Abstract
sed heart rate variability (HRV) and high-frequency content of the terminal region of the ventricular activation of signal-averaged ECG (SAECG) have been reported in athletes. The present study investigates HRV and SAECG parameters as predictors of maximal aerobic power (VO 2max) in athletes. HRV, SAECG and VO 2max were determined in 18 high-performance long-distance (25 ± 6 years; 17 males) runners 24 h after a training session. Clinical visits, ECG and VO 2max determination were scheduled for all athletes during thew training period. A group of 18 untrained healthy volunteers matched for age, gender, and body surface area was included as controls. SAECG was acquired in the resting supine position for 15 min and processed to extract average RR interval (Mean-RR) and root mean squared standard deviation (RMSSD) of the difference of two consecutive normal RR intervals. SAECG variables analyzed in the vector magnitude with 40-250 Hz band-pass bi-directional filtering were: total and 40-µV terminal (LAS40) duration of ventricular activation, RMS voltage of total (RMST) and of the 40-ms terminal region of ventricular activation. Linear and multivariate stepwise logistic regressions oriented by inter-group comparisons were ad- justed in significant variables in order to predict VO 2max, with a P < 0.05 considered to be significant. VO 2max correlated significant- ly (P < 0.05) with RMST (r = 0.77), Mean-RR (r = 0.62), RMSSD (r = 0.47), and LAS40 (r = -0.39). RMST was the independent predic- tor of VO 2max. In
by inter-group comparisons were ad- justed in significant variables in order to predict VO 2max, with a P < 0.05 considered to be significant. VO 2max correlated significant- ly (P < 0.05) with RMST (r = 0.77), Mean-RR (r = 0.62), RMSSD (r = 0.47), and LAS40 (r = -0.39). RMST was the independent predic- tor of VO 2max. In athletes, HRV and high-frequency components of the SAECG correlate with VO 2max and the high-frequency content of SAECG is an independent predictor of VO 2max. Correspondence P.R. Benchimol Barbosa Departamento de Cardiologia Universidade Gama Filho Rua Pompeu Loureiro, 36/702 20551-030 Rio de Janeiro, RJ Brasil Fax: +55-21-2255-1409 E-mail: ecgar@yahoo.com Research partially supported by CNPq and CAPES. Received February 7, 2006 Accepted November 13, 2006 Key words •Long-distance runners •Signal-averaged electrocardiogram •Maximal aerobic power •Mathematical model
200 Braz J Med Biol Res 40(2) 2007 M. Marocolo et al. Introduction Regular aerobic exercise training brings about beneficial changes in the cardiovascu- lar system, with impact on individual and community health (1-3). The cardiovascular remodeling that follows aerobic fitness, re- flected by both structural and functional changes of the heart, can be easily assessed by routine clinical examination. Mechanical remodeling following progressive upgrade in training level aiming at maximal aerobic power fitness is characterized by increased cardiac performance and a mild to moderate increase in left ventricular mass, and by increased contractile force of the cardio- myocyte in laboratory animal models (1,4- 7). Mainly characterized by parasympathetic dominance, autonomic remodeling is char- acterized by decreased resting heart rate and increased heart rate variability (HRV) in- dexes. On the other hand, the electrical re- modeling that occurs with both mechanical and autonomic remodeling is characterized by a redistribution of the activation time throughout the ventricular wall, which is expressed on the surface electrocardiogram (ECG) (4-7). The presence of high-frequency compo- nents in the terminal region of the surface ECG has been described on athletes (6,8,9). The nature and the genesis of these compo- nents on the surface ECG in athletes, how- ever, are still subject to investigation, and have been associated with both improve- ment of aerobic training level and incipient ventricular hypertrophy (6,8,9). Conven- tional methods aiming at estimating maxi- mal aerobic power consist of exercise test- ing (either field or ergospirometry) usually lasting 12 to 25 min. The objectives of the present study were: 1) to investigate HRV parameters and high- frequency content of the signal-averaged ECG (SAECG) as predictors of maximal aerobic power in trained athletes and healthy untrained subjects, and 2) to propose a math- ematical model based on HRV and SAECG parameters to classify both trained and un- trained healthy subjects according to maxi- mal aerobic power. Subjects, Material and Methods The study protocol was in accordance with the principles of the Helsinki Declara- tion and was approved by the Ethics Com- mittee in Clinical Research of the Laranjeiras National Institute of Cardiology, Rio
ematical model based on HRV and SAECG parameters to classify both trained and un- trained healthy subjects according to maxi- mal aerobic power. Subjects, Material and Methods The study protocol was in accordance with the principles of the Helsinki Declara- tion and was approved by the Ethics Com- mittee in Clinical Research of the Laranjeiras National Institute of Cardiology, Rio de Ja- neiro, RJ, Brazil, and all subjects provided written informed consent before enrollment. The study population was recruited from January to July 2004 and comprised 36 sub- jects divided into two groups: 18 trained long-distance runners (athletes) and 18 healthy untrained subjects (controls). The sample size used in the present study was calculated from numerical variables as de- scribed by Smith et al. (6) and Raineri et al. (10). From reference data, the sampling pro- cedure was based on the difference of the 40-µV terminal (LAS40, ms) of at least 10 ms, with values of α = 0.05 and ß = 0.1, on the vector magnitude, filtered by a 4-pole bi- directional bandpass Butterworth filter with cut-off frequencies at 40 and 250 Hz, and using XYZ Frank orthogonal leads. The sub- jects participating in the present study were slightly younger but had similar anthropo- metric characteristics and gender distribu- tion as those in the studies used as reference in the sampling procedure (6,10). Inclusion criteria were volunteers over 18 and less than 40 years old, in good mental and physi- cal health, without a previous history of cardiovascular disease or systemic arterial hypertension. Subjects less than 18 and more than 40 years old, with a previous history of diabetes mellitus, thyroid dysfunction or liver disease, alcohol consumers, or tobacco smok- ers, or those currently taking any medicine were excluded. Subjects had their maximal oxygen con- sumption (VO 2max) estimated by the Coo-
201 Braz J Med Biol Res 40(2) 2007 Heart rate variability and signal-averaged ECG in top athletes per 12-min field test (Table 1), and calcu- lated by the following equation (11,12): (Equation 1) where D is the distance achieved in the field test, expressed in meters, and VO 2max is estimated as mL kg -1 min -1 . Next, the result- ing VO 2max was divided by the constant 3.5 mL kg -1 min -1 to be converted into metabolic equivalents (METs). Control and athlete groups were separated according to VO 2max as estimated by the Cooper test, arbitrarily defined as less than 11.5 METs for controls and more than 16.0 METs for athletes. The VO 2max gap between groups was defined a priori with the purpose of enhancing even- tual differences in SAECG parameters caused by conditioning level. The athletes discontinued training 24 h before testing, and fasted for 4 h before signal recording. All subjects replied to a question- naire about age, health condition, use of medi- cation, and physical activities, and their height, weight, anteroposterior and laterolateral tho- racic diameters were measured with the super- vision of a primary care physician. The two groups were matched for age, gender, and body surface area calculated by a nomogram (Table 1) in order to minimize inter-group physiological and anthropometric variability (e.g., chest diameter) and to reduce the poten- tial effect of thoracic geometry on the surface ECG signals. 12-Lead surface ECG acquisition Each subject also had a 12-lead surface ECG acquired with a CardiMax electrocar- diograph (Fukuda Denshi, Seattle, WA, USA) in resting supine position. ECG sig- nals were analyzed for left ventricular hy- pertrophy and assessed by the voltage-dura- tion product, as previously described (13). The voltage-duration product was considered to be abnormal if greater than 2440 mV/s, defining left ventricular hypertrophy. ECG acquisition and signal averaging SAECG was acquired shortly after appli- cation of the questionnaire and physical ex- amination. Before a 15-min continuous sig- nal acquisition, the subjects remained in the supine position for 5 min for stabilization of autonomic modulation after the change from the
was considered to be abnormal if greater than 2440 mV/s, defining left ventricular hypertrophy. ECG acquisition and signal averaging SAECG was acquired shortly after appli- cation of the questionnaire and physical ex- amination. Before a 15-min continuous sig- nal acquisition, the subjects remained in the supine position for 5 min for stabilization of autonomic modulation after the change from the orthostatic position, thus preventing an “autonomic memory” in the acquired sig- nals (14,15). The ECG signals were acquired from the body surface using the modified bipolar Frank XYZ-orthogonal leads and an electro- cardiograph amplifier model AEG03 (Lynx Tecnologia Eletrônica, São Paulo, SP, Bra- zil), with 10-GΩ input impedance, 120-dB/ channel common mode rejection, and a 12- VDC power supply. Each ECG lead was amplified (gain = 1000), analogically fil- tered from 0.05 to 300 Hz by a 3rd order Butterworth filter, and converted into digital format by a data acquisition system with a 14-bit AD converter with a ±10-V dynamic range, and -1000 Hz/channel sampling fre- quency (Lynx Tecnologia). The ECG was digitized and immediately stored in the PC Table 1. Anthropometric and demographic char- acteristics of the subjects who participated in the study. Controls Athletes Age (years) 28.5 ± 5.2 25.0 ± 6.0 Gender (F/M) 1/17 1/17 BSA (m²) 1.75 ± 0.16 1.77 ± 0.15 APTD (cm) 21.1 ± 1.8 21.3 ± 1.1 LLTD (cm) 27.7 ± 2.9 27.9 ± 1.4 VO 2max (METs) 8.4 ± 2.1 19.8 ± 1.4* Data are reported as means ± SD for 18 subjects in each group. BSA = body surface area; APTD = anteroposterior thoracic diameter; LLTD = latero- lateral thoracic diameter; VO 2max = estimated maximum aerobic power; MET = metabolic equivalent (mL kg min -1 ). *P = 0.001 compared to control subjects (unpaired Student t-test). See text for details.
202 Braz J Med Biol Res 40(2) 2007 M. Marocolo et al. for off-line processing. The skin was carefully prepared with a mildly abrasive pad and washed with alco- hol for ECG acquisition, and 6 self-adhesive silver/silver-chloride electrodes were at- tached to the horizontal X-lead electrodes, placed on the right and left mid-axillary lines at the fourth intercostal space. The vertical Y-lead electrodes were placed on the middle clavicle line and second intercos- tal space and costal border, and the sagittal Z lead was placed anteriorly and posteriorly at the level of the 4th intercostal space just to the left of the midline. A seventh ground electrode was placed on the tenth rib on the mid-axillary line. The paired electrode-to- electrode impedance was measured and was less than 7 kΩ for all electrode pairs. ECG acquisition was carried out from 8 am to 1 pm in a quiet and air-conditioned environ- ment, with a temperature of 25ºC. Digital data were processed with pattern recogni- tion software to reject ectopic or excessively noisy beats (16,17). The coherent weighted averaging was carried out independently on each lead using a previously described and validated technique for R-wave alignment and was stopped at a final noise below 0.35 µV (16,17). The SAECG was analyzed in the time domain on the vector magnitude (Vm), defined by: (Equation 2) where X, Y and Z are the averaged orthogo- nal leads after 4-pole bi-directional Butter- worth filtering, with band pass ranging from 40 to 250 Hz. Four parameters were calculated from the vector magnitude (Figure 1): 1) duration of the signal-averaged filtered QRS com- plex (DUR, ms), 2) duration in the terminal region of ventricular activation with a mag- nitude below <40 µV (LAS40, ms), 3) the root mean squared (RMS) value of the po- tentials in the 40-ms terminal of the filtered QRS (RMS40, µV), and 4) the RMS value of the total filtered QRS (RMST, µV). Although terminal portion analysis has been used to identify the risk of ventricular arrhythmias, these parameters have been also tested as markers of cardiac fitness in athletes in gen-
root mean squared (RMS) value of the po- tentials in the 40-ms terminal of the filtered QRS (RMS40, µV), and 4) the RMS value of the total filtered QRS (RMST, µV). Although terminal portion analysis has been used to identify the risk of ventricular arrhythmias, these parameters have been also tested as markers of cardiac fitness in athletes in gen- eral. SAECG was considered abnormal if two of three (DUR, RMS40, LAS40) param- eters were outside the normal range (18). Heart rate variability The raw ECG was processed simulta- neously on the X and Z leads to extract the series of normal RR intervals by rejecting ectopic or extremely noisy beats. The fol- lowing parameters were thus calculated for the analysis of HRV: Mean-RR (ms), the average duration of RR intervals; RRmin (ms), the shortest RR interval value; RRmax (ms), the largest RR interval value; RMSSD (ms), the RMS standard deviation of the difference between two consecutive normal RR intervals. During the analysis of RR interval, rejected beats per acquired ECG ranged from 0.16 to 5.22% with a mean (± SD) of 0.66 ± 0.93%. Statistical analysis Continuous variables are reported as mean Figure 1. Vector magnitude parameters in the time domain. Onset and offset represent the limits of the ventricular activation, automatically detected (see text for details). For abbrevi- ations see legend to Table 2. 0.10 0.05 0 mV DUR = 134.0 ms RMST = 55.7 µV RMS40 = 5.9 µV LAS40 = 68 ms noise = 0.3 µV 0 50 100 150 200 250 300 350 400 450 500 550 ms 40 ms Vector magnitude QRS complex DUR LAS40 RMS40 offsetonsetP-wave 40 µV
203 Braz J Med Biol Res 40(2) 2007 Heart rate variability and signal-averaged ECG in top athletes ± SD. SAECG (RMST, RMS40, DUR, LAS40) and HRV (RMSSD, mean-RR, RRmax, RRmin) parameters had their prob- ability distributions assessed. The RMST and RMS40 variables showed asymmetric probability distributions and were normal- ized by natural logarithm transformation (Lnt) (18,19). The means were compared by the unpaired Student t-test. Categorical vari- ables were compared by the Fisher exact test. Linear regression analysis using the Pearson correlation coefficient (r) was car- ried out between SAECG parameters and VO 2max, as well as between HRV param- eters and VO 2max. The level of significance was set at α = 0.05 in all tests. Data were analyzed using the Statgraphics Plus 5.1 (Manugistic Inc., Rockville, MD, USA), MS Excel 2000 (Microsoft Corporation, Redmond, WA, USA), and Epi-Info 6.04 (Center of Disease Control, Atlanta, GA, USA) software. Mathematical model A multivariate stepwise logistic regres- sion model was adjusted to significant HRV and SAECG variables in order to estimate VO 2max (P < 0.05 in inter-group compari- son). In the logistic model, variables with r² ≥ 0.64 and hierarchically lower signifi- cance by the Wald test were considered to be co-variates and were excluded from the mo- del (exclusion started with highest P value). The final model was achieved when all co- variates were excluded. A binary variable Code, assuming a value of zero for controls and of one for athletes was defined as: (Equation 3) where P is given by: (Equation 4) Table 2. Signal averaged electrocardiogram and heart rate variability of control sub- jects and athletes. Control Athletes Signal-averaged electrocardiogram DUR (ms) 98.40 ± 14.6 98.10 ± 20.5 Lnt (RMST) (Ln[µV]) 1.93 ± 0.14 2.16 ± 0 .15* LAS40 (ms) 31.80 ± 8.1 24.40 ± 9.5* Lnt (RMS40) (Ln[µV]) 1.40 ± 0.21 1.73 ± 0 .31* Beats 329 ± 127 323 ± 109 Final noise (µV) 0.23 ± 0.11 0.25 ± 0.08 Heart rate variability RRmin (ms) 734 ± 165 869 ± 152* Mean-RR (ms) 1034 ± 160 1351 ± 241* RRmax (ms) 881 ± 125 1127
.15* LAS40 (ms) 31.80 ± 8.1 24.40 ± 9.5* Lnt (RMS40) (Ln[µV]) 1.40 ± 0.21 1.73 ± 0 .31* Beats 329 ± 127 323 ± 109 Final noise (µV) 0.23 ± 0.11 0.25 ± 0.08 Heart rate variability RRmin (ms) 734 ± 165 869 ± 152* Mean-RR (ms) 1034 ± 160 1351 ± 241* RRmax (ms) 881 ± 125 1127 ± 194* RMSSD (ms) 39.70 ± 16.0 74.10 ± 36.6* DUR = duration of the signal-averaged filtered QRS complex; Lnt (RMST) = natural logarithm transformation of root mean squared value of the total filtered QRS; LAS40 = duration of potentials below 40 µV at terminal QRS complex; Lnt (RMS40) = natural logarithm transformation of RMS value of the potentials in the 40-ms terminal of the filtered QRS; Beats = number of coherent-averaged beats; Final noise = final noise of coherent averaging; RRmin = the shortest RR interval value; Mean-RR = the average duration of RR intervals; RRmax = the largest RR interval value; RMSSD = root mean squared standard deviation. *P < 0.05 compared to subjects (unpaired Student t-test). where K is a fitting constant and A i the regression constant of each X i variable. A decision-three algorithm based on the final variables of the model was built to stratify VO 2max. The threshold cut-off lev- els of the final variables were detected by logistic regression model analysis, and were not included in the interval range as refer- ence for higher VO 2max estimates. Results The voltage-duration product was not significantly different (1,676.1 ± 711 for athletes vs 1,325.0 ± 548.5 mV/s for con- trols) between groups. An electrical marker of left ventricular hypertrophy was detected in 3 subjects in the athlete group and in no subject in the control group (P = 0.23). The HRV and SAECG parameters of each group and respective inter-group com- parison are presented in Table 2. All vari- ables demonstrated significant differences between the control and athlete groups, ex-
control group (P = 0.23). The HRV and SAECG parameters of each group and respective inter-group com- parison are presented in Table 2. All vari- ables demonstrated significant differences between the control and athlete groups, ex-
204 Braz J Med Biol Res 40(2) 2007 M. Marocolo et al. correlation, however, was found between DUR and VO 2max. The prevalence of abnormal SAECG pa- rameters in the control and athlete groups was 11 and 8.3%, respectively (P = NS). Both RMS40 and LAS40 were abnormal in one athlete and in three control subjects. DUR was within the normal range in all subjects. Mathematical model After excluding the significant co-vari- ates, the variables included in the model were: LAS40, RMST, RMSSD, and Mean- RR. The model showed 88.9% positive pre- dictive, 94.4% negative predictive, and 91.7% total predictive values, explaining 69.33% of total deviance (P < 0.001). RMST was the only independent predictor of VO 2max (Table 3). The threshold cut-off values of the vari- ables involved in the classification of the highest VO 2max were (according to the op- timal dichotomization level as computed by logistic regression model): RMST > 120.9 µV, LAS40 < 28.1 ms, Mean-RR > 1003.8 ms, and RMSSD > 30.3 ms. The decision algorithm that best defined physical conditioning with 72.2% sensitivi- ty, 94.4% specificity, and 83.3% total pre- dictive accuracy was (pseudo-code lan- guage): IF “RMST > 120.9” THEN IF “LAS40 < 28.1” OR “Mean-RR > 1003.8” OR “RMSSD > 30.3” THEN “Athlete” ELSE “Control” where {IF...THEN} pairs represent condition commands in hierarchical order, and quoted parameters the condition codes; {ELSE} com- mand points to the final alternative, and {OR} command allows logical non-excluding as- sociation of condition codes. The quoted “Control” and “Athlete” refers to control and athlete groups, respectively. cept for DUR, for equivalent numbers of coherent-averaged beats and final noise val- ues. Linear correlation analysis was carried out jointly on the control and athlete groups. Positive correlations were observed between VO 2max and Lnt (RMST), and between VO 2max and Lnt (RMS40) (r = 0.77, P < 0.001; r = 0.6, P < 0.001, respectively; Fig- ure 2A and B). The correlation between LAS40 and VO 2max was -0.393 (P = 0.018). These two parameters, Lnt (RMST) and Lnt (RMS40), also showed a positive correla- tion (r = 0.718,
Description
This study investigates HRV and SAECG parameters as predictors of maximal aerobic power in athletes.