Participants

Seven competitive swimmers (4 men and 3 women; body height: 179.0 ± 5.9 cm; body mass: 69.05 ± 5.05 kg; age: 19.97 ± 2.18 years; sum of 7 skin folds: 60.98 ± 22.37 mm) volunteered to participate in this study. All were members of the Spanish Senior National Team and had 6–8 years competitive experience, with their average personal best times reaching 803 ± 45 points, according to the World Aquatics points (WAPs). According to McKay et al.'s (2022) classification framework, these personal best times corresponded to a classification competitor of level 4 (international level). The study adhered to the Declaration of Helsinki guidelines and was approved by the Institutional Ethics Committee of the Facultad de Ciencias de la Actividad Física y del Deporte (INEF), Universidad Politécnica de Madrid, Madrid, Spain (protocol code: 20240207; approval date: 13 January 2025). For participants who were minors, written informed consent was obtained from their legal guardians and all of them and their coach provided written approval for retrospective analysis.

Measures

Before this study began, each swimmer's best time in their main event during the major competition of the previous season (MT_p) was collected and expressed as WAPs. Then, during a 40-week training period from the beginning to the end of the 2022–2023 season (September to July), swimmer training data were registered. The season included three macrocycles that covered three main competitions: the winter national championships, the spring national championships, and the summer national championships—all held in a 50-m pool, with the water temperature varying between 26.6 and 27.3°C (Figure 1). Swimmers typically completed nine in-water training sessions per week (2 h each) and four supervised dry-land training sessions per week (75 min each), following standardized team warm-up routines, nutritional guidance, and medical support under the supervision of experienced national-team staff. Training adherence was consistently high (>98%) for both in-water and dry-land sessions, with minimal inter-individual variability, ensuring uniform exposure to the prescribed training program. The menstrual cycle of the female participants was not monitored in the present study, and, given the small sample size, it was unlikely to influence the training load-performance relationships across a full competitive season.

Figure 1

Illustration of a 40-week season structured into three macrocycles, each associated with the major competition.

The first macrocycle spanned a 13-week training block, the second covered a 14-week training block, and the third included a 13-week training block, with the main competitions held in the last week of the macrocycles. In these macrocycles, the performance of the swimmers in their main events was evaluated. These performances were quantified using WAPs and recorded as Meet Time (MT) 1, 2, or 3, respectively.

Following Foster et al.’s (2001) recommendations, participants received standard instructions to report their RPE within 30 min after each training session using the CR-10 scale (Borg et al., 1987). The volume of each training session per swimmer, both in water and during dry-land training, was quantified as time (in minutes) and expressed as distance (in km) for swimming sessions and load (in kg) for dry-land sessions, respectively, and recorded in an Excel sheet.

Design and Procedures

A descriptive design of a collective case series, as previously used in a study on a group of athletes who underwent the same or similar training regimes over a certain period (Hellard et al., 2019), was used with the group of international-level swimmers. Although each swimmer followed a specific training program, the research design was primarily observational, as removing such training from a control group was neither ethical nor possible in such trained populations. Therefore, the training programs were designed and implemented by the coaches without the researchers' interference.

The ITL was quantified using the _SRPE method, as described by Foster et al. (2001). According to this method, the _SRPE was calculated by multiplying the RPE of the swimmer by the duration (in minutes) or volume (in km for swimming and kg for dry-land sessions) of each training session (Dhahbi et al., 2024a). The dry-land training load was quantified as the total lifted volume (in kg), calculated as the product of the load, repetitions, and sets for each exercise. This method does not reflect the relative intensity of the exercise performed by each swimmer (Dhahbi et al., 2024b). Consequently, this metric should be interpreted primarily as an individualized, practical indicator of the ITL rather than an absolute measure of training intensity. The results were quantified as the in-water training load in km (W__SRPE_km) and minutes (W__SRPE_min), and the dry-land training load variables in kg (D__SRPE_kg) and minutes (D__SRPE_min). Weekly _SRPE values were determined by adding the _SRPE values of all sessions in the week and reporting the final values in arbitrary units (AU). Wallace et al. (2009) validated this approach for swimming.

Furthermore, consistently with the Murray et al.'s (2016) study, the ACWR was used as an index of athlete’s training readiness, using exponentially weighted moving averages to compare training loads completed in a recent training block (seven days) with the chronic training load completed over a period of four weeks. Exponentially weighted moving averages were calculated using a decay factor defined as λ = 2 / (N + 1), where N represented the time window. A coupled ACWR approach was adopted, whereby the acute workload week was included in the chronic workload calculation. The moving average model was used to assign the athlete's earlier workloads with progressively diminishing weight compared to training sessions completed more recently, which represented a greater weight. This resulted in the following variables: the in-water and dry-land ACWR in minutes (W_ACWR_min and D_ACWR_min, respectively), the in-water ACWR in km (W_ACWR_km), and the dry-land ACWR in kg (D_ACWR_kg). The change in performance (Δ) was quantified using WAPs as the difference between the swimmer's best time in the main competition of the current training cycle (MT) and their best time in the main competition of the previous training cycle (MT_p).

Statistical Analysis

Training and competitive data were statistically analyzed using R software (v.4.3.0 for Windows). Descriptive statistics, including mean ± SD values and dispersion coefficients, were calculated for all variables. According to the objectives of this study, the training load indicators (the _SRPE and the ACWR) with the personal best times of the previous training cycle (MT_p) were used to predict the change in swimmers’ performance at the end of each macrocycle (ΔMT₁, ΔMT₂, and ΔMT₃). As a preliminary analysis, a correlation matrix for the training load indicators (W__SRPE_km, D__SRPE_kg, W__SRPE_min, D__SRPE_min, W_ACWR_km, D_ACWR_kg, W_ACWR_min, and D_ACWR_min) was built to check for multicollinearity in the performance regression models, as multicollinearity could cause some explanatory variables to be non-significant.

Then, given the repeated-measures structure of the data (three macrocycles per swimmer), linear mixed models (LMMs) were used to account for within-subject dependence. The swimmer’s identity was included as a random intercept, while training load indicators (the _SRPE, the ACWR, and MT_p) were included as fixed effects. All continuous predictors were z-standardized (e.g., z_W__SRPE_km7) before the analysis. This was done to capture physiologically meaningful changes related to each swimmer's typical training load while accounting for repeated observations of swimmers. This approach allowed the LMMs to distinguish the true relationships between training load and performance from stable between-swimmer differences, which might explain the discrepancies observed with simpler regression models. Based on tapering physiology and periodization literature (Hellard et al., 2019; Mujika and Padilla, 2003), training-load indicators were summarized over seven weeks before each target competition. This time window was selected prior to the lead-up competition period, during which the training load is typically reduced while intensity is maintained.

Session Rating of Perceived Exertion (_SRPE) and Performance

Training load indicators (D__SRPE_kg and W__SRPE_km) throughout the 40 weeks peaked approximately six weeks before major competitions (weeks 7–10), yielding 12.46 and 24.63 AU, respectively. These peak loads aligned with linear periodization models that progressively built workloads before a taper phase to enhance performance (Mujika and Padilla, 2003). In the weeks immediately prior to the competition (13, 26, and 40), a clear decrease in training load was observed that could be associated with an improved race-day performance (Hellard et al., 2013; Mujika and Padilla, 2003; Seiler, 2010). In fact, during the last four weeks of cycles 1, 2, and 3, W__SRPE_km progressively decreased to 50.0%, 53.3%, and 50.0%, respectively. Similarly, D__SRPE_kg decreased by 54.8%, 38.1%, and 48.2% in cycles 1, 2, and 3. Mujika and Padilla (2003) reported that a significant 50–90% reduction in training volume and training frequency in the lead-up to a competition can induce 0.5–6% improvements in personal best times in cyclical sports. Changes in performance in the present study comprised a 2.3% decrement in cycle 1, but 2.3% and 0.2% improvements in cycles 2 and 3, respectively. This could reflect a lower importance in the competition phase for cycle 1 compared to cycles 2 and 3. At the beginning of the swimming season, swimmers were changing from general to more specific preparation, and race-specific adaptations probably did not occur. Similar patterns have been reported in cyclical sports (Mujika and Padilla, 2003) and swimming (Hellard et al., 2019; Mujika et al., 2002), with better results generally seen later in the season.

During the final seven weeks leading to competition, the LMM analysis showed high levels of simulation-based sensitivity and significant associations between training load indicators and swimming performance. The higher swimming training load expressed as z_W__SRPE_km7 was negatively associated with competition performance (β = −15.5), showing a greater simulation-based sensitivity than the time-based indicator (z_W__SRPE_min7). This statement is in line with established tapering frameworks in swimming (Hellard et al., 2019; Mujika and Padilla, 2003). In particular, the _SRPE is widely used as a practical indicator of the ITL and has been shown to align well with training intensity (García-Ramos et al., 2014) and with heart rate-derived measures and overall training volume (Wallace et al., 2009). However, the present results are the first to associate the swimming ITL with performance of elite, competitive swimmers.

For the dry-land variables, a possible association was observed between z_D__SRPE_kg7 (β = −20.3) and competition performance, suggesting that excessive dry-land training in the final weeks may compromise readiness. Accordingly, reducing the dry-land training load during this period could enhance neuromuscular recovery and facilitate performance gains, thereby ensuring optimal performance in water (Wirth et al., 2022). This represents an important and novel finding of the present study, as no previous reports on competitive swimmers have examined the combined influence of in-water and dry-land training loads on competition performance. This is so despite dry-land training being a consistent component of elite swimming programs. Nevertheless, it is important to emphasize that dry-land training is not detrimental to competition performance. Experimental evidence demonstrates that well-designed, specific strength interventions can improve torque production and correct muscular imbalances (Agrebi et al., 2024). Therefore, the observed associations likely reflect excessive or poorly timed loading rather than the inherent value of strength training.

ACWR and Performance

The weekly ACWR values in the present study (W_ACWR_km: 0.99 ± 0.09; D_ACWR_kg: 0.91 ± 0.12) were maintained within the established “safe” range of 0.8 to 1.3 (Gabbett and Whiteley, 2017; MacMillan et al., 2020) and showed standard deviations consistent with the 10–15% variation that Jones et al. (2016) proposed. The highest values of W_ACWR_km (1.07) and D_ACWR_kg (1.27) were recorded in weeks 7 and 10 prior to the major competition, respectively. These values during the last seven weeks before the competition indicated that the weekly and monthly training loads were consistent and that the drop in training load was not drastic enough to accelerate the downside of the ratio. This was aligned with the guidelines for reducing swimming training volume while maintaining intensity to maintain an effective training stimulus and adequate fitness levels (Hellard et al., 2019; Mujika and Padilla, 2003) and to maintain the ACWR in the preferred range. However, the models, including the ACWR (Table 2c–2d), did not demonstrate stability, showing singular fits and limited explanatory value, particularly when the ACWR was expressed as W_ACWR_min. These findings agree with previous reports on tennis in which longitudinal analyses of junior players showed no significant association between the internal or external ACWR and match outcomes. Although ACWR markers were maintained close to 1, they did not predict match results, suggesting that a balanced workload may be insufficient to explain competitive performance (Myers et al., 2021). This study's results indicated that when ACWR values were close to 1, they were not independently associated with MT performance, suggesting that the ACWR alone may be insufficient to explain the performance outcome in the context of overall training load. It is crucial to understand that the ACWR was primarily used in swimming as an injury management index (Collette et al., 2018) and should not be interpreted merely as a training load metric, but rather as an index of fatigue reflecting the balance between recent and chronic load (Gabbett, 2016; Hulin et al., 2013). The misapplication of W_ACWR_km as a direct prescription for tapering volume reductions could potentially lead to suboptimal performance outcomes.

Relationship between MTp and Performance

In elite sport, the initial performance level constrains the magnitude of achievable improvement (Costa et al., 2010). For instance, Del Castillo et al. (2022) report for elite swimmers that performance gains over one training cycle can predict swimming success in the following cycle (R² = 0.68–0.73). Therefore, the role of previous performance MT_p as a predictor of the competition outcome is statistically and physiologically expected among international-level swimmers. Under these conditions, training-load variables are expected to exert a comparatively smaller influence on performance. However, by modeling performance change between consecutive training cycles (ΔMT) rather than MT_p, the present analysis substantially reduced the confounding effect of stable expertise-related factors, allowing for the clearer identification of training-load effects. Within this framework, this study's findings indicate that MT_p is not merely a retrospective marker, but a good predictor of an athlete's adaptive response to training (Table 2). This underscores the need for coaches to consider athletes' baseline performance when prescribing training loads, as a higher initial capacity may facilitate greater physiological adaptations (Allen et al., 2014; Mujika et al., 2002).

Practical Implications and Limitations

According to this study, integrating W__SRPE_km and D__SRPE_kg into training assessments and session programming can allow coaches to manage the ITL and recovery more effectively, particularly during competitive phases. From an applied perspective, during the seven weeks before a major competition, a gradual reduction in W__SRPE_km with the clear moderation of D__SRPE_kg appears to be beneficial to competition outcomes. In addition, monitoring the ACWR could be useful for describing training load stability; however, this should not be interpreted as a direct measure of training load or competition readiness. Therefore, rather than targeting specific ACWR thresholds for swimming and dry-land training, coaches may benefit more from identifying trends in load management and supporting optimal performance timing. Moreover, the timing of _SRPE peaks relative to competition phases should be carefully managed to align with tapering goals and maximize performance outcomes. Personalized training plans that reflect past performance, set incremental goals, and incorporate intermediate competitions can boost athletes' confidence and create positive momentum.

Future research should aim to replicate this study with swimmers across different events and distance specializations, allowing for a deeper, swimming-specific analysis of how training load metrics interact to influence performance. To further advance athletic performance optimization, contemporary approaches that go beyond traditional volume- and intensity-based paradigms, such as machine learning and AI-driven methods, could enhance personalized training prescriptions and support injury prevention in sports such as swimming (Souaifi et al., 2025).