Cardiorespiratory coordination during exercise recovery: a novel measure for health assessment

Óscar Abenza

Lluc Montull

Casimiro Javierre

Natàlia Balagué

Abstract

Cardiorespiratory coordination (CRC), a recently identified biological variable assessing the interaction among parameters derived from cardiopulmonary exercise tests (CPET), has demonstrated a heightened sensitivity to both training effects and functional disorders. Given the critical role of exercise recovery in detecting functional dysregulations, this study aimed to explore CRC during exercise recovery in healthy and physically inactive adults. Fifteen participants underwent a pyramidal CPET performing identical workloads (in inverse order) during the incremental and recovery (decremental) phases. A principal component analysis of selected cardiorespiratory variables was carried out to evaluate CRC. The first principal component eigenvalue and information entropy were calculated. Participants were categorized based on whether they exhibited one or two principal components (1PC and 2PCs groups) during the recovery phase of the CPET. The results revealed that: a) CRC was higher during the recovery phase compared to the incremental phase (t = –2.67; p < .01; d = –0.72), b) the 1PC group displayed higher eigenvalues (t = 3.756; p <.01; d = 2.09) and lower information entropy (U = 0.00; p <.01; d = 15.83) than the 2PCs group, and c) the 1PC group had lower heart rate, ventilation, and rating of perceived exertion at the end of the recovery phase than the 2PCs group (d = 1.21, d = 0.57, d = 0.71, respectively). In conclusion, CRC increased during exercise recovery, and participants with fewer principal components in the recovery phase exhibited greater cardiorespiratory efficiency and better fitness.

Introduction

Cardiopulmonary exercise testing (CPET) is widely applied in clinical practice for the evaluations of the cardiac reserve and the cardiorespiratory fitness of all types of populations (Balady et al., 2010). However, isolated physiological variables (e.g., heart rate [HR], expired minute volume [VE], oxygen consumption [VO₂], etc.) and related parameters (e.g., ventilatory threshold, respiratory exchange ratio, etc.), hardly inform about the dynamic coordination of the cardiorespiratory function during exercise (Balagué et al., 2016; Garcia-Retortillo et al., 2017). 

Cardiorespiratory coordination (CRC), proposed as a variable to measure the interactions among parameters derived from CPET, has been highly useful to sensitively detecting differences across various conditions, such as repeated maximal exercises (Garcia-Retortillo et al., 2017), diverse populations with and without chromosomal disorders (Oviedo et al., 2021), training levels under hypoxia effects (Krivoshchekov et al., 2021), different training regimes (Balagué et al., 2016; Garcia-Retortillo et al., 2019) and dietary manipulations (Esquius et al., 2022). However, its testing and diagnosing possibilities within sport and medical fields remains largely underexplored. 

CRC has been commonly assessed through principal component analysis (PCA), a technique that identifies the underlying covariation patterns among time series data from different cardiorespiratory variables (Balagué et al., 2016). When variables share low covariation, the number of principal components (PCs) increases and vice versa. PCs capture the shared variability between these variables in decreasing order of importance. Thus, the first principal component (PC₁) represents the highest covariation between variables, and it is considered the coordinative variable, whereas each subsequent component (PC₂, PC₃, etc.) captures progressively less shared variance (Balagué et al., 2016). The total number of PCs reflects the level of coordination among the studied cardiorespiratory variables, with fewer PCs suggesting more efficient coordination (Balagué et al., 2016). In contrast, an increase in the number of PCs may indicate the formation of new coordinative patterns (Haken, 2000), and consequently, lower efficiency. This lower efficiency has been identified as a consequence of effort accumulation (Garcia-Retortillo et al., 2017; Garcia-Retortillo et al., 2019), cardiorespiratory disorders (Oviedo et al., 2021), and the effects of training under hypoxia (Krivoshchekov et al., 2021).

As a complement of PCA, the information entropy measure serves as a tool for assessing the complexity of coordinative states within the system (Seely & Macklem, 2012). Higher entropy indicates a more disordered state of the system, requiring a greater amount of information to describe its current state. In contrast, as the system stabilizes to fewer states, its entropy decreases (Naudts, 2005). The sensitivity of CRC to the effects of effort accumulation and general overload appears to offer more relevant information about internal load compared to commonly used variables such as maximal workload or maximal oxygen consumption (Balagué et al., 2016; Garcia-Retortillo et al., 2017).

Common CPET protocols often include incremental and maximal exercises with short resting periods (3 to 10 min) or submaximal active recovery phases at constant workload. However, the evaluation of the exercise recovery period contains crucial information about the adaptivity of regulatory systems during exercise. Exercise recovery represents a dynamic phase in which various activated mechanisms try to return to their initial states (Bartels et al., 2018; Romero et al., 2017). The cardiovascular system, in particular, assumes a crucial role in redistributing blood to satisfy the changing energy and oxygen demands, and rapid post-exercise regulation is crucial for cardiovascular health (Pocock et al., 2013). Therefore, exercise recovery may be considered as a crucial phase for detecting coordinative dysfunctions between the cardiovascular and respiratory systems.

A pyramidal exercise protocol (see Figure 1), which involves progressively increasing and then symmetrically decreasing workloads, may reveal potential dysregulation of cardiorespiratory functions. The gradual reduction in intensity could provide sensitive indicators of recovery efficacy and efficiency of the cardiorespiratory system. While a pyramidal exercise protocol has not yet been employed to assess CRC, it proves particularly valuable for comparing the cardiorespiratory responses to the same workload between incremental and recovery (decremental) phases. Consequently, it allows an assessment of the exercise recovery after being stressed by increasing workloads (Montull et al., 2020). 

Figure 1

See Full Size

This study aimed to explore the potential of CRC assessment during exercise recovery using a submaximal pyramidal protocol for evaluating the cardiorespiratory fitness of healthy and physically inactive adults. Our hypothesis was that the recovery phase would present a reduced number of PCs and lower information entropy compared to the incremental phase. Additionally, individuals with lower cardiorespiratory fitness were expected to demonstrate a higher number of PCs, increased information entropy, and poorer psychobiological recovery during the recovery phase.

Methodology

Participants

Fifteen healthy but inactive (< 30 min of daily physical activity) adults (seven males and eight females: 53.07 ± 3.31 y.o., 169.27 ± 13.26 cm, 80.24 ± 13.26 kg, and BMI 28.43 ± 6.57 kg·m^–2) participated voluntarily in the study. Conducting a power analysis using G*Power 3.1 (Faul et al., 2007) to determine the sample size, and considering large effect sizes of CRC during exercise (Balagué et al., 2016), we estimated a sample size of 12 participants (d = 1, α < .05, power (1 – β) = .80). 

The exclusion criteria were the following: (a) cardiovascular diseases; (b) contraindications to exercise; (c) use of medications that may influence the exercise response of HR. After being informed about the study procedures, participants signed an informed consent before the intervention. Experimental procedures were approved by the Clinical Research Ethics Committee of the local Sports Administration (ref. 07/2015/CEICEGC) and carried out according to the Helsinki Declaration.

Procedures

A pyramidal exercise protocol using a cycle ergometer (Excalibur, Lode, Groningen, Netherlands) was performed (see Figure 1). After a warm-up cycling at 30 W, the workload was increased 30 W/min (males) and 15 W/min (females), until participants reported a Rating of Perceived Exertion (RPE) ≥ 15 (hard) at the Borg’s 6-to-20 scale. At this point, the workload was reduced inversely until 30 W. The cycling frequency was always maintained at 70 rpm. Participants were familiarized with the testing procedures and the RPE (6–20) scale (Borg, 1998) during submaximal incremental exercises at least two times in the month prior to the experiment. 

Data acquisition

During the test, participants breathed through a valve (Hans Rudolph, 2700, Kansas City, MO, USA) and respiratory gas exchange was determined using an automated open-circuit system (Metasys, Brainware, La Valette, France). Oxygen (O₂) content, carbon dioxide (CO₂) content, and air flow rate were recorded breath by breath. The Metasys system currently recorded HR during the same period. Prior to each trial, the system was calibrated using a known composition mixture of O₂ and CO₂ (O₂ 15%, CO₂ 5%, N₂ balanced) (Carburos Metálicos, Barcelona, Spain) and ambient air. Participants were monitored continuously via a 12-lead electrocardiogram (CardioScan v.4.0, DM Software, Stateline, Nevada, USA). 

Testing took place in a well-ventilated lab with a room temperature of 23 ºC and relative humidity of 48%, with minimal variations of no more than 1 ºC in temperature and 10% in relative humidity. Participants carried out the test at least 3 h after a light meal, and without previously practicing vigorous physical activity for 72 h (Balagué et al., 2016).  

Data analysis 

Firstly, data series of the selected cardiorespiratory variables (expired fraction of oxygen ‘FeO₂’, expired fraction of carbon dioxide ‘FeCO₂’, HR and VE) were divided into incremental and recovery phases. Other cardiorespiratory variables commonly used in CPET (e.g., respiratory exchange ratio, oxygen pulse, VO₂, systolic volume, etc.) were excluded from the analysis due to their mathematical relation with the previous ones (Balagué et al., 2016). 

Principal component analysis 

PCA was then performed on the selected cardiorespiratory variables for each phase separately. Prior to perform PCA, Barlett’s sphericity test and Kaiser-Meyer-Olkin index (KMO) were computed for each participant to assess the suitability of applying PCA (Jolliffe, 2002). The number of PCs was determined using the Kaiser-Gutmann criterion, treating PCs with eigenvalues λ ≥ 1.00 as significant (Jolliffe, 2002). The optimal parsimony solution of the extracted PCs was obtained through Varimax orthogonal rotation (Meglen, 1991). 

Given the interest in studying CRC during exercise recovery, two groups were distinguished based on the number of PCs in the recovery phase: participants with 1PC (1PC group; n = 6) and participants with 2 PCs (2PCs group; n = 9). 

Information entropy analysis

To quantify the degree of coordination among the involved cardiorespiratory subsystems of each participant during both phases, the information entropy measure was calculated as follows:

H ± ∑ [1/2 ln (λ) + 1/2 ln (π) + 1/2]

Where H is is the information entropy of the system, λ is the PC’s eigenvalue and π is the Ludolph’s number. This sum included all PC’s eigenvalues of the CPET (e.g., in a test with 2PCs, the sum was conducted over two eigenvalues belonging to PC₁ and PC₂, respectively).

Comparison of incremental and recovery phases

To compare the PC₁ eigenvalues and information entropy between the incremental and recovery phases of the test, Mann-Whitney U test and matched pairs t-test were used, respectively. PC₁ was chosen due to containing the largest proportion of data variance. Additionally, differences in both outcomes between 1PC and 2PCs groups during the recovery phase were assessed using Mann-Whitney U test and independent t-test, respectively. Standardized differences were demonstrated through Cohen’s d, considering medium and large effect sizes (d ≥ 0.5 and d ≥ 0.8, respectively, Cohen, 1988). 

Orthogonal projections of PC₁ during the recovery phase between 1PC and 2PCs groups

A comparison of PC₁ projections of all variables between 1PC and 2PCs groups during the recovery phase was carried out using Mann-Whitney U test and effect sizes (Cohen’s d).

Psychobiological recovery efficacy and performance between 1PC and 2PCs groups

The values of HR, VE, VO₂ and RPE at the end of the recovery phase and the entire test were compared between 1PC and 2PCs groups through Mann-Whitney test and effect sizes (Cohen’s d). 

All statistical analyses were performed using RStudio (RStudio, Inc., 2023), and the predetermined significance level was set at p < .05.

Results

PCA and information entropy of cardiorespiratory variables during incremental and recovery phases

Barlett’s sphericity test presented a highly significant result (p < .001), confirming the suitability of PCA data. The Kaiser-Meyer-Olkin (KMO) index indicated satisfactory sampling adequacy in both the incremental phase (0.60 ± 0.07) and the recovery phase for both groups: 1PC (0.69 ± 0.06) and 2PCs (0.52 ± 0.07). 

During the recovery phase, the percentage of participants exhibiting one PC evidenced an increase of 228% compared to the incremental phase, as shown in Figure 2.

Figure 2

See Full Size

Figure 3 illustrates that the eigenvalues of PC₁, representing the predominant portion of data variance, were slightly greater in the incremental phase (2.57 ± 0.18) than in the recovery phase (2.28 ± 0.51) (d = –0.76). Furthermore, information entropy was also significantly higher during the incremental phase (2.54 ± 0.39) compared to the recovery phase (2.20 ± 0.54) (t = –2.67; p < .01; d = –0.72).

Figure 3

See Full Size

When comparing the two groups within the recovery phase, PC₁ eigenvalues were significantly higher in the 1PC group (2.71 ± 0.32) compared to the 2PCs group (1.99 ± 0.39) (t = 3.756; p < .01; d = 2.09). In the same phase, the 1PC group displayed markedly lower information entropy (1.57 ± 0.06) compared to the 2PCs group (2.62 ± 0.07) (U = 0.00; p < .01; d = 15.83) (see Figure 4).

Figure 4

See Full Size

Orthogonal projections of PC₁ during the recovery phase between 1PC and 2PCs groups

Table 1 reveals that participants in the 1PC group presented relative uniform physiological responses during the recovery phase. Specifically, three cardiorespiratory variables (FeCO₂, HR and VE) displayed high positive projections/values on PC₁, while FeO₂ exhibited negative values. Notably, FeO₂ and FeCO₂ projections onto PC₁ were significantly higher in the 1PC group compared to the 2PCs group (FeO₂: U = 5.00; p < .01; d = 1.99, FeCO₂: d =1.66). In contrast, participants in the 2PCs group did not present uniform cardiorespiratory projections on PC_1.

Table 1

 Orthogonal projections of variables onto PC₁ for participants in both groups during the recovery phase.

See Table

Figure 5 illustrates the differences between the results of 1PC and 2PCs groups in CRC. During the incremental phase, both groups demonstrated similar outcomes, with their variables (FeCO₂, HR, and VE) displaying a higher degree of co-linearity with PC₁, while FeO₂ predominantly aligned with PC₂. However, in the recovery phase, the variance of the four cardiorespiratory variables in the 1PC group was encapsulated by a singular PC₁, whereas the 2PCs group showed three variables (FeO₂ or FeCO₂, HR, and VE) with PC₁ and FeCO₂ or FeCO₂ with PC₂. 

Figure 5

See Full Size

Psychobiological recovery efficacy and performance between 1PC and 2PCs groups

Table 2 shows the values for both groups of HR, VE, VO₂ and RPE at the end of the recovery phase. In comparison to the 2PCs group, the 1PC group presented lower values across all studied variables, with statistically significant differences in HR (U = 7.50; p = .03). Effect sizes indicated intermediate to large magnitudes for HR, VE and RPE (d = 1.21; d = 0.57; d = 0.71, respectively). Furthermore, the total duration of the entire tests by the 1PC group (968.17 ± 179.66), including both incremental and recovery phases, was significantly longer than that of those performed by the 2PCs group (848.22 ± 234.75) (U = 0; p < .01; d = 0.56).

Table 2

Comparison of median and Interquartile Range (IQR) Final Values of HR, VE, VO₂ and RPE between groups.

See Table

Discussion

The main findings of this study were that: a) CRC increased during the recovery phase of a pyramidal CPET, b) participants with 1PC in the recovery phase displayed higher PC₁ eigenvalues and lower information entropy compared to those with 2PCs, and c) HR, VE and RPE values recorded at the end of the CPET were lower in the 1PC group compared to the 2PCs group. 

The reduction in the number of PCs and information entropy during the recovery (decremental) phase, in comparison to the incremental phase of the pyramidal test, suggests a more efficient CRC when the workload decreases. Although this may seem contradictory to previous findings, which indicated impaired CRC after previous exertion (Garcia-Retortillo et al., 2017), the current results can be elucidated by a heightened involvement of anaerobic metabolism during the incremental phase as opposed to the recovery phase. During the incremental phase, the inertia of aerobic metabolism, counterbalanced by the activation of lactic metabolic pathways, produced hyperventilation (Binder et al., 2008; Molkov et al., 2014). This response, which is attributed to the production of non-metabolic CO₂, was plausibly responsible for increasing the number of PCs during the incremental phase, even though the same workloads were performed in the recovery phase. 

Montull et al. (2020) also reported a lack of symmetry in cardiorespiratory values between the incremental and recovery phases of a pyramidal exercise test, with higher values during recovery (decremental phase) for the same workload. In contrast, CRC in this study showed a higher efficiency during the active recovery. These findings confirm the interest of complementing the results of current CPET, such as cardiac reserve and cardiorespiratory fitness, with CRC parameters (i.e., number of PCs and information entropy) (Garcia-Retortillo et al., 2017).  

The higher PC₁ eigenvalues and lower information entropy observed in the 1PC group during the recovery phase, in comparison with the 2PCs group, signifies a higher order, synchronization, efficiency, and adaptability in the cardiorespiratory response (Balagué et al., 2016; Garcia-Retortillo et al., 2017). This implies a more trained cardiorespiratory system and may extend to other interrelated and embedded physiological processes, for example, chemical-sensitive receptors, limbic-hypothalamic system, sympathetic/parasympathetic activity or muscle activity, operating at different timescales to ensure adaptation to workload demands (Garcia-Retortillo & Ivanov, 2022; Kairiukstiene et al., 2020; Pocock et al., 2013; Qammar et al., 2022; Velicka et al., 2019). 

During the recovery phase, the 1PC group demonstrated similar covariation across the four cardiorespiratory variables, with FeO₂ showing elevated values that inversely correlated with FeCO₂, HR, and VE. This sustained elevation in FeO₂, while other variables were recovering, suggests a delayed oxygen demand response (Bahr & Sejersted, 1991). Despite this delay, the 1PC group did not form a new PC, likely due to their superior performance, which resulted in a lower frequency of cardiorespiratory values (i.e., impacting fewer data points) compared to the 2PCs group (Balagué et al., 2016).

In contrast, some participants in the 2PCs group contributed to the formation of PC₂ primarily through FeO₂, likely attributed to heightened hyperventilation at the onset of the recovery phase, indicating less efficient gas exchange. Others in the 2PCs group formed PC₂ through FeCO₂, which suggests they surpassed their ventilatory anaerobic threshold, resulting in substantial hyperventilation induced by excess non-metabolic CO₂ (Binder et al., 2008; Molkov et al., 2014). 

The 1PC group exhibited lower quantitative values of HR, VE and RPE at the end of the recovery phase compared to the 2PCs group, despite undergoing a longer test duration, indicating potentially better workload tolerance. This suggests that participants with superior performance demonstrated not only greater efficiency but also more efficacy in the recovery of the cardiorespiratory system during pyramidal workloads. Indeed, the increased information entropy in the 2PCs group highlights potential dysfunctions in the feedforward and feedback mechanisms mediated by chemoreceptors to regulate ventilation (Zebrowska et al., 2020). This inefficiency may contribute to a more pronounced impairment in the control and regulation of cardiorespiratory function, ultimately leading to earlier exhaustion and higher quantitative values of psychobiological variables. 

The CRC assessment of exercise recovery using pyramidal protocols has relevant clinical implications in CPET. This emphasis on evaluating exercise recovery with gradual changes in workloads provides valuable insights into post-exercise cardiorespiratory regulation, offering information about the internal load stress provoked by the preceding exercise (Bartels et al., 2018; Romero et al., 2017). The application of PCA and information entropy notably demonstrated the potential to inform about the efficiency and efficacy of the cardiorespiratory system in front of both workload increase and decrease, which reinforces these measures as a valuable objective assessment of individuals’ cardiorespiratory fitness during exercise (Balagué et al., 2016; Garcia-Retortillo et al., 2017). This approach introduces new possibilities for the diagnosis and prediction of health and performance states in CPET, including the identification of physiological disorders or pathologies and, most importantly, the prevention of cardiac arrest ((Kairiukstiene et al., 2020; Velicka et al., 2019). 

Dimensional compression techniques, such as PCA, reduce the high dimensionality of time-series data into a few components, offering a more comprehensive understanding of individual dynamics (Denis, 2016). This approach aligns with the perspective that such analyses are more integrative and realistic compared to traditional physiological measures relying on isolated and static quantitative values (Balagué et al., 2020; Garcia-Retortillo et al., 2017). Moreover, it is worth noting that submaximal pyramidal exercises offer a highly relevant information about the state of the cardiorespiratory system and allow avoiding maximal tests, which may pose certain risks, especially in adult and inactive populations. 

This work presented certain methodological limitations and future perspectives of research. The strict inclusion criteria of including only inactive and healthy adults in the sample size limited the statistical significance of the results. Future investigations are warranted to increase this sample size and investigate diverse age groups and varying fitness and health statuses to validate these preliminary findings. Secondly, considering that this study established the workloads based only on RPE, future research should add other objective measurements such as HR or predetermined workloads. Finally, future studies should consider incorporating systolic and diastolic arterial blood pressure into PCA, alongside other cardiorespiratory variables to integrate more relevant variables.

Lastly, although PCA as a linear dimension technique is validated as a valuable and sensitive tool for detecting cardiorespiratory changes during CPET (Garcia-Retortillo et al., 2017), further data analysis techniques should be explored to capture the nonlinear dynamics of CRC. In this sense, not only nonlinear PCA methods (Tenenbaum et al., 2000) may be interesting, but also other analyses promoted by Network Physiology of Exercise (Garcia-Retortillo et al., 2020; Garcia-Retortillo & Ivanov, 2022; Garcia-Retortillo et al., 2024; Montull et al., 2023; Vázquez et al., 2016).

Conclusion

This study demonstrated that cardiorespiratory coordination in healthy inactive adults increased during exercise recovery. Participants with fewer number of principal components in this phase exhibited greater recovery efficiency and efficacy of the cardiorespiratory system. Hence, cardiorespiratory coordination is reinforced as a valuable biological variable for providing integrative and sensitive insights into cardiopulmonary exercise testing and, accordingly, the fitness status. Additionally, the submaximal pyramidal exercise protocol appears to be a suitable tool for assessing adult populations and identifying potential cardiorespiratory dysregulation. 

Acknowledgements

We would like to thank the participants for their willingness to contribute to this research. Additionally, we are grateful to the Department of Physiological Sciences at the University of Barcelona for their assistance in collecting the data.