Introduction

Systematic monitoring of training load (TL) is a central component in modern elite soccer. TL is commonly conceptualized through the distinction between external training load (ETL) and internal training load (ITL) [1]. ETL describes the physical work performed by the athlete, whereas ITL reflects the individual psychophysiological response to that work [2, 3]. ETL is commonly derived from wearable tracking systems, including GPS/GNSS and inertial sensors, whereas heart rate (HR)-derived measures are widely used to quantify ITL, as they provide a non-invasive, continuous assessment of cardiovascular load during training.

Advances in wearable technology have enabled practitioners and researchers to objectively quantify ETL. GPS devices with integrated inertial measurement units (IMUs) are now widely used to capture physical demands during training sessions [4,5]. Commonly reported ETL variables include total distance, high-speed running distance, accelerations, decelerations, and player load [6–8]. When combined with HR monitoring, these GPS–IMU-derived variables provide an opportunity to directly examine how specific locomotor and mechanical demands translate into cardiovascular load. Importantly, internal load integrates both exercise intensity and duration, suggesting that accumulated session volume may be a primary driver of cardiovascular strain [9].

Soccer is characterized by repeated accelerations, decelerations, changes of direction, and high-speed running interspersed with periods of low-intensity activity [10–12]. These demands impose substantial mechanical and cardiovascular stress on players, with marked variability both between individuals and within the same players across training sessions [6], even when the external load is identical. Within this framework, the relationship between ETL and ITL represents a key link between prescribed training stimuli and the resulting physiological stress imposed on the athletes. The importance of TL monitoring is further amplified in youth soccer, as adolescent players are exposed to progressively increasing training demands, while undergoing growth- and maturation-related changes in body composition and body size, neuromuscular coordination, and cardiovascular function [13–15]. These developmental processes may further modulate cardiovascular responses to ETL, potentially resulting in greater inter- and intra-individual variability in HR responses compared with adult players [16].

Previous research has examined the associations between GPS-derived ETL variables and HR-derived ITL metrics, demonstrating that cardiovascular responses reflect the cumulative physiological stress imposed by ETL [17]. Volume-based metrics, including total distance and player load, have demonstrated moderate to very large associations with HR-derived measures, whereas intensity variables (e.g., total distance per minute) have shown more variable associations [17,18]. However, many studies have relied on pooled correlation analyses that do not distinguish between between-player and within-player correlations, potentially overestimating associations, when repeated observations are collected across time. Between-player correlations reflect whether changes in one variable are paired with analogous changes in another variable among different individuals. In contrast, within-player correlations describe whether changes in one variable over time are associated with concurrent changes in another variable within the same individual.

Methodological concerns are evident in studies, which include simple correlation analyses, as these may violate the assumptions of independent observations when repeated measures are collected over time and may conflate between-player and within-player effects [19]. Recent research has emphasized the importance of analytical approaches that explicitly account for the longitudinal and clustered structure of TL data, including within-individual correlations, generalized estimating equations, and linear mixed-effects models [18,20,21]. Such approaches allow for the isolation of intra-individual associations between ETL and ITL across training sessions. This provides a more accurate representation of how changes in external load are related to physiological responses within the same player, which is essential for individualizing training prescription and TL monitoring.

Another important consideration is whether ETL should be expressed in volume or in intensity metrics. Volume-based ETL metrics primarily reflect overall training volume, whereas intensity metrics expressed in relation to training time (in minutes) provide insight into training intensity independent of session length [8,22]. While both approaches are widely used in practice, their relative contribution to variability in HR response remains unclear in elite youth populations [23,24]. Given that HR responses are inherently time-dependent, it is less understood whether absolute volume-based metrics or intensity-based variables better explain HR-derived ITL in elite youth soccer training.

Despite the expanding literature on TL monitoring in youth soccer, limited evidence is available on the relationship between absolute and intensity-based GPS metrics using a within-player analytical approach during structured in-season training sessions. Furthermore, few studies have explored the independent predictive contribution of specific ETL variables to HR-based ITL, while accounting for repeated observations within players.

Therefore, the aims of this study were (i) to quantify within-player associations between GPS-derived ETL variables and HR exertion in in-season training sessions in elite under-17 soccer players, and (ii) to determine which intensity-based ETL metrics independently predict HR-based ITL.

Methods

Anthropometry and maturity assessment

Anthropometric measurements were performed by an ISAK level 2-accredited physiotherapist during the football academy’s routine screening on 12 January 2023. Standing height, sitting height (93.5 ± 3.2 cm), and body mass were recorded, and leg length was calculated as standing height minus sitting height (84.4 ± 3.1 cm). Chronological age was calculated on the date of the anthropometric assessment. Maturity offset, representing the estimated number of years before or after the age at peak height velocity (PHV), was calculated using the equation developed for boys by Mirwald et al. [25]:

Positive maturity offset values indicate years after PHV, whereas negative values indicate years before the age at PHV.

External training load monitoring

ETL was quantified using 10-Hz GPS units with integrated 100-Hz IMUs (Catapult S7 Vector, Catapult Sports, Melbourne, Australia). Each player wore the same device throughout the monitoring period to minimize the inter-unit variability. Devices were positioned between the scapulae using a manufacturer-provided sports vest. The tracking unit provides locomotor and mechanical TL variables derived from GPS and tri-axial accelerometer, gyroscope, and magnetometer data. In the present study, the following volume-based ETL variables were used: total distance covered (TDC; m), high-speed running distance (19.8–25.1 km·h ⁻ ¹, HSR; m), explosive efforts (EXP; count), and PlayerLoad^TM (PL; arbitrary unit, AU). These parameters were also expressed in relation to time, representing intensity-based metrics. HSR was defined according to the manufacturer’s default velocity thresholds. EXP were identified using the proprietary Catapult algorithm based on combined high intensity accelerations and changes of direction (> 3.5 m·s ⁻ ²), and deceleration events (<−3.5 m·s ⁻ ²). PL was calculated from the vector magnitude of tri-axial accelerometer data and represents a composite measure of the accumulated mechanical TL, according to the manufacturer’s algorithm, originally described by Boyd et al. [26]:

where represents the acceleration at the y, x, and z axes, respectively.

Previous studies have reported a coefficient of variation (CV) typically below 5% for total distance and player load, whereas higher variability has been observed for high-speed running and accelerations in youth populations (10–15% CV) [1,27–29]. The Catapult Vector unit system has demonstrated acceptable validity for distance-based metrics (typical error ~1–2.5%) and high inter-unit reliability (ICC = 0.95–0.99, 1–3% CV) [30]. In line with recent methodological recommendations for transparent GPS/GNSS reporting in youth soccer proposed by Havanecz et al. [31], key data-quality indicators were reported. GPS signal quality was considered acceptable when the mean number of connected satellites was > 6 and horizontal positioning quality was < 2.0; in the present study, the average satellite count exceeded 6 and mean horizontal dilution of precision (HDOP) remained < 1.0, thereby meeting standardized data quality criteria [29].

Results

Descriptive statistics for the examined variables are presented in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics for the examined external training and internal training load variables.

https://doi.org/10.1371/journal.pone.0357176.t001

As illustrated in Figs 1 and 2, HR exertion and ETL variables demonstrated similar temporal patterns across training days, with peak values observed mid-week (Wednesday) and lower values toward the end of the training week (Friday). Notably, this alignment was more pronounced for volume-based ETL variables (Fig 1) compared to intensity-based variables (Fig 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Daily variation in HR exertion and volume-based ETL variables within a microcycle.

TDC: total distance covered, HSR: high-speed running distance, EXP: explosive efforts, PL: PlayerLoad^TM.

https://doi.org/10.1371/journal.pone.0357176.g001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Daily variation in HR exertion and intensity-based ETL variables within a microcycle.

TDC: total distance covered, HSR: high-speed running distance, EXP: explosive efforts, PL: PlayerLoad^TM.

https://doi.org/10.1371/journal.pone.0357176.g002

Repeated measures correlation revealed nearly perfect within-player associations between HR exertion and both TDC (r = 0.921, 95% CI: 0.903–0.946, p < 0.001) and PL (r = 0.924, 95% CI: 0.906–0.948, p < 0.001). HSR (r = 0.662, 95% CI: 0.587–0.726, p < 0.001) and EXP (r = 0.621, 95% CI: 0.542–0.690, p < 0.001) demonstrated large associations with HR exertion. Intensity metrics showed smaller effect sizes compared to absolute variables. Large correlations were observed for TDC·min ⁻ ¹ (r = 0.543) and PL·min ⁻ ¹ (r = 0.503), and a moderate correlation was observed for HSR·min ⁻ ¹ (r = 0.390), whereas EXP·min ⁻ ¹ (r = 0.195) demonstrated a small association (p = 0.002).

The linear mixed-effects model predicting HR exertion·min ⁻ ¹ showed that TDC·min ⁻ ¹ demonstrated the strongest independent effect, followed by EXP·min ⁻ ¹. In contrast, HSR·min ⁻ ¹ and maturity offset were not independently associated with HR exertion·min ⁻ ¹ (Table 2). Multicollinearity diagnostics indicated no evidence of collinearity among the predictors (VIF = 1.087–1.121).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Linear mixed-effects model predicting HR exertion·min ⁻ ¹ by intensity-based ETL variables and maturity offset.

https://doi.org/10.1371/journal.pone.0357176.t002

Including playing position in the model did not alter the overall pattern of associations. TDC·min ⁻ ¹ (β = 0.482, p < 0.001) and EXP·min ⁻ ¹ (β = 21.920, p < 0.001) remained significant predictors, whereas HSR·min ⁻ ¹ (p = 0.265) and maturity offset (p = 0.739) remained non-significant. No statistically significant position-related differences were observed.

Discussion

The present study examined within-player associations between GPS-derived ETL variables and HR-based ITL during in-season training in elite youth soccer players. The main finding was that HR exertion was most strongly associated with TDC and PL. In contrast, TDC·min ⁻ ¹ and EXP·min ⁻ ¹ were the main independent predictors of HR exertion·min ⁻ ¹, whereas HSR·min ⁻ ¹ did not provide additional explanatory value. These findings indicate that HR-based load metrics reflect different aspects of the external–internal training load relationship.

The daily patterns provide additional insights for these associations (Figs 1 and 2). HR exertion and volume-based ETL variables followed a broadly similar weekly pattern, with the highest values observed around the mid-week training day. This is consistent with the structure of the weekly microcycle, in which MD-3 sessions typically represented the highest-volume stimulus. In contrast, the results showed greater variability in the external–internal load relationship, especially on Monday and Tuesday, which may reflect the more heterogeneous training aims on these days, including technical activities, agility tasks, and individualized load management following the preceding match. In these training days, increases in, particularly, EXP were not paired with an analogous increase in HR exertion. Similar day-to-day variation in TL across soccer microcycles has been reported previously, when training content differs according to match-day proximity and age category-specific programming [40]. These descriptive patterns should not be interpreted as evidence of day-specific external–internal load relationships, as training day was not formally included as a predictor in the analysis.

The nearly perfect within-player associations between HR exertion and both TDC and PL indicate that accumulated cardiovascular load closely reflected the volume of external work performed within individual players. However, this finding should be interpreted in light of the cumulative nature of HR exertion. Because HR exertion integrates both exercise intensity and duration, strong associations with volume-based ETL variables are partly expected. Therefore, the main practical value of the present findings is not merely the strength of these associations, but the distinction between cumulative and time-normalized interpretations of internal load. In this context, HR exertion appears more suitable for evaluating accumulated cardiovascular load across the full training session, whereas HR exertion·min ⁻ ¹ provides a more specific indication of per-minute internal load intensity. Accordingly, greater total distance and greater mechanical load are likely to be accompanied by higher cardiovascular strain across training sessions. Previous studies have also demonstrated strong relationships between accumulated external load and cardiovascular responses in team sports [7,18]. In addition, several studies have also demonstrated the concept that HR-derived internal load reflects the cumulative physiological stress imposed by exercise, which is mainly driven by the total volume of work performed [3,9]. In soccer-specific research, TDC and PL have been identified as key load variables, integrating a wide range of movement-specific activities performed during training [6,7], which may explain their strong association with HR exertion in the present within-player analysis.

When HR exertion was expressed relative to training duration, TDC·min ⁻ ¹ emerged as the strongest independent predictor of HR exertion·min ⁻ ¹. This suggests that the overall locomotor intensity of training remained closely associated with per-minute cardiovascular strain. This interpretation is consistent with the concept that HR-derived load reflects the physiological response to the external work performed, but that the strength and meaning of this relationship depends on whether load is expressed in absolute or relative terms [3,9,18]. EXP·min ⁻ ¹ also contributed independently, suggesting that explosive actions may provide additional information about the intensity of the internal response. This may be relevant because acceleration-, deceleration-, and change-of-direction-based actions impose distinct mechanical and physiological demands that are not fully captured by distance-based variables alone [41]. In contrast, HSR·min ⁻ ¹ was not independently associated with HR exertion·min ⁻ ¹ after accounting for TDC·min ⁻ ¹ and EXP·min ⁻ ¹, suggesting that high-speed running intensity did not provide additional explanatory value in this training context.

An important methodological aspect of the present study is the use of a within-player analytical approach to examine the relationship between ETL and ITL. Previous studies have often relied on pooled or group-level analyses, reporting associations between external and internal load without separating between- and within-player effects [7,17,42,43]. While these studies consistently demonstrate moderate-to-very large relationships between ETL and ITL variables, such approaches may be influenced by inter-individual variation and may therefore not reflect intra-individual changes across repeated training sessions [19]. By accounting for repeated observations within players, repeated-measures correlation provides a more direct estimate of how fluctuations in ETL are accompanied by changes in ITL within the same player [38]. This is relevant for training monitoring, where practitioners are often interested in whether an individual player shows a higher or lower internal response than expected for a given external training stimulus.

Regarding youth physical development, adolescent soccer players are exposed to ongoing growth- and maturation-related changes that influence physiological responses to TL [14,16]. These developmental factors may contribute to greater inter- and intra-individual variability in HR responses compared with adult athletes [18]. In previous research with sub-elite youth soccer players [44] the strength of the correlations progressively decreased with the decrease in age, with the lowest associations being found in the age group close to the age at peak height velocity (under-15). In our study, maturity offset was not associated with HR exertion·min ⁻ ¹. This may partly reflect the relatively narrow post-PHV range, as most athletes were from two to three years after their age at PHV, and contradicts previous findings reporting maturity-related differences in training load across age and maturity groups in youth soccer [45]. The present within-player findings suggest that changes in external load were closely reflected by changes in HR data within individual players, especially for volume-based variables. These strong correlations also indicate that in the under-17 age group, the associations between ETL and ITL are of similar magnitude to that observed for adult players. This further highlights the importance of individualized monitoring of training load in youth soccer, which may contribute to more individualized load management and training prescription.

Based on the present findings, the use of solely external metrics (e.g., total distance, high-speed running distance) may overlook individual variability in physiological adaptation. From a practical point of view, the HR exertion appears useful for evaluating the cardiovascular cost of a training session, in relation to volume-based GPS variables such as TDC and PL. In contrast, HR exertion·min ⁻ ¹ may provide additional insight into the intensity of the internal response, with TDC·min ⁻ ¹ and EXP·min ⁻ ¹ being the most relevant ETL indicators in the present model. Therefore, combining ETL and ITL metrics, alongside appropriate within-player analytical approaches, may provide more precise TL monitoring and support further decision-making in training prescription.

Conclusion

In elite youth soccer training, HR exertion showed very strong within-player associations with TDC and PL. TDC·min ⁻ ¹ and EXP·min ⁻ ¹ emerged as the main independent predictors of HR exertion·min ⁻ ¹. These findings indicate that cumulative and time-normalized HR-based load metrics should be interpreted separately, as they reflect different aspects of the external–internal load relationship. Within-player analytical approaches may therefore provide a more individualized interpretation of training responses in elite under-17 youth soccer.

Acknowledgments

We would like to sincerely thank the young football players for their valuable contribution to this study. During the preparation of this manuscript, the authors used ChatGPT (OpenAI, GPT-5.5) to assist with the preparation of Figs 1 and 2. The figures were based on the authors’ original data, and all outputs were reviewed, verified, and edited by the authors. The authors take full responsibility for the accuracy and integrity of the figures and the manuscript.