ArticlePDF Available

Correlation Properties of Heart Rate Variability during a Marathon Race in Recreational Runners: Potential Biomarker of Complex Regulation during Endurance Exercise

July 2021
Journal of Sports Science & Medicine 20(4):557-563

DOI:10.52082/jssm.2021.557

Project: Fractal correlation properties of heart rate variability: Exploration and implementation of a new biomarker for intensity distribution in endurance exercise and training prescription

Authors:

Thomas Gronwald

Medical School Hamburg

Bruce Rogers

University of Central Florida

Laura Hottenrott

University of cooperative education sport and health

Olaf Hoos

University of Wuerzburg

Show all 5 authorsHide

There is only very limited data examining cardiovascular responses in real-world endurance training/competition. The present study examines the influence of a marathon race on non-linear dynamics of heart rate (HR) variability (HRV). Eleven male recreational runners performed a self-paced marathon road race on an almost flat profile. During the race, heart rate and beat-to-beat (RR) intervals were recorded continuously. Besides HRV time-domain measurements, fractal correlation properties using short-term scaling exponent alpha1 of Detrended Fluctuation Analysis (DFA-alpha1) were calculated. The mean finishing time was 3:10:22 ± 0:17:56 h:min:s with a blood lactate concentration of 4.04 ± 1.12 mmol/L at the end of the race. Comparing the beginning to the end segment of the marathon race (Begin vs. End) significant increases could be found for km split time (p < .001, d = .934) and for HR (p = .010, d = .804). Significant decreases could be found for meanRR (p = .013, d = .798) and DFA-alpha1 (p = .003, d = 1.132). DFA-alpha1 showed an appropriate dynamic range throughout the race consisting of both uncorrelated and anti-correlated values. Lactate was consistent with sustained high intensity exercise when measured at the end of the event. Despite the runners slowing after halfway, DFA-alpha1 continued to fall to values seen in the highest intensity domain during incremental exercise testing in agreement with lactate assessment. Therefore, the discrepancy between the reduced running pace with that of the decline of DFA-alpha1, demonstrate the benefit of using this dimensionless HRV index as a biomarker of internal load during exercise over the course of a marathon race.

Graphic view of DFA-alpha1 and km split times over time (mean ± standard deviation).

…

Figures - uploaded by Thomas Gronwald

Content may be subject to copyright.

Content uploaded by Thomas Gronwald

Content may be subject to copyright.

Content uploaded by Thomas Gronwald

Content may be subject to copyright.

©Journal of Sports Science and Medicine (2021) 20, 557-563

http://www.jssm.org DOI: https://doi.org/10.52082/jssm.2021.557

Received: 21 January 2021 / Accepted: 25 June 2021 / Published (online): 10 July 2021

Correlation Properties of Heart Rate Variability during a Marathon Race in

Recreational Runners: Potential Biomarker of Complex Regulation during

Endurance Exercise

Thomas Gronwald 1, Bruce Rogers 2, Laura Hottenrott 3, Olaf Hoos 4 and Kuno Hottenrott 5

1 Department of Performance, Neuroscience, Therapy and Health, Faculty of Health Sciences, MSH Medical School

Hamburg, University of Applied Sciences and Medical University, Am Kaiserkai 1, 20457 Hamburg, Germany; 2 Depart-

ment of Internal Medicine, College of Medicine, University of Central Florida, 6850 Lake Nona Boulevard, Orlando,

Florida, 32827-7408, USA; 3 Institute of performance diagnostics and health promotion, Martin-Luther-University Halle-

Wittenberg, Von-Seckendorff-Platz 2, 06120 Halle (Saale), Germany; 4 Center for Sports and Physical Education, Julius-

Maximilians-University of Wuerzburg, Am Hubland, 97074 Wuerzburg, Germany; 5 Institute of Sports Science, Martin-

Luther-University Halle-Wittenberg, Von-Seckendorff-Platz 2, 06120 Halle (Saale), Germany

Abstract

There is only very limited data examining cardiovascular re-

sponses in real-world endurance training/competition. The pre-

sent study examines the influence of a marathon race on non-lin-

ear dynamics of heart rate (HR) variability (HRV). Eleven male

recreational runners performed a self-paced marathon road race

on an almost flat profile. During the race, heart rate and beat-to-

beat (RR) intervals were recorded continuously. Besides HRV

time-domain measurements, fractal correlation properties using

short-term scaling exponent alpha1 of Detrended Fluctuation

Analysis (DFA-alpha1) were calculated. The mean finishing time

was 3:10:22 ± 0:17:56 h:min:s with a blood lactate concentration

of 4.04 ± 1.12 mmol/L at the end of the race. Comparing the be-

ginning to the end segment of the marathon race (Begin vs. End)

significant increases could be found for km split time (p < .001, d

= .934) and for HR (p = .010, d = .804). Significant decreases

could be found for meanRR (p = .013, d = .798) and DFA-alpha1

(p = .003, d = 1.132). DFA-alpha1 showed an appropriate dy-

namic range throughout the race consisting of both uncorrelated

and anti-correlated values. Lactate was consistent with sustained

high intensity exercise when measured at the end of the event.

Despite the runners slowing after halfway, DFA-alpha1 contin-

ued to fall to values seen in the highest intensity domain during

incremental exercise testing in agreement with lactate assess-

ment. Therefore, the discrepancy between the reduced running

pace with that of the decline of DFA-alpha1, demonstrate the ben-

efit of using this dimensionless HRV index as a biomarker of in-

ternal load during exercise over the course of a marathon race.

Key words: Autonomic nervous system, Detrended Fluctuation

Analysis, non-linear dynamics, endurance exercise.

Introduction

Marathon running has been extensively studied in terms of

underlying cardiovascular physiology, substrate metabo-

lism, biomechanical efficiency, and even motivational fac-

tors (Zinner and Sperlich, 2016). However, although there

are data describing behavior of heart rate (HR) variability

(HRV) during the actual event (Franco et al., 2014), the

specific indices studied have been less than optimal for as-

sessment of real-time exercise intensity because of a lack

of appreciable changes during exercise. Although time-

and frequency-domain related linear HRV measures such

as the standard deviation of all normal beat-to-beat (RR)

intervals SDNN, the root mean square of successive differ-

ences (RMSSD) and high-frequency (HF) power have been

shown to change with exercise intensity, they generally fol-

low a trajectory showing a loss of dynamic range around

the first ventilatory threshold (VT1) (Tulppo et al., 1996;

Cottin et al., 2007; Karapetian et al., 2008).

In this context, non-linear HRV analysis is a prom-

ising approach for internal load monitoring and physio-

logic threshold determination (Gronwald et al., 2020a).

Detrended Fluctuation Analysis (DFA; Peng et al., 1995)

with its dimensionless short-term scaling exponent alpha1

(DFA-alpha1) reveals fractal correlation properties of HR

time series within a range of 0 and 2 and shows clear de-

pendency on changes to organismic demands related to ex-

ternal load markers such as intensity, cadence and duration,

leading to insights in complex regulation during endurance

exercise (for review see Gronwald and Hoos, 2020; for ap-

plication see Gronwald et al., 2020a). A recent investiga-

tion has shown that this non-linear index based on fractal

correlation properties of HRV changes with exercise inten-

sity with the midpoint of its range being near the VT1,

thereby demonstrating potential as a physiological marker

for a wide range of workloads (Gronwald et al., 2020a;

Rogers et al., 2021). Therefore, this approach shows great

potential regarding the investigation of Network Physiol-

ogy of Exercise (NPE) recently introduced by Balagué et

al. (2020). However, despite the growing body of literature

from laboratory studies in this field, HRV data from real-

world endurance training/competition is rarely available

for linear HRV indices (Sztajzel et al., 2006; Cottin et al.,

2007; Franco et al., 2014). For non-linear DFA-alpha1,

Gronwald et al. (2020b) conducted a pilot study during pro-

longed severe to maximal endurance exercise during a

10km road race and found a clear reduction in correlation

properties in RR fluctuations that might be attributed to in-

creased organismic demands. Furthermore, they showed

that a higher performance outcome in terms of a faster total

running time implies the necessity of an earlier loss of cor-

relation properties during the beginning of the exercise reg-

imen, indicating an accelerated alteration of cardiac

Research article

Non-linear HRV during a Marathon

558

autonomic regulation and cardio-pulmonary onset-kinetics

at a higher performance standard. The fact that a self-paced

competitive endurance race performance provides more in-

sights into the complex organismic regulation during se-

vere to maximum intensity exercise, elucidates the need for

further research in this field. In addition, since much of the

current information related to exercise and DFA-alpha1

has been obtained during short-term incremental ramp pro-

tocols (Gronwald and Hoos, 2020), further insights may be

gleaned by observing the behavior over the course of

longer exercise sessions such as a marathon road race.

The purpose of this report is to gain more general

insight into non-linear HRV dynamics during prolonged

endurance exercise competition and to evaluate the poten-

tial of DFA-alpha1 as a marker for internal load and com-

plex regulation during long duration exercise. Most of the

exercise intensity during a marathon race is performed near

aerobic threshold (Zinner and Sperlich, 2016) which is not

considered a strenuous work rate (Jamnick et al., 2020), al-

beit for a very prolonged period of time. However, the

measurement of the running speed at the aerobic threshold

is generally done via short duration constant power or in-

cremental ramp testing (Coen et al., 2001; Faude et al.,

2009). Few studies have examined real time blood lactate

measurements during a long road race, but evidence indi-

cates values between 2 to 3 mmol/L (Santos et al., 2006),

mostly well above typical threshold values for the aerobic

threshold (Faude et al., 2009). Thus, there appears to be a

possible discrepancy between a measure of internal load

intensity (e.g. blood lactate concentration) with that of ex-

ternal load intensity (e.g. running speed at aerobic thresh-

old). Insights into the issue of internal vs. external load

monitoring are possible given the dynamic range of DFA-

alpha1 throughout the spectrum of exercise intensities.

Since direct sampling of blood lactate is not practical dur-

ing a competitive long-distance event, only immediate

post-race values will be obtained. Therefore, non-linear

DFA-alpha1, along with standard linear time-domain

measures of HRV will be examined for potential usefulness

as internal load monitors. We assume that DFA-alpha1 add

complementary information about complex cardiovascular

regulation during marathon racing in addition to the rather

low but steady increases in HR due to cardiovascular drift.

Methods

Participants

Fifteen participants were recruited by newspaper advertise-

ment or personal approach two hours before the marathon

race in the warm-up area and voluntarily participated in the

study. All runners were free from any medical issues. After

being informed extensively about the procedures and ob-

jectives of the study, they provided informed consent in ac-

cordance to the institutional review board and the guide-

lines of the Helsinki World Medical Association Declara-

tion. After exclusion of RR-interval artefacts eleven male

runners (age: 36.6 ± 9.6 years; height: 1.77 ± 0.07 m; body

mass: 70.9 ± 9.8 kg) with a weekly training volume of 50

to 100km, and a personal best marathon time of 2:45 to

3:30 h:min were included in the study analysis.

Study Design

The data were collected during an official marathon race

on a certified asphalt road course (German Road Races As-

sociation) with an almost flat profile (overall elevation be-

tween start and finish line: 80m; temperature: 22 °C, alti-

tude: 282 m above sea level). During exercise, heart rate

(HR) and RR-intervals were recorded continuously during

the race (beat-to-beat-modus) using a HR-monitor with a

time resolution of 1ms (Polar S810, Polar Electro GmbH,

Germany; Kingsley et al., 2005; Nunan et al., 2009). Addi-

tionally, blood lactate (La) and blood glucose (Glu) con-

centration were assessed from each runner immediately

(within the first minute) after crossing the finish line and

were analyzed using an enzymatic-amperometric method

(Super GL ambulance, Dr. Mueller Geraetebau GmbH,

Germany; Faude and Meyer, 2008).

HRV Analysis

Collected raw data were transferred to a PC via an infrared

interface. The NN intervals were stored as ASCII files for

further data analysis. Using Kubios HRV software (Pre-

mium Version 3.4.1, Biosignal Analysis and Medical Im-

aging Group, Kuopio, Finland; Tarvainen et al., 2014), the

HRV analysis was conducted on data collected from 5 min

every 10% (10-100%) of the race for each runner. Raw data

were visually inspected and artefacts were corrected with

the Kubios proprietary automatic artefact correction

method. Only participants whose data contained artefact

rates of <3% were included for analysis. Besides standard

HRV indices obtained from time-domain analysis includ-

ing the mean of normal RR-interval length (meanRR in

ms), the total variability as standard deviation of all normal

RR-intervals (SDNN in ms) and the root mean square of

successive differences (RMSSD in ms; Ciccone et al.,

2017), the scaling behavior for assessing correlation prop-

erties was calculated using the non-linear short-term scal-

ing exponent DFA-alpha1 (window width: 4 ≤ n ≤ 16

beats). DFA-alpha1 values indicate time series correlation

properties, i.e. type of noise: approx. 1.5 for strongly cor-

related Brownian noise and ≤ 0.5 for uncorrelated white

noise with random signals. Approx. 1 signifies a mix of un-

correlated and maximally correlated signal components

with 1/f noise (represents a balance between the complete

unpredictability (randomness) of white noise and the pre-

dictability (strong correlations) of Brownian noise) (Platisa

and Gal, 2008).

Statistical analysis

The statistical analysis was performed with SPSS 23.0

(IBM Statistics, USA) for Windows (Microsoft, USA).

The Shapiro-Wilk test was applied to verify the Gaussian

distribution of the data. The degree of variance homogene-

ity was verified by the Levene test. Subsequently, to ana-

lyze the effects of the exercise bout on dependent variables

(HR, HRV parameters) and 1 km split times between meas-

uring intervals 10% (Begin) and 100% (End) paired sample

T-tests were applied. For all tests, statistical significance

was accepted for p ≤ .05. Cohen’s d was used to denote

effect sizes in comparison of the measurement intervals

(small effect = 0.2, medium effect = 0.5, large effect = 0.8;

Cohen, 1988).

Gronwald et al.

559

Results

The mean finishing time was 3:10:22 ± 0:17:56 h:min:s

(minimum values: 2:43:18 h:min:s; maximum values:

3:33:14 h:min:s) with a blood lactate concentration of 4.04

± 1.12 mmol/L and a blood glucose concentration of 6.15

± 1.21 mmol/L at the end of the race. Comparing the be-

ginning to the end segment of the marathon race (Begin vs.

End) significant increases with large effect size were found

for km split time (p < .001, d = .934) and for HR (p = .010,

d = .804), showing higher HR with slower km split times.

Significant decreases with large effect size were found for

meanRR (p = .013, d = .798) and DFA-alpha1 (p = .003, d

= 1.132). There were no significant changes in SDNN (p =

.802, d = .072) and RMSSD (p = .188, d = .379). Descrip-

tive values (mean ± standard deviation) of all analyzed pa-

rameters during the race (10-100%) are provided in Table

1. In addition, Figure 1 and Figure 2 show the km split

times, HR and DFA-alpha1 over time. Figure 3 provides an

additional view (km split time, HR, DFA-alpha1, RMSSD

and SDNN) of one subject with the time-varying analysis

function of Kubios HRV Software (Version 3.4.1).

Table 1. Km split times, HR, and HRV measures (mean ± standard deviation) during the marathon race (10%-100%).

Parameters 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Km split times

[min:sec]