis protocol, the percentage of anchor points remaining for analysis in 15-min ECG segments reduced, on average, from 48.2% using all anchor points to 25.1% using only faster anchor points (number of RR intervals: before analysis protocols: 930 ± 154; all DC anchor points: 452 ± 100; only DC-faster anchor points: 232 ± 47). Performance of Modified Index In the present study, Modified Index algorithm adaptation was able to improve classification of healthy controls from athletes as compared to standard DC. In addition to showing the highest AUC values among all methods tested, the Modified Index also showed (Fig. 3) the lowest standard deviation, thus featuring best reproducibility. Table 3 present coefficient of correlation (r) values for all DC indexes vs. classical cardiac vagal modulation variables (RMSSD and HF). Both DC-faster and Modified Index, which measured the velocity of change in the PRSA curve, showed the highest correlation with classic vagal-mediated variables. This finding indicates that the vagal contribution to HRV as expressed in DC method is bound to the slope of ascent of the PRSA. To the best of our knowledge, this is the first report of this correlation. Limitation of the study Study limitations include the following: (i) small sample size, (ii) two physiologically well-defined groups, and (iii) VO2MAX was estimated indirectly according to the 12-min Cooper test [1]. Although a high correlation with spirometrically assessed VO2MAX has been demonstrated, a direct oxygen consumption measurement may be necessary to validate the present results [18].
310
O. Nasario-Junior et al. / Journal of Electrocardiology 47 (2014) 306–310
The original DC index was developed for 24-h Holter recordings and, to suppress errors due to artifacts; RR interval prolongations of more than 5% are excluded [6]. In this study, less restrictive criteria of exclusion (RR interval prolongations N 20%) were adopted. Notwithstanding, this restriction criteria have already been validated [12], and applied [13,15] to compensate the utilization of short-term recordings. It has already been demonstrated that a significant correlation exists between the physical conditioning and DC index [16]. In the present study, the proposed index was applied to evaluate aerobic fitness. Nevertheless, it should be emphasized that the present method has not been tested as a risk stratification tool for clinical conditions, particularly myocardial infarction. Further studies are required to demonstrate its potential clinical utility. The advantages of this approach in assessing physical fitness are its non-invasive nature, lower cost, and an intrinsic lower risk for unexpected arrhythmic events. However, the ability of the method in assisting conventional exercise stress tests needs further demonstration.
Conclusion The modified deceleration capacity index (Modified index), which focuses on the transient state, neglecting quasi-periodic oscillations prior to and after anchor points, appropriately classifies athletes and healthy sedentary subjects and improves the AUC diagnostic accuracy of deceleration capacity index to assess physical conditioning. Acknowledgment This work was partially supported by the Brazilian Research Council (CNPq/MCTI) and PROEX Grant (CAPES/MEC). References [1] Marocolo M, Nadal J, Benchimol-Barbosa PR. The effect of an aerobic training program on the electrical remodeling of heart high-frequency components of the signal-averaged electrocardiogram is a predictor of the maximal aerobic power. Braz J Med Biol Res 2007;40:199–208.
[2] Melanson EL, Freedson PS. The effect of endurance training on resting HR variability in sedentary adult males. Eur J Appl Physiol 2001; 85:442–9. [3] Trevizani GA, Benchimol-Barbosa PR, Nadal J. Effects of age and aerobic fitness on heart rate recovery in adult men. Arq Bras Cardiol 2012;99:802–10. [4] Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Heart rate variability. Standards of measurement, physiological interpretation, and clinical use. Eur Heart J 1996;17:354–81. [5] Rajendra AU, Paul JK, Kannathal N, et al. Heart rate variability: a review. Med Biol Eng Comput 2006;44:1031–51. [6] Bauer A, Kantelhardt JW, Barthel P, et al. Deceleration capacity of heart rate as a predictor of mortality after myocardial infarction: cohort study. Lancet 2006;367:1674–81. [7] Bauer A, Kantelhardt JW, Bunde A, et al. Phase-rectified signal averaging detects quasi-periodicities in nonstationary data. Physica A 2006;364:423–34. [8] Kantelhardt JW, Bauer A, Schumann AY, et al. Phase-rectified signal averaging for the detection of quasi-periodicities detection of quasiperiodicities. Chaos 2007;17:015112. [9] Pan Q, Gong Y, Gong S, et al. Enhancing the deceleration capacity index of heart rate by modified-phase-rectified signal averaging. Med Biol Eng Comput 2010;48:399–405. [10] Lipsitz LA, Mietus J, Moody GB, Goldberger AL. Spectral characteristics of HR variability before and during postural tilt. Relations to aging and risk of syncope. Circulation 1990;81:1803–10. [11] Barbosa PR, Barbosa-Filho J, De Sá CA, et al. Reduction of electromyographic noise in the signal-averaged electrocardiogram by spectral decomposition. IEEE Trans Biomed Eng 2003;50:114–7. [12] Clifford GD, McSharry PE, Tarassenko L. Characterizing artifact in the normal human 24-hour RR time series to aid identification and artificial replication of circadian variations in human beat to beat heart rate using a simple threshold. Comput Cardiol 2002;29:129–32. [13] Campana LM, Owens RL, Clifford GD, et al. Phase-rectified signal averaging as a sensitive index of autonomic changes with aging. J Appl Physiol 2010;108:1668–73. [14] Uijen GJH, Weerd JPC, Vendrik AJH. Accuracy of QRS detection in relation to the analysis of high-frequency components in the electrocardiogram. Med Biol Eng Comput 1979;17:492–502. [15] Guzik P, Piskorski J, Contreras P, Migliaro ER. Asymmetrical properties of heart rate variability in type 1 diabetes. Clin Auton Res 2010;20:255–7. [16] McNarry MA, Lewis MJ. Interaction between age and aerobic fitness in determining HR dynamics. Physiol Meas 2012;33:901–14. [17] Ricca-Mallada R, Migliaro ER, Piskorski J, Guzik P. Exercise training slows down HR and improves deceleration and acceleration capacity in patients with heart failure. J Electrocardiol 2012;45:214–9. [18] Cooper KH. A means of assessing maximal oxygen intake. Correlation between field and treadmill testing. JAMA 1968;203:201–4.