Effective monitoring of internal training load is central to optimizing performance and minimizing bad overreaching in team sports, with heart rate (HR)–based measures among the most frequently used tools in professional football scenarios (1). HR indices collected at rest, during exercise, and in recovery provide integrative information about fitness, fatigue, and endurance performance, and are widely regarded as core components of contemporary athlete monitoring systems (2). Also resting HR can be classified as a marker that may reflect both cardiorespiratory fitness status and short-term readiness/fatigue state, acknowledging that its interpretation is context-dependent and influenced by recovery, illness, and non-training stressors (2). Small-sided games (SSGs) are constrained, drill-based forms of play (3) that allow coaches to simultaneously target technical–tactical behaviors and high cardiovascular loading in a time-efficient manner (4). Experimental studies has shown that appropriately designed SSGs can elicit HR responses comparable to or even exceeding those induced by intermittent running protocols or official match play, reinforcing their use as a conditioning stimulus (5). At the same time, repeated-measures studies reveals substantial within- and between-player variability in HR responses during standardized SSG protocols, indicating that players may experience different internal loads despite sharing identical task constraints (6). Recent syntheses indicate that internal-load measures (e.g., HRmean, %HRmax) generally show low variability compared with external-load and technical outcomes, and that it is important to distinguish within-session bout-to-bout stability from between-session reliability when interpreting variability magnitudes (4). Although HR-derived indices are not injury outcomes, internal-load monitoring is a core component of contemporary workload–response frameworks in which inappropriate load progression and inadequate recovery are theorized to contribute to maladaptation and injury risk (7, 8). In applied football context, HR-based internal-load information from SSGs is commonly used to (i) verify whether the intended cardiovascular stimulus was achieved, (ii) individualize exposure via bout number/duration and player rotations, and (iii) identify disproportionate responses (higher-than-expected HR for a given constraint set) that may indicate transient fatigue, heat strain, illness, or reduced readiness and therefore warrant altered training content or recovery emphasis (2).

Manipulating SSG format, particularly pitch size and number of players, significantly alters HRmean, rating of perceived exertion, and locomotor demands, with smaller formats generally associated with higher internal and external loads (9). For example, comparisons across formats consistently show that 2v2 configurations elicit higher HRmean and perceived exertion than larger formats such as 4v4, while also modifying technical actions and ball involvement patterns (10). Recent data in youth soccer further indicate that HR responses during 2v2 and 4v4 SSGs are closely related to technical performance indicators such as lost balls and successful shots, underscoring the sensitivity of HRmean to contextual task constraints (11). Beyond average intensity, within- and between-player variability in HR across repeated SSG bouts, often quantified using the coefficient of variation (CV%), has emerged as an importante marker for assessing the stability and responsiveness of different SSG formats (12). CV% provides an absolute index of within-player variability (typical error expressed relative to the mean) and is commonly used to quantify reliability/consistency of repeated physiological measures (13). Longitudinal work in professional football shows that internal and external load measures exhibit considerable session-to-session fluctuations during preseason and in-season periods, reinforcing the need to understand factors that influence the consistency of HR responses in SSGs (14).

At the individual level, resting HR is inversely associated with cardiorespiratory fitness and positively associated with cardiovascular risk, and typically declines in response to chronic endurance training (15). Cross-sectional comparisons between endurance-trained and untrained adults demonstrate that athletes display lower resting HR together with higher maximal oxygen uptake, supporting the use of resting HR as a global index of aerobic conditioning (16). In team sports, conceptual frameworks highlight that resting and exercise-related HR measures, along with heart rate variability, provide non-invasive insights into autonomic balance and aerobic fitness, and can be embedded within multivariate monitoring systems (17). Standardized submaximal fitness tests have gained popularity because exercise HR recorded at a fixed submaximal intensity shows high reliability and large inverse correlations with endurance performance tests in team sport athletes (18). Longitudinal observations indicate that changes in submaximal HR during running or shuttle-based protocols are sensitive to training-induced alterations in aerobic capacity in professional players, supporting their use for routine status monitoring (19). Nevertheless, recent surveys of load-monitoring practice in elite football emphasize substantial heterogeneity in the implementation and interpretation of HR-based indices, and point to a lack of evidence on how simple measures such as resting HR and submaximal HR relate to HR responses in football-specific SSG formats (20).

The Yo-Yo Intermittent Recovery (YYIR) tests are established field assessments of high-intensity intermittent running capacity in team sports and show strong associations with external running demands and high-intensity activity during match play (21). Foundational work on the YYIR tests demonstrated large correlations between YYIR performance and high-speed running distance in football, confirming their ecological validity for intermittent sport demands (22). Subsequent studies in youth soccer players have confirmed that YYIR1 exhibits good test–retest reliability and can discriminate between performance levels and positional roles (23). YYIR performance is also associated with HR-derived indices of match intensity, including time spent in high-intensity HR zones and post-exercise HR recovery, indicating that YYIR outcomes reflect integrated cardiovascular and neuromuscular function during intermittent play (24). Emerging evidence links YYIR results with technical and tactical performance in small-format games such as 2v2, suggesting that players with higher intermittent running capacity may better tolerate the demands of constrained SSG scenarios (25). Nonetheless, recent analyses emphasize that YYIR performance is influenced by multiple morphological and neuromuscular factors and do not clarify how YYIR levels compare with resting HR and submaximal HR in explaining HRmean and its stability across repeated SSG bouts of different formats (26).

Current evidence establishes SSGs as powerful yet variable training tools and confirms that HR-based markers (including resting HR, submaximal HR, and YYIR performance) each capture partially distinct aspects of aerobic fitness and intermittent exercise tolerance, while leaving their relative contributions to explaining HR responses during SSGs largely unresolved (27). The stability of HRmean across repeated SSG bouts, reflected by lower CV% in HR responses, is increasingly recognized as a key property for using SSGs to monitor training-induced aerobic adaptations, yet the physiological determinants of maintaining low HR variability over repeated 2v2 and 4v4 formats remain nuclear (28). In this context, stability refers specifically to within-session bout-to-bout consistency of HRmean (i.e., lower CV% across the three bouts), which is typically interpreted as a more predictable cardiovascular stimulus under standardized constraints, however low CV% should be interpreted pragmatically (predictable internal-load delivery) rather than as a proxy for superior adaptation per se. Meta-analytic work on sided-games further shows that exposure and intra-individual reliability of external high-speed and sprint running loads are strongly influenced by game format and constraints, reinforcing the need for format-specific analyses of internal load stability (29). However, HR responses during SSGs are multi-determined by task constraints (rules, area per player), game dynamics (ball-in-play time, transitions), and psychophysiological factors (competitive arousal and pacing) (4), so physiological markers may explain only a modest fraction of variance in HR outcomes.

Against this background, and in line with calls to integrate simple HR-derived markers into practical monitoring frameworks, it is pertinent to determine which HR-based measures best explain the internal load experienced during standardized SSGs (30). Therefore, the main aim of the present study was to identify which of three readily obtainable measures (resting HR, HR recovery post 60-s of a standardized warm-up running test, and HRmax at YYIR test level) best explains HRmean during 2v2 and 4v4 SSGs in football players, and a secondary aim was to determine which of these measures is most closely associated with the capacity to maintain lower CV% in HR responses across repeated SSG bouts and whether these relationships are format dependent, thereby providing format-specific guidance for integrating simple HR markers into SSG-based monitoring and prescription.

Across the sample, resting heart rate averaged 72.7 ± 8.3 bpm (range 58–84 bpm), HRR during the submaximal warm-up test was 145.1 ± 9.9 bpm (128–164 bpm), and the maximal heart rate recorded during the Yo-Yo IR1 test (HRmax) was 183.7 ± 5.4 bpm (174–194 bpm). Mean heart rate during 2v2 SSGs was 166.9 ± 5.1 bpm (155.0–176.3 bpm), whereas mean heart rate during 4v4 SSGs was 165.3 ± 6.0 bpm (152.0–176.6 bpm). The within-session coefficient of variation of HR across bouts was 2.05 ± 1.16% (0.35–5.07%) in 2v2 and 2.85 ± 1.43% (0.89–7.34%) in 4v4.

When comparing formats, mean HR was slightly higher in 2v2 than in 4v4, but the estimate was imprecise and did not reach statistical significance (166.9 ± 5.1 vs. 165.3 ± 6.0 bpm; mean paired difference Δ = 1.60 bpm, 95% CI − 0.80 to 4.00; t₃₉ = 1.35, p = 0.19, Cohen’s d = 0.21). In contrast, HR variability across bouts was significantly lower in 2v2 than in 4v4 (2.05 ± 1.16 vs. 2.85 ± 1.43%; mean paired difference Δ = −0.80, 95% CI − 1.44 to −0.16; t₃₉ = −2.52, p = 0.016, Cohen’s d = −0.40), indicating more stable cardiovascular responses across repeated bouts in the smaller format.

Associations between the three physiological markers and mean HR during SSGs (Figure 2) were generally small to moderate. In 2v2, Pearson correlations between HRmean and resting HR (r = 0.17, p = 0.30), HRR (r = −0.04, p = 0.80) and Yo-Yo HRmax (r = −0.03, p = 0.86) were all trivial to small and non-significant. In 4v4, resting HR and Yo-Yo HRmax again showed non-significant relationships with HRmean (r = −0.13, p = 0.43 and r = 0.23, p = 0.16, respectively), whereas HRR displayed a significant positive association with mean HR (r = 0.36, p = 0.023; 95% CI 0.05–0.60). Thus, players who exhibited higher HR during the standardized submaximal warm-up tended to experience higher mean HR during 4v4, while no such relationship was evident in 2v2.

Multiple linear regression models confirmed these patterns. For 2v2, the combined model including resting HR, HRR and Yo-Yo HRmax explained only 3.1% of the variance in HRmean (R² = 0.03, adjusted R² = −0.05, overall p = 0.77), and none of the individual predictors reached significance (resting HR: β = 0.11 bpm·bpm⁻¹, p = 0.31; HRR: β = −0.03 bpm·bpm⁻¹, p = 0.77; Yo-Yo HRmax: β ≈ 0.00 bpm·bpm⁻¹, p = 0.99). For 4v4, the same set of predictors accounted for 16.3% of the variance in HRmean (R² = 0.16, adjusted R² = 0.09, overall p = 0.090). In this model, HRR emerged as the only significant independent predictor of mean HR (β = 0.20 bpm·bpm⁻¹, p = 0.047), whereas resting HR (β = −0.10 bpm·bpm⁻¹, p = 0.38) and Yo-Yo HRmax (β = 0.13 bpm·bpm⁻¹, p = 0.46) were not significant. Variance inflation factors for the three predictors were close to 1.0 in all models, indicating negligible multicollinearity. In the combined-format analysis, we fitted a linear mixed-effects model (random intercept for player) including format (2v2 vs. 4v4), the three centered predictors (resting HR, submaximal warm-up HRR, Yo-Yo HRmax), and predictor × format interaction terms. The model provided limited overall explanatory value (marginal R² = 0.13; likelihood-ratio test vs. intercept-only model: p = 0.12). The only format-dependent association supported by the interaction terms was for HRR (HRR × format: β = 0.228 bpm·bpm⁻¹, p = 0.049), indicating that HRR related positively to HRmean in 4v4 (β = 0.201 bpm·bpm⁻¹, 95% CI 0.027 to 0.376, p = 0.024) but not in 2v2 (β = −0.026 bpm·bpm⁻¹, p = 0.77). Interactions involving resting HR and Yo-Yo HRmax with format were not statistically significant (both p ≥ 0.12).

For the stability of cardiovascular load, correlations between the physiological markers and the CV% of HR across bouts were generally small, with only one borderline association (Figure 3). In 2v2, CV% showed a moderate, negative but non-significant correlation with resting HR (r = −0.30, p = 0.065), such that players with higher resting HR tended to exhibit slightly lower variability of HR across bouts. Correlations with HRR (r = 0.25, p = 0.12) and Yo-Yo HRmax (r = 0.02, p = 0.92) were small and not statistically significant. In 4v4, none of the predictors were significantly associated with CV% (resting HR: r = 0.21, p = 0.20; HRR: r = −0.15, p = 0.35; Yo-Yo HRmax: r = −0.12, p = 0.45).

Format-specific multiple regressions with CV% as the outcome were consistent with the correlation analyses. In 2v2, the model including resting HR, HRR and Yo-Yo HRmax explained 16.7% of the variance in CV% (R² = 0.17, adjusted R² = 0.10, overall p = 0.083), but none of the individual predictors reached conventional significance (resting HR: β = −0.06%·bpm⁻¹, p = 0.16; HRR: β = 0.04%·bpm⁻¹, p = 0.18; Yo-Yo HRmax: β = 0.02%·bpm⁻¹, p = 0.43). In 4v4, the same predictors accounted for only 7.2% of the variance in CV% (R² = 0.07, adjusted R² = −0.00, overall p = 0.43), and again none of the individual coefficients were significant (all p ≥ 0.20).

To examine whether the relationships between physiological markers and HR variability differed by format, we fitted a linear mixed-effects model (random intercept for player) with CV% as the dependent variable, including format, the three centered predictors, and predictor × format interactions. The overall model supported modest explanatory value (marginal R² = 0.19; likelihood-ratio test vs. intercept-only model: p = 0.020). The main effect of format was significant, with CV% higher in 4v4 than 2v2 by 0.80 percentage points at average predictor values (β = 0.799%, p = 0.0033), consistent with the paired t-test. The resting HR × format interaction was also significant (β = 0.081%·bpm⁻¹, p = 0.016), indicating a more negative resting-HR slope in 2v2 (β = −0.045%·bpm⁻¹, p = 0.059) and a small, non-significant positive slope in 4v4 (β = 0.036%·bpm⁻¹, p = 0.13). In interpretive terms, a 10-bpm higher resting HR would correspond to 0.45 percentage points lower CV% in 2v2 (i.e., slightly greater within-session stability), whereas the analogous change in 4v4 would be ~0.36 percentage points higher CV% (a small, non-significant effect). Although these differences are modest in absolute terms, they are potentially meaningful for monitoring because they are of similar magnitude to typical within-session CV% values (2–3%). Interactions involving HRR and Yo-Yo HRmax did not reach conventional significance (p = 0.057 and p = 0.95, respectively).

To illustrate whether the negligible group-level difference reflects consistent individual responses or heterogeneous changes, Figure 4 present within-player paired-line plots for HRmean and HR CV% across the 2v2 and 4v4 formats.

Recent research highlight that variability is intrinsic to SSGs, with internal-load measures (e.g., HRmean, %HRmax) typically showing lower variability than technical and high-speed external-load outputs, and with variability differing depending on whether it is quantified within-session (bout-to-bout) or between-session (day-to-day). Building on this literature, the present study focused specifically on within-session bout-to-bout stability (CV% across three repetitions) and tested whether commonly collected HR-derived markers (resting HR, HRR during sub-maximal warm-up test, and Yo-Yo HRmax) help explain individual differences in HRmean and stability during 2v2 versus 4v4 formats. The primary findings were that HRR was the only measure associated with HRmean, and this relationship emerged exclusively during 4v4. Resting HR and Yo-Yo IR1 HRmax showed no meaningful associations with HRmean in either format, and none of the markers showed strong explanatory capacity for HR variability across repeated bouts, although resting HR displayed a modest format-dependent pattern. A plausible explanation for the null findings for Yo-Yo HRmax is that peak HR in maximal field tests often clusters near an individual ceiling (limited between-player spread), whereas Yo-Yo performance outcomes (distance/level) or running speed at standardized physiological thresholds typically provide stronger discriminatory information about intermittent running capacity. Additionally, HR variability was systematically lower in 2v2 than in 4v4, indicating greater bout-to-bout cardiovascular stability in more constrained formats. Practically, lower CV% in 2v2 suggests that this format may be preferable when coaches aim to deliver a standardized cardiovascular stimulus or use HRmean as a routine monitoring readout (reduced within-session ‘noise’), whereas relatively higher CV% in 4v4 may be acceptable (or even desirable) when the goal includes exposing players to more variable tactical–contextual demands while maintaining a broadly similar mean internal load.

The finding that mean HR did not differ significantly between 2v2 and 4v4 parallels previous research showing that smaller-sided games can yield high cardiovascular responses, and that variation in format does not always translate to mean HR differences. For example, a study (5) found comparable cardiovascular loads across multiple SSG formats. In our data the lack of mean HR difference suggests that, from a cardiovascular-load viewpoint, both formats can be used interchangeably to elicit similar mean HR. A likely explanation is that although the number of players changed, other constraints (e.g., pitch size, bout/rest durations) were held constant, thereby preserving internal-load consistency. Coaches frequently manipulate player numbers, pitch size, and bout structure to target intensity (9). Therefore, our results reinforce that mean HR alone may not be sensitive to format changes when other variables are controlled. However, because area per player was closely matched across formats and SSGs were preceded by a standardized warm-up, HRmean may have been partially ‘ceilinged,’ reducing sensitivity to player-number changes.

In contrast, the significantly lower HR CV% in 2v2 suggests greater physiological response stability in that format compared to 4v4. This aligns with findings (6) who reported smaller within-player variability in HR in the 3v3 format compared to 5v5. The probable mechanism is that fewer players amplify involvement and intensity per individual and reduce heterogeneity of effort, thus leading to more homogeneous responses. In our 2v2 scenario each player is more consistently active, reducing drop-off periods, hence lowering CV. Thus, if coaches seek more reproducible cardiovascular stimulus across players and sessions, smaller formats (such as 2v2) may be preferable. A plausible explanation is that 2v2 increases individual involvement and reduces low-engagement phases, however, without ball-in-play time, technical involvement, or external-load tracking, this mechanism cannot be confirmed. For applied monitoring, interpretability will be highest when the SSG is performed at a consistent microcycle location (e.g., a comparable day relative to match day), time of day, and standardized warm-up. Using the observed within-session CV% (2–3%), practitioners may consider individual deviations in HRmean that exceed typical within-session ‘noise’ (e.g., several bpm beyond a player’s usual response under the same constraints) as potential flags for follow-up rather than as deterministic indicators of maladaptation.

The fact that only HRR (sub-maximal warm-up HR) predicted mean HR in the 4v4 format, explaining about 16% of the variance, while resting HR and Yo-Yo IR1 HRmax did not, merits discussion. Previous research has shown that aerobic fitness markers (such as HRmax or VO₂max) sometimes correlate with internal load during SSGs, but often the associations are weak or inconsistent (41). The limited predictive power might reflect that SSGs impose complex multifactorial demands (technical, tactical, motivational) rather than pure aerobic challenge (42). In our case, HRR may reflect immediate cardiovascular readiness or autonomic state prior to session, and thus correlate with how intensely the player engages physiologically during the bout. That this predictive association emerges only in the 4v4 (and not in 2v2) suggests that when format allows more variation in involvement (i.e., larger team size), pre-session physiological readiness becomes a more important determinant of cardiovascular load, whereas in highly constrained, high-engagement formats (2v2) the format’s intensity may override these individual differences. However, this association may also reflect coupling between submaximal exercise HR responsiveness and HR during game-based activity, rather than acute readiness per se.

It is also important to interpret the non-significant prediction of HR variability (CV%) by our physiological markers. Despite resting HR showing a format-interaction, no marker reliably predicted variability across formats. This suggests that variability of internal load in SSGs depends more on format constraints, game dynamics, player behavior and possibly motivation and tactics more than on baseline physiological status (43). Earlier work reported that while mean %HRmax responses were high and relatively stable (CV around 2.6–3.8%), other metrics such as RPE and blood lactate exhibited larger inter-player variability (44). Hence, practitioners should not rely solely on baseline physiological tests to anticipate variability. Rather, controlling session design (player number, area, bouts) is more important for managing variability.

Several limitations deserve acknowledgement. Generalisability is limited because the sample comprised two amateur teams from the same regional context. Training density, match congestion, and tactical/physical demands can differ substantially in elite environments, which may alter both HR responses to SSG constraints and the interpretability of HR-based monitoring signals. Our sample consisted of amateur male players, which may limit the applicability of our findings to elite or female populations. Although we manipulated number of players (2v2 vs. 4v4), other potential influences (such as pitch size per player, tactical instructions, ball-in-play time) were standardized and may nonetheless have influenced responses. Because CV estimates are sensitive to the number of repeated observations, CV% derived from three bouts should be interpreted as a pragmatic within-session stability indicator rather than a definitive reliability estimate. Moreover, because external-load measures (e.g., distance, accelerations, impacts) were not measured, similar HRmean across formats should be interpreted as similar cardiovascular strain only, not necessarily similar mechanical loading. Future research should manipulate these design variables. We used only heart rate metrics as internal load indicators, thus inclusion of external load measures (GPS, accelerometry) or perceptual measures (RPE) would offer a more complete picture.

From a coaching and sport science perspective, our findings suggest that measuring sub-maximal warm-up HR (HRR) may offer some predictive insight into cardiovascular load—particularly in 4v4 formats—but this predictive utility is limited. Given that HR variability (CV%) is lower in the 2v2 format, coaches seeking more reproducible cardiovascular stimuli across players and sessions might prefer 2v2 for certain training blocks. However, since mean HR did not differ between formats, both 2v2 and 4v4 may be used to elicit comparable average cardiovascular load. Ultimately, session design should emphasize format constraints (player number, pitch size, bout/rest durations) rather than relying solely on baseline physiological tests for regulating cardiovascular stimulus in SSG training.

This study examined whether three physiological markers (resting HR, submaximal warm-up HRR, and Yo-Yo IR1 HRmax) could explain individual differences in cardiovascular responses during two commonly prescribed small-sided game formats in football. Under the present protocol (2v2 vs. 4v4 with near-equal area per player), the specific heart-rate–derived markers examined (HRrest, HRR from a standardized submaximal test, and Yo-Yo IR1 HRmax) explained only a small-to-modest proportion of variance in SSG HR outcomes, and this explanatory value was format-dependent. Across both formats, mean HR was similar, yet 2v2 produced more stable bout-to-bout heart-rate responses than 4v4, indicating that game format exerts a stronger influence on HR variability than on average internal load. Among the physiological markers, only HRR demonstrated meaningful explanatory value, and exclusively in 4v4, where it accounted for a modest portion of HRmean variance. Resting HR and Yo-Yo IR1 HRmax did not predict mean HR or its variability in either format. These findings show that cardiovascular markers have limited utility for predicting internal load during small-sided games, and that their predictive capacity is format-dependent. The consistency of cardiovascular responses appears to be shaped more by structural constraints of the game (likely player number) than by baseline physiological status. Thus, practitioners should rely on careful manipulation of SSG design variables, rather than on pre-session physiological measures, when the objective is to control or anticipate cardiovascular load in football training. Although the present findings are performance-focused, improving the precision of day-to-day load management through SSG design may also have downstream relevance for athlete health by supporting injury-prevention and public-health–aligned objectives (e.g., minimizing excessive training stress and avoidable time-loss from sport).