1. Introduction
The heart rate (HR) and the R-R interval (RRi) show complex oscillations, which is required for rapid adaptation to different conditions to maintain body homeostasis with adequate operation. Cardiac oscillation can be characterized by heart rate variability (HRV), as a variation between consecutive heartbeats. These parameters are determined by complex interactions between sympathetic and parasympathetic nervous systems, therefore, their repeated detections are valuable and non-invasive tools to evaluate the autonomic nervous system’s (ANS) influence on the heart (Draghici & Taylor, 2016; Rosenwinkel et al., 2001). The root mean square successive difference of RRi (RMSSD) is an important and most commonly used time domain HRV parameter, which is sensitive to short-term high-frequency fluctuation in the HR, and it is mostly mediated by the fast parasympathetic action independently from respiration rate (Ernst, 2017; Plews et al., 2013; Shaffer & Ginsberg, 2017). Several factors can influence these measurements, including age, sex, physical activity, body mass index, respiration, body position, time of recordings, sleep and stress level (Shaffer & Ginsberg, 2017). It is also well-known that HRV is significantly associated with mean HR (and RRi) through the influence of physiological phenomena and mathematical constraints (Sacha & Pluta, 2008). Since changes in cardiac autonomic activity are known to reflect a deviation from homeostasis, the measurement of HRV may be important in the evaluation of the stress level or recovery status in athletes (Lloria-Varella et al., 2023).
Team handball is a professional sport determined by individual performance of each player, the tactical elements and cooperation of the team. It has a complex nature due to the high intensity activities, including sprints, jumps, and sudden direction changes for appropriate performance in the different game situations (Gorostiaga et al., 2006; Leuciuc et al., 2022). Furthermore, each specific position in handball requires unique physiological and physical attributes relating to the post-specific technical and tactical requirements (Vila et al., 2012). Thus, successful execution is highly dependent on several complex and interrelated factors, including anthropometric, cardiorespiratory, neuroendocrine characteristics. Most of the reports investigated the relationship between body composition and HRV values applied cross-sectional study design with single measurements, or they focused primarily to pathologic conditions (Cvijetic et al., 2023; Eyre et al., 2014). Measures of an athlete’s HRV can be applied for the prescription of training. It has been suggested that the frequently registered HRV values can provide an effective biofeedback and it a simple method for improving performance and/or preventing overload-related injuries (Jiménez Morgan & Molina Mora, 2017). The beneficial phenomenon of the HRV analyses is due to the fact that HRV may reflect the training-induced level of stress and overload (Bellenger et al., 2016). Ample studies showed that the heart rate variability in handball players can be changed after different manipulations, including caffeine intake, special exercises, cold water immersion, naps duration (Djientcheu et al., 2024; Kayacan et al., 2023; Klatt et al., 2021; Lopes-Silva et al., 2021; Nishida et al., 2021; Ravier et al., 2022; Ravier & Marcel-Millet, 2020). However, little reports exist for prolonged period on handball players, who are regularly exposed to high training loads, and even no data are available about the connection between RRi and C_RMSSDs values in athletes (Buchheit, 2015).
It is well-known that the standard HRV analysis is mathematically biased because of the nonlinear relationship between RR interval and HR, especially if subjects differ in terms of their mean HR (Billman et al., 2015; Plaza-Florido et al., 2023; Sacha & Pluta, 2005). To solve this problem, while preserving the significance of HRV, corrected HRV values were estimated by division and multiplying the HRV values with different powers of RRi (Gąsior et al., 2018; Sacha et al., 2013; Sacha & Pluta, 2008). The same process was performed for RMSSD in this study, supposing that the corrected RMSSD (C_RMSSD) parameters may be reliable for cardiovascular assessment in this healthy athlete population.
The goal of this study was to clarify the post- and age-dependent influences of anthropometric measures on the heart rate (HR) and its variability (HRV) in young male handball players (14–21 years) during resting condition, and to reveal associations between the cardiovascular parameters by classical and machine learning data analytic methods. The analyses of the large amount of data can provide a general mean value of young handball players in different post and age groups, and the personal analyses of the deviations might help to find the factors leading to the alterations, and to the correction of the training load. This method might provide an effective way for the prevention of physical overload or even psychological burning out.
2. Materials and Methods
85 male handball players were involved for 11 months (between 01 July 2022 and 31 May 2023) in this analytic cohort study. Athletes were informed in advance about the test procedure (and were given the opportunity to practice performing the HRV measurement accurately), and all participants or their parents signed documents providing written informed consent to the study protocol. These procedures were conducted in accordance with the ethical standards of the Declaration of Helsinki and approved by the local research ethics committee under number OGYEI/56606-5.
Participants’ anthropometric data including body weight and height, body composition (fat and muscle amount and percentage, the ratio of fat and muscle amount, and body mass index [BMI]; Table 1) were measured monthly using the Seca mBCA 555 device (Hamburg, Germany).
Table 1
Anthropometric and HRV characteristics at group level.
| PARAMETERS | MEAN | SD |
|---|---|---|
| Body weight (kg) | 81.7 | 12.27 |
| Body height (cm) | 187.0 | 7.74 |
| Muscle amount (kg) | 34.4 | 3.89 |
| Muscle ratio (%) | 42.4 | 2.76 |
| Fat amount (kg) | 12.3 | 6.78 |
| Fat ratio (%) | 14.3 | 6.01 |
| Fat/Muscle ratio (%) | 34.8 | 16.97 |
| BMI | 23.3 | 3.06 |
| HR | 65.8 | 8.84 |
| RRi | 939.4 | 112.97 |
| SDNN | 130.4 | 35.36 |
| pNN50 | 43.1 | 14.74 |
| RMSSD | 88.4 | 30.84 |
| lnRMSSD | 4.4 | 0.38 |
[i] See text for abbreviations.
Participants were organized into subgroups based on their age (A1: ages between 14–15, n = 18; A2: ages between 16–17, n = 34; A3: between 18–19, n = 20; A4: ages between 20–21, n = 13) and their playing positions (goalkeeper n = 13, pivot n = 12, back n = 19, center back n = 20, and wing n = 21). At the time of measurements all participants were in good health and had no major injury during this period.
After a familiarization session, HRV measurements were taken between 6 and 11 A.M. in home circumstances. Participants were required to avoid strenuous activities, caffeine, or other factors that could influence HRV results prior to measurement. Providing this information would ensure that external influences were minimized.The cardiovascular parameters (Table 1) were recorded for 5 min on supine position using a portable heart rate monitor (Polar H10 model; POLAR® Electro Oy, Kempele, Finland; sampling rate: 1000 Hz), which is reliable for routinely detection of these parameters (Schaffarczyk et al., 2022). Time domain HRV analyses were performed using ELITE HRV app (Version 5.5.1; Gloucester, MA), which supports several input data formats for electrocardiogram (ECG), and calculates all time domain HRV parameters (Table 1).
2.1. Data and statistical analyses
Data were involved in the analyses, if the participant had at least 30 data during the investigated period. Altogether 12213 cardiovascular and 513 anthropometric data were available for the 85 players.
Data are presented as mean values and standard deviations in the text, Table 1 and figures. Since most of the parameters did not pass the normality test, nonparametric tests were used (or see below the “Preparatory technics for deep-learning method” paragraph). For comparison the different groups, the Kruskal-Wallis test was performed. To establish the relationship between the different HRV predictors, Spearman correlation analysis was applied.
The analyzed HRV parameters were the standard deviation of RRi (SDNN), the proportion of NN50 (pNN50): the number of pairs of successive RRi that differs by more than 50 ms divided by the total number of RRi, the root mean square of consecutive RRi (RMSSD) and its logarithmic value (lnRMSSD; Table 1). Besides these parameters, corrected values of RMSSD (C_RMSSDs) were also analyzed, which can provide information about the RR independent components of RMSSD, and the correlation analyses were performed between RRi and C_RMSSD values to preclude the mathematical consideration of HR effects (Sacha & Pluta, 2005, 2008). Thus, the C_RMSSD values were build up by dividing and multiplying the RMSSD values by the power 1, 2, 4, 8, 16 of RRi (signed as C/P16, C/P8, C/P4, C/P2, C/P1, C/P0 (the original RMSSD), C*P1, C*P2 C*P4 C*P8, and C*P16; See Figures 3H and 4H). After this conversion, Spearman’s rank-order correlations were performed between RRi and all the different HRV values.
For these analyses Statistica software (TIBCO Statistica®, 14.01.25; Arlington, VA, USA) was applied, and significance was accepted at the p < 0.05 level.
2.2. Data Preprocessing for Machine Learning
In this chapter, we delve into the crucial process of data preprocessing for machine learning. The quality and suitability of data feeding into our machine learning models significantly impacts their performance. To ensure a solid foundation for our analysis various machine learning and data science methods were experimented, including clustering, decision trees, and dimensionality reduction. However, it’s important to note that some of these methods can be sensitive to issues such as skewness, missing values, and extreme outliers. Given the presence of significant variance and skewness in our dataset, our first step was to prepare and clean the data. We began by addressing skewness in the individual characteristics of the dataset including RRi and RMSSD. To achieve this, a two-step process was applied scaling the values and transforming them into a normal distribution using the Box-Cox power transformation. The Box-Cox method is a widely recognized statistical technique designed to convert non-normally distributed data into a normal distribution (Osborne, 2019). This method involves taking the natural logarithm of the original data and transforming it using a maximum likelihood estimated power. The estimated power is determined by the level of skewness in the data. This transformation is known to enhance the accuracy of various data science methods, particularly distance-based clustering algorithms. Our dataset, which represents physiological data over time, takes the form of a time series, as a sequential record of observations. However, to extract meaningful insights or make comparisons, the aggregation or transformation of the data are frequently required. To facilitate such analyses, cumulative attributes were generated for each characteristic after completed the initial data preparation steps.
Even though clustering methods are capable of handling multidimensional data, it can be challenging for humans to comprehend the cross-correlations and visualize data in multidimensional space. To address this issue, various dimension reduction techniques were employed. These techniques effectively reduced the feature space from 10–20 dimensions (RMSSD, mean RMSSD, standard deviation of RMSSD, RRi, mean RRi, etc.) to a more manageable 2D or 3D space without significant loss of information.
For dimensionality reduction, we explored three methods:
Principal Component Analysis (PCA): It is a widely used statistical technique for dimensionality reduction accomplished through linear transformation (Greenacre et al., 2022). It enables us to visualize large datasets of high dimensionality while retaining the maximum amount of information from the original data.
t-Distributed Stochastic Neighbor Embedding (t-SNE): This is a non-linear method that transforms high-dimensional data into lower-dimensional representations, often in 2D or 3D space (Zheng et al., 2022). The primary objective of t-SNE is to preserve both local and global structures within the data.
Uniform Manifold Approximation and Projection (UMAP): It is another non-linear dimension reduction method (Diaz-Papkovich et al., 2021), which shares similar goals with PCA and t-SNE, focusing on preserving local structures and patterns within the data. UMAP represents the data as a distance-weighted graph, where closer neighbors carry more weight in the representation.
2.3. Data Analysis with Machine Learning
Following the rigorous processes of data cleaning, power transformation, and dimension reduction, a comprehensive analysis of our dataset was embarked. The goal was to gain deeper insights into the recorded participants’ data and uncover the underlying “habits” within their multidimensional features. Additionally, the distances between participants were also measured, ultimately providing a comprehensive understanding of the data.
To achieve these goals, the K-Means clustering method was employed, a well-established technique for comparing and measuring the distances between data points in multidimensional space. K-Means is a popular unsupervised machine learning method known for vector quantization (Likas et al., 2003). Its primary purpose is to assign each data point to a cluster with the nearest centroid, effectively minimizing the variance within clusters, while maximizing the differences between them. One limitation of K-Means is that the number of clusters must be predetermined. Therefore, the optimal number of clusters were 3 in our case determined by the elbow method (Syakur et al., 2018).
Based on this optimal cluster number, multiple runs were performed, and their results are illustrated in the Results section (Figure 5). Visualizing clustering results is crucial for understanding the participants’ proximity to cluster centroids and to one another. However, this approach alone may not fully reveal the complex cross-correlations between the features and the clusters. To address this limitation and gain deeper insights into the data, we turned to Decision Tree-based methods, which are commonly used in supervised learning (Kőrösi & Vinkó, 2021). Decision Trees are adept at capturing non-linear relationships between features and the target variable. Moreover, their decisions are interpretable, making them ideal for elucidating correlations and cross-correlations within the data. Therefore, we employed a Decision Tree regressor to unravel the relationships between participant information (features) and the selected cluster values.
3. Results
3.1. Anthropometric data
The total group anthropometric and HRV characteristics were presented in Table 1. The mean ages were similar at the different positions (p = 0.81), and the distribution of the different positions were also similar in the different age groups (p = 0.82). Most of the anthropometric values increased significantly with age, while the percentage of muscle amount had the opposite tendency (Figure 1). As regards the post-based differences in these factors, most of them were higher in the goalkeeper and pivot players compared to the other posts (especially to the center backs and wings), except the percentage muscle amount that showed the opposite pattern (Figure 2).
Figure 1
Age-based (A1–A4) anthropometric parameter differences. A: body weight, B: body height, C: BMI, D: muscle amount, E: ratio of muscle amount, F: Fat amount, G: ratio of fat amount, H: ratio of fat and muscle amount. * Signs significant differences from the A1 age group, # from the A2 age group.
Figure 2
Post-based anthropometric parameter differences. A: body weight, B: body height, C: BMI, D: pNN50, E. ratio of muscle amount, F: Fat amount, G: ratio of fat amount, H: ratio of fat and muscle amount. * Signs significant differences from the wing players, # from the center back players.
3.2. HRV parameters
Regarding the analysis of cardiovascular data, neither age nor the post had significant influence on the HR and RRi values. In contrast, age caused significant effects in two parameters (SDNN and RMSSD/RRi; Figure 3.) with a trend to decrease with age. Furthermore, the post-based group analyses revealed significant effects in the SDNN, RMSSD and RMSSD/RRi data with highest levels in the pivots (Figure 4).
Figure 3
Age-based (A1–A4) cardiovascular parameter differences. A: HR, B: RRi, C: SDNN, D: pNN50. E: RMSSD, F: lnRMSSD, G: RMSSD/Ri. * Signs significant differences from the age-based group with the same colour as their bar. H: Changes in correlation values between the different C_RMSSD values and RRi in the different age-based groups. The case of /P0 corresponds to RMSSD. * Signs significant differences between A1 and A3, # between A2 and A3, o between A1 vs A4 and + between A2 and A4 age groups.
Figure 4
Post-based cardiovascular parameter differences. A: HR, B: RRi, C: SDNN, D: pNN50. E: RMSSD, F: lnRMSSD, G: RMSSD/Ri. * Signs significant differences from the post-based group with the same colour as their bar. H: Changes in correlation values between the different C_RMSSD values and RRi in the different post-based groups. The case of /P0 corresponds to RMSSD. * Signs significant differences between Wing and goalkeeper, # between center back and goalkeeper, o between center back and pivot, $ between wing and back and § between center back and back posts.
3.3. Correlation between the anthropometric and cardiovascular parameters
The correlation analysis between the anthropometric and the cardiovascular parameters (n = 513) detected high level significant correlations (R > 0.3) in some age- and post-based subgroups (Table 2). Thus, players only in the A4 subgroup had correlations with high level between the HR and most of the anthropometric parameters (Table 2). Regarding the post-based correlations, the pivot players had the highest number of significant correlations, while the back players the lowest one. The wing players had high levels of correlations only between the different HRV values and the body composition parameters.
Table 2
Spearman’s correlation with >0.3 regression values between the anthropometric and cardiovascular parameters in the different age- and post-based subgroups.
| HR | SDNN | pNN50 | RMSSD | lnRMSSD | RMSSD/RRi | |
|---|---|---|---|---|---|---|
| A4 | ||||||
| Body weight | 0.42 | |||||
| Muscle ratio (%) | –0.45 | |||||
| Fat mass (kg) | 0.56 | |||||
| Fat ratio (%) | 0.56 | |||||
| Fat/Muscle Ratio (%) | 0.52 | |||||
| BMI | 0.34 | |||||
| Goalkeeper | ||||||
| Body height | –0.51 | 0.35 | 0.32 | |||
| Muscle ratio (%) | –0.36 | |||||
| Pivot | ||||||
| Body weight | 0.45 | –0.24 | –0.34 | |||
| Body height | –0.57 | –0.52 | –0.59 | –0.59 | –0.69 | |
| Muscle mass (kg) | 0.40 | –0.43 | –0.55 | –0.47 | –0.47 | –0.43 |
| Muscle ratio (%) | –0.34 | |||||
| Fat mass (kg) | 0.42 | |||||
| Fat ratio (%) | 0.39 | |||||
| Fat/Muscle Ratio (%) | 0.38 | |||||
| BMI | 0.38 | |||||
| Back | ||||||
| Body weight | 0.45 | |||||
| BMI | 0.32 | |||||
| Center back | ||||||
| Body weight | –0.32 | |||||
| BMI | 0.39 | –0.41 | –0.34 | |||
| Wing | ||||||
| Body weight | –0.37 | –0.57 | –0.57 | –0.56 | –0.61 | |
| Body height | –0.43 | –0.39 | –0.39 | –0.43 | ||
| Muscle mass (kg) | –0.62 | –0.62 | –0.61 | –0.66 | ||
| Fat mass (kg) | –0.34 | –0.32 | –0.30 | –0.35 | ||
| BMI | –0.43 | –0.44 | –0.43 | –0.47 |
3.4. Results of corrected RMSSD analysis
The correlation analysis of all the available data (12213) between the RRi and the different HRV parameters showed that pNN50 and RMSSD had the highest level of positive correlations with the RRi (R = 0.63; and 0.50; respectively), while the SDNN had the lowest level (R = 0.32), and this phenomenon also occurred in each age- post-based subgroups.
In agreement with Sacha et al. the detailed analysis of C_RMSSD values at total group level showed that the RMSSD decreased its dependence on RRi after being divided by RRi (from R = 0.50 to R = 0.24), and even had inverse correlation appeared from division by RRiP2 (Sacha, 2014b; Sacha & Pluta, 2005, 2008). The inspection of curves of different age-based groups revealed that players in the A1 and A2 groups had significantly lower level of correlations between RMSSD/RRiP2 and RMSSD*RRiP1 values compared to A3 and A4 groups (Figure 3H). Regarding the post-based analysis, the wing, goalkeeper and center back players showed inverse correlations from division by RRiP2, while the back and pivot players had this phenomenon from division by RRiP4. Thus, the curves of different post-based groups showed that the wing and center back players had significant differences from the other players between RMSSD/RRiP4 and RMSSD*RRi (Figure 4H).
3.5. Machine Learning Investigation Results
The initial exploration of raw sequenced data, supported by visual representations in the form of charts, provided valuable insights into the characteristics of participants. However, to gain a deeper and more detailed understanding, a thorough investigation employing machine learning techniques were conducted such as clustering and decision trees. These methods allowed us to uncover complex cross-correlations among variables like RMSSD, playing position, and age.
In the first step of our investigation, we applied the K-Means clustering method with k = 3 to group participants based on their measured and cumulative features. The results of this clustering are presented in Figure 5A, revealing that when visualizing the t-SNE 2D data in conjunction with the Age groups, distinct separation of participants into different clusters is evident. Thus, age groups A4 and A3 were predominantly associated with cluster 2, age groups A2 and A3 with cluster 1, and age groups A1 and A2 with cluster 0. A similar pattern is observed when considering player positions; the t-SNE 2D data, when linked with player positions, generates distinct groupings (Figure 5B). Notably, ‘back’ and ‘goalkeeper’ are primarily connected to cluster 2, ‘wing’ and ‘center back’ are associated with cluster 1, and ‘pivot’ is linked to cluster 0.
Figure 5
Visualization of the t-Distributed Stochastic Neighbor Embedding (t-SNE) 2D data by clustering method (K-means) in conjunction with the age groups (A), or the player positions (B). Displaying the exploration of decision tree classification on the cluster labels and participants’ features (C).
The main characteristics of clusters are:
Cluster 0: the scaled mean RMSSD <0.464 and muscle weight minimum <0.225
Cluster 1 the scaled mean RMSSD <0.464 and muscle weight minimum >0.225
Cluster 2 the scaled mean RMSSD >0.464
To gain a deeper understanding of how the clustering method assigned cluster labels, the decision tree regression was performed on cluster labels and participants’ features (Figure 5C). The results indicate that the most crucial feature for determining cluster assignment is the mean RMSSD, followed by the participant’s body weight and muscle weight. Surprisingly, the RMSSD/RRi did not influence the clustering process.
4. Discussion
Prevention is a very important field in sport medicine. The repeated detection of body composition together with cardiovascular parameters are key issues in sport practice due to its link to performance and injury risk prevention. In agreement with earlier data, the anthropometric characteristics of handball players differ according to the position of play and the age (Dengel et al., 2014; Ghobadi et al., 2013; Krüger et al., 2014), and significant associations were detected between several anthropometric parameters and HRV factors (Plaza-Florido, Migueles, Sacha, et al., 2019), and the degree of these relationships were age and post dependent. The present study aimed to identify the main determinants of HRV in young handball players aged 14–21 years and to integrate these determinants as normative values of several HRV parameters, including C_RMSSD values, as main mirror of the activity of autonomic system. The results of the correlation analyses between HR/RRi and HRV parameters are in agreement with a recent report detecting HRV parameters in football and ice hockey players (Ziadia et al., 2022). Regarding the reports of HRV parameters in handball players, monthly resting HR and HRV measures were monitored in highly-trained adolescent handball players, suggesting only a limited usefulness this relatively rare monitoring in practice (Buchheit, 2015).
One of our important goals was to determine whether different groups of participants exhibited distinct RMSSD patterns, or if we could identify specific participant groups that shed light on the significance of RMSSD. K-Means clustering unveiled distinctive participant „habits” and facilitated the measurement of distances between participants. By using the Decision Tree regressor, intricate relationships and cross-correlations were explored between features and clusters, shedding light on the underlying patterns and nuances within the data. This combination of unsupervised and supervised machine learning techniques has empowered us to extract meaningful insights and drive our research forward. Data preprocessing is a critical step in preparing our data for machine learning models. Addressing skewness, aggregating attributes, and employing dimension reduction techniques were vital processes that paved the way for more effective analysis and model building. These techniques not only enhance the quality of the data but also make them more accessible and interpretable, allowing for deeper insights and improved model performance. The findings achieved by t-SNE, K-Means and Decision tree were particularly interesting, as they highlighted the significance of RMSSD in relation to different age groups and player positions. This suggests that RMSSD carries essential information about a participant’s state, playing style, and body parameters. Importantly, this significance has been established through the rigorous application of machine learning methods. While our initial data investigation successfully validated the theory regarding the importance of RRi on RMSSD, it served as a solid foundation for future time-series analysis and further exploration into the relationships between these variables. These results also emphasized the power of machine learning techniques in unravelling complex associations within the dataset. Thus, it is clearly proved the importance of stratified analyses of routinely recorded cardiovascular parameters in athletes. These normative values could be used in practice as a reference during evaluation of the load and stress (environmental factors) and follow-up to recovery. The results of machine learning method of HRV parameters with RRi, age, BMI, etc. showed that RRi (and HR) was the strongest determinant for all HRV parameters for the general group and for all age subgroups, and the RMSSD was main factor for cluster assignment, while not the RMSSD/RRi.
It is well-known that the close correlation between HR/RRi and HRV parameters is due to at least partially to mathematical bias, therefore, Sacha et al. developed a model to modify the HRV dependence on HR; i.e. by decreasing or increasing this relationship (Sacha, 2014a). They showed that the shape of correlation curves of HR (or RRi) and the corrected HRV values may be a sign of pathological conditions (Sacha, 2014a; Sacha & Pluta, 2005). The investigation of the C_RMSSD values provided several new results in our condition. While most of the HRV measures did not differ between age subgroups, the age was a significant predictor of the majority for C_RMSSD data, i.e. the curves of the Spearman’s correlation values showed lower levels of correlations among the RRi and C_RMSSD between C_RMSSD/P2 and C_RMSSD*P1 interval in the younger subgroups (A1 and A2) compared to the older ones (A3 and A4). As regards the post-based analysis, the back and wing players had diminished correlations compared to other players between in this respect. These data suggest a higher level of sympathetic activities in subgroups with lower C_RMSSD correlation values.
Our study can be viewed as a complement to some earlier studies performed in healthy and obese children and in contact sport athletes (ice hockey and football players), since it provides normative values for handball players aged between 14–21 years achieved by repeated measurements (Gąsior et al., 2018; Plaza-Florido, Migueles, Mora-Gonzalez, et al., 2019; Veijalainen et al., 2019; Ziadia et al., 2022). These earlier studies suggest that decreased cardiorespiratory fitness may be associated with autonomic nervous system dysfunction. In young adults lower values on resting HRV parameters were associated with higher values on cardiovascular disease risk factors represented by metabolic syndrome markers and insulin resistance (Plaza-Florido et al., 2020; 2022). In agreement with a recent study (Plaza-Florido et al., 2023), the relationship with RRi (or HR) with C_RMSSD values significantly decreased by division with power 2 in this young healthy competitive handball players with good physical condition.
4.1. Limitations and future recommendations
This study has some limitations that should be mentioned. First, this is a single-center study, therefore, some caution should be applied when expanding our results to other handball players. Furthermore, studies are also needed to determine HRV values specific to a female population.
The findings have to be interpreted with caution as several other factors could influence the results, including respiratory rate, weather parameters and/or level of physical and psychotic stress, which were not characterized. While the participants were required to perform the measurements during resting condition, this condition was not controlled in this study. However, it might be supposed that the mean of the huge amount of data during this relatively long period in the 85 players could cover the effects of daily changes in these factors.
Altogether, knowledge of the mean of these special cardiovascular parameters in different groups of healthy sedentary subjects or athletes might help in revealing the outlier data and find the causes of these alterations. The prolonged effects of the training performed one day before should also be considered. To decrease the methodological limitations of this study, a modified design would be more reliable, i.e. cardiovascular parameter detection should be performed before the training session after a short resting period.
5. Conclusion
In conclusion, the present study firstly provides baseline cardiovascular values for corrected HRV parameters specific to male handball players aged 14–21 playing at different posts. The systemic data analyses explored age- and post-based differences in autonomic function, indicating that individual alterations from the mean values in different subgroups might indicate changes in performance. Thus, SDNN and RMSSD/RRi parameters decreased with age, and the pivot players had the highest values. The detailed analyses by applying machine learning methods and C_RMSSD delivered several new data which may help to fill some important gaps in the literature of sport-specific relationships between anthropometric and cardiovascular parameters in handball with important practical implications for trainers and coaches. Future research should also use similar methodology to investigate the effects of other factors, including the participant psychological conditions and/or environmental parameters (e.g. weather). This method might also be appropriate for the evaluation of the development in the cardiovascular functions during the different training phases.
Competing Interests
The authors have no competing interests to declare.
Author Contributions
All persons listed as authors have contributed to preparing the manuscript, and no person or persons other than the authors listed have contributed significantly to its preparation. All authors discussed the results and contributed to the final manuscript.
Specifically:
Zita Petrovszki was involved in the data collection and analyses and preparation of figures. Oliver Czimbalmos developed the software for machine learning analyses and was involved in the data processing as well. Vera Gal was involved in the data collection and anthropometric measurements. Gabor Korosi took part in the development of the software for machine learning analyses, and the consideration of the results obtained with these methods. Edit Nagy was involved in the data collection and preparation of the manuscript. Rita Mikulan was involved in the data analyses and explanations/considerations of the results. Gyöngyi Horvath conceived and designed research and wrote the manuscript involving all the other authors.
We did not apply any types of AI methods for the writing of this manuscript.
Zita Petrovszki and Olivér Czimbalmos authors have made an equal contribution.