The research was conducted using a single-group repeated-measures, pre–post-test design, in which each participant was assessed before and after the intervention. All participants completed three test sessions: two test sessions within 2 weeks, in a rested state, prior to the Lidingöloppet race, and one post-test immediately after the race. The first test session, performed 5–10 days prior to Lidingöloppet, consisted of a treadmill running test to measure submaximal and maximal physiological and biomechanical parameters. The second test session consisted of an anthropometric assessment of body composition and was performed 3–7 days following the first test. Post-tests were conducted immediately after the race in a field-laboratory constructed at the finish line of the race. The Swedish Ethical Review Authority approved the study (2022-04035-01), and participants provided verbal and written consent after being informed of any potential risks associated with the experiments prior to participation. An overview of the experimental design is provided in Figure 1.

Thirteen runners (twelve male and one female, average age 35 ± 4 years, 10 km PB = 36:46 ± 03:17 min:ss) were recruited for the study through digital advertising on social media, sports associations, and sports clubs. Participants were included if they were healthy individuals between 18 and 40 years old, willing to participate in the 30 km cross-country race “Lidingöloppet,” and able to verify their 10 km personal best. Individuals were excluded if they had any current or recent injury (within the past 6 months), were smokers, or had a 10 km personal best slower than 40 min, as verified through official race results or GPS files.

Following the baseline assessment, the participants completed the XC running race Lidingöloppet—a 30-km race with a 550-m elevation gain, run on gravel roads and grass on Lidingö island (Stockholm, Sweden) at sea level. During the race, the runners aimed to perform maximally. Performance during the race was measured as the time to completion from the race start, based on the official chip time. The temperature during the race was ∼14 °C, with no rainfall.

At the finish line of the race, an in-field laboratory was set up inside a tent to ensure stable environmental conditions. Following race completion, the participants immediately entered the in-field laboratory for the measurement of body mass. Thereafter, the athletes completed 5-min of submaximal running at 14 km h⁻¹. For the measurement of physiological and biomechanical variables, the same equipment and standardization procedures as in the pre-test were used. Only one athlete could be tested at a time; therefore, participants were tested in the order they finished the race. This resulted in an average waiting time of 45:15 ± 24:28 (min:ss) following race completion.

Data were tested for normality using the Shapiro–Wilk test. For analysis of variables related to performance, a hierarchical multiple linear regression model was used to explain variance in performance. Independent variables were added in order of their correlation to performance. To limit collinearity in the model, the variance inflation factor and tolerance of the regression model were calculated, with an upper limit of 2.0 for the variance inflation factor and a lower limit of 0.5 for tolerance. The regression model resulting in the highest R ² without collinearity between variables will be presented. Correlation analyses were performed between physiological variables, Stryd biomechanical variables (power output, ground contact time, vertical oscillation, leg spring stiffness, cadence, and step length), and performance using Pearson’s r. Differences between pre–post tests for physiological and biomechanical variables at 14 km h⁻¹ were analyzed using paired sample t-tests for normally distributed variables and Wilcoxon rank for skewed variables. To study the relationship between changes in biomechanical variables and changes in RE, computed Δ variables (post−pre) for RE and biomechanics were analyzed using Pearson’s r correlation. If skewed, Spearman’s rho was used. To further investigate changes in RE, correlation analyses using Pearson’s r were performed between ΔRE and pre-test measures of VO₂max (mL·min^-1·kg^-1), VO₂ at θLT, RE_ox, RER, and race time (s). Furthermore, correlations between ΔRE and ΔRER, ΔVE (L·min^-1), Δwb (kcal·kg^-1·km^-1), and Δbody mass (kg) were analyzed. Statistical significance was set at p ≤ 0.05. Data are presented as the mean ± SD, if not stated otherwise. All analyses were performed using Jamovi (2.3.21, The Jamovi Project, 2022).

The participants ran the race with an average speed of 13.8 ± 1.7 km h⁻¹, resulting in an average finishing time of 02:12:20 ± 00:14:48 (hr:min:ss), 35% ± 15% slower than the winning time and with a race placement of 398 ± 342.

Pre-test anthropometric and physiological data from both submaximal and maximal tests, along with the results of the correlation analysis between each variable and race time, are presented in Table 1. The only anthropometric variable related to performance was body fat mass (kg) (r = 0.646; p = 0.032). All other anthropometric variables were not related to race performance. All the pre-test physiological variables considered were correlated with the race time under both conditions (submaximal and maximal), except for RE_ox. No significant correlation was found between the race time and power output, ground contact time, vertical oscillation, leg spring stiffness, cadence, or step length measured using the Stryd device.

The multiple regression model, with the race time as a dependent variable resulting in the highest R ², is shown in Table 2. After controlling for collinearity, the best regression model included the independent variables Fat % at 14 km h^-1 and VO₂ at θLT and RE_allometric. This model explained 91.8% of the variance in the race time (R ² = 0.918, adjusted R ² = 0.887, F = 29.7, and p < 0.001). The results of the classical endurance performance model, including VO₂max, RE_ox, and VO₂ at θLT, are presented in Table 3. Collinearity between variables existed, and R ² did not reach above 0.918 (R ² = 0.834, adjusted R ² = 0.772, F = 13.4, and p = 0.002). Although the studied group was fairly homogeneous, with most ages clustering between the early and late 30s and a CV of only ∼11%, additional analyses including age as a covariate showed that age was not a significant predictor of the race time (p = 0.566) and did not improve model fit (adjusted R ² = 0.877 vs. 0.887 without age).

All participants were tested after finishing the race. Physiological variables measured in both pre- and post-tests are presented in Table 4. Post-race VO₂ relative to body mass (mL· min⁻¹·kg⁻¹) increased significantly by 3.4%, whereas absolute VO₂ (L·min⁻¹, −0.73%) and VE (4.55%) showed no significant changes. The race also negatively affected RE_ox, which increased by 3.4% from pre- to post-race (Figure 2). However, when running economy was allometrically scaled (RE_allometric), it did not differ significantly (2.5%). Furthermore, RE_ec also showed no significant difference between pre- and post-tests (2.1%). Substrate utilization (Figure 3) changed significantly post-race with decreased RER (−6.18%), increased Fat% (62.22%), and decreased Carb% (−25.32%).

None of the MotionMetrix biomechanical variables, such as cadence (1.2%), ground contact time (−2.3%), vertical displacement (−1%), overstride (−6.7%), or vertical force (2.6%), differed significantly between pre- and post-test for any variable (p > 0.05 for all the variables).

To examine whether differences in waiting time between the end of the race and the post-test influenced changes in RE, a Pearson’s correlation analysis between ΔRE_ox and waiting time was conducted. The results of this analysis showed that there was no significant correlation between ΔRE_ox and waiting time (r = 0.039; p = 0.898).

Participants’ pre-test measurements of VO₂max (mL·min⁻¹·kg⁻¹; r = 0.181, p = 0.573), VO₂ at θLT (r = 0.253, p = 0.428), and RER at 14 km h⁻¹ (r = −0.178, p = 0.560) did not significantly relate to ΔRE_ox. Furthermore, participants’ performance did not relate to ΔRE_ox (r = −0.440, p = 0.132). Moreover, no significant relationships between ΔRE_ox and ΔWB (r = 0.470, p = 0.105), ΔVE (r = 0.481, p = 0.096), Δbody mass (r = −0.493, p = 0.087), ΔH-la (r = 0.496, p = 0.084), and ΔRER (r = −0.388, p = 0.191) were observed.

No significant relationships were found between changes in biomechanical variables obtained from MotionMetrix and ΔRE_ox. ΔCadence (r = 0.215, p = 0.480), Δ ground contact time (r = 0.234, p = 0.442), ΔOverstride (r = −0.250, p = 0.411), Δvertical force (r = −0.174, p = 0.569), and Δvertical displacement (r = −0.286, p = 0.343) did not relate to ΔRE_ox.

Consistent with the classical endurance performance model (Bassett and Howley, 2000), performance in the 30 km XC Lidingöloppet correlated with VO₂max, fractional utilization (speed and VO₂ at θLT), speed at RCP, and RE (allometrically scaled oxygen and energy cost), but not RE without scaling. Additionally, race performance was correlated to substrate utilization at submaximal speed and fat mass. However, a regression with only VO₂max, RE, and fractional utilization was limited by multicollinearity (R ² = 0.834), whereas a model including fat% at 14 km h⁻¹, VO₂ at θLT, and RE_allometric explained more variance (R ² = 0.918).

Although VO₂max correlated strongly with performance, it was excluded from the optimal regression, somewhat contradicting previous trail-running studies (Ehrström et al., 2018; Alvero-Cruz et al., 2019; Pastor et al., 2022). As described by Costill et al. (1973), VO₂max is not a good predictor among athletes with similar VO₂max; in our cohort, values were homogeneous, except for two outliers. Thus, VO₂max did not add explanatory power, despite its correlation. Likewise, vVO₂max—linked to VO₂max and RE and shown to predict trail-running performance (Alvero-Cruz et al., 2019; Coates et al., 2021; Sabater-Pastor et al., 2023)—also correlated with performance but was excluded. Nonetheless, VO₂max and vVO₂max remain important determinants of performance.

RE_allometric and RE_ec correlated with performance and featured in our regression, contrasting earlier trail-running studies of long races (Ehrström et al., 2018; Alvero-Cruz et al., 2019; Pastor et al., 2022; Coates et al., 2021) and short races (Björklund et al., 2019). Those studies, except for that by Ehrström et al. (2018), who measured incline RE, assessed RE on level terrain despite greater elevation gain and technical demands, which may have obscured its impact. Longer races may prioritize fractional utilization and muscle preservation over RE (Millet et al., 2012; Pastor et al., 2022; Coates et al., 2021). Lidingöloppet’s moderate elevation and technicality resemble hilly road races, where RE strongly predicts marathon performance (Barnes and Kilding, 2015). Importantly, only allometrically scaled oxygen cost was related to performance, implying that VO₂/body mass is not strictly proportional (Svedenhag, 1995) across a wide mass range (the body mass range in this investigation was 54–83.6 kg). Additionally, RE_ec correlated with performance, underscoring the role of substrate utilization in RE (Fletcher et al., 2009).

θLT and RCP were strongly related to performance; VO₂ at θLT featured in the optimal model, aligning with the study by Scheer et al. (2019). Conversely, VT did not predict performance in other trail races (Ehrström et al., 2018; Alvero-Cruz et al., 2019; Pastor et al., 2022), likely due to greater technical difficulty, elevation gain, and longer race duration, where low intensity diminishes the influence of VT (Millet et al., 2012). In Lidingöloppet’s shorter, less severe course, participants ran at 93% ± 5% of vθLT, indicating race pace and performance VO₂ near vθLT and VO₂ at θLT.

Substrate utilization correlated with performance; higher fat and lower carbohydrate oxidation at submaximal speeds likely spare glycogen. Pastor et al. (2022) also confirmed the importance of substrate utilization.

Fat mass was positively related to performance, consistent with previous studies (Alvero-Cruz et al., 2019; Pastor et al., 2022; Björklund et al., 2019). Higher fat mass increases gravitational work without enhancing capacity, explaining this relationship.

Parameters from the Stryd device (power output, ground contact time, vertical oscillation, leg spring stiffness, cadence, and step length) did not correlate with performance and were excluded, simplifying practical assessments by relying solely on metabolic-cart data.

RE_ox significantly increased post-race, while absolute oxygen cost, RE_allometric, and RE_ec remained unchanged.

These changes reflect increased fat reliance: a 62.22% increase in fat utilization demands more oxygen per energy yield, increasing RE_ox, while RE_ec remains stable. RE_allometric also remains constant due to non-linear body mass scaling. Post-race, H-la increased, RER decreased (reflecting altered substrate use), WB increased, and body mass decreased. VE and biomechanical parameters did not change, although biomechanical responses varied individually.

Absolute oxygen cost did not increase post-race; instead, reduced body mass elevated relative oxygen cost and RE_ox. Allometric scaling neutralized this change, suggesting that body mass influences relative oxygen cost (Svedenhag, 1995), although ΔRE_ox was not significantly related to Δbody mass, contradicting this notion. Maintaining the same absolute oxygen uptake while carrying less body mass inevitably increases RE_ox (mL·kg^-1·km^-1), which can give the appearance of poorer running performance. At the same time, the substantially decreased RER implies greater fat utilization, a substrate that requires more oxygen per unit of energy produced than carbohydrate oxidation. This indicates that the post-race increase in RE_ox may reflect the combined influence of body weight loss (via sweat and glycogen depletion) and a substrate shift toward greater fat oxidation once glycogen availability is reduced. However, ΔRE_ox did not correlate with ΔRER. Other studies show large RER shifts with increased energetic cost (Kyröläinen et al., 2000; Nicol et al., 2007; Sabater Pastor et al., 2021), likely due to longer, more demanding races, although short trials also report notable oxygen cost increases (Candau et al., 1998). Variations in race elevation, duration, population, and test conditions likely explain discrepancies. The lack of significant VE change and absence of ΔRE_ox correlations with ΔVE or ΔWB are in contrast with prior findings linking ΔVE and ΔRE (Thomas et al., 1995). Moreover, unchanged biomechanics and their non-association with ΔRE_ox align with studies showing no relation between RE and biomechanical changes (Hunter and Smith, 2007; Kyröläinen et al., 2000; Nicol et al., 2007). Thus, RE_ox changes might be likely driven by body mass loss (sweating and glycogen depletion) and decreased RER, requiring more oxygen for fat oxidation.

The performance level did not correlate with ΔRE_ox, consistent with findings of Sabater Pastor et al. (2021), indicating that performance does not influence the degree of RE degradation in long trail races or XC races such as Lidingöloppet. This contradicts hypotheses linking RE deterioration and “fatigue resistance” (Jones et al., 2021; Brueckner et al., 1991), warranting further studies. Aerobic fitness, measured as VO₂max, θLT, and submaximal running substrate use, also did not influence ΔRE_ox. Pre-test RE_ox negatively correlated with ΔRE_ox: runners with higher initial oxygen cost changed the least, and only two with high pre-test RE_ox reduced cost post-race (Figure 2). Notably, the top performer had the best pre-race RE_ox but the largest worsening (+23%), suggesting that higher initial RE_ox reserves allow a greater margin for decline. This variability underscores the importance of individual responsiveness post-race. This challenges the principle of Jone and s (2024) that the best resilience (least decline in critical power/velocity) confers an advantage. Although we did not consider the increase in energy cost at the individual critical speed or changes in critical power/velocity, our findings suggest that the subject with the best RE_ox in a fresh state might still have an advantage despite the post-competition decline. However, the heterogeneity of our sample makes it difficult to draw definitive conclusions, and further investigation is warranted.

There are some methodological limitations in the present study that need to be considered and discussed. The only speed tested that was common to all runners included in the study was 14 kmh^-1; therefore, it was not possible to capture physiological and biomechanical responses post-race at different intensities. During the post-tests, waiting time before testing was minimized, resulting in a practical upper limit on the number of participants. On average, post-race measurements were delayed by 45:15 ± 24:28 (min:sec). This relatively long and inconsistent waiting period may have allowed participants to recover to varying degrees, potentially affecting the recorded physiological values and masking the true magnitude of immediate fatigue or metabolic disturbances. To avoid waiting times exceeding ∼45 min, approximately 15 participants could have been included, depending on the ranges in finishing times. With a sample size of 13, however, the statistical power to detect moderate effect sizes (Cohen’s d = 0.6) is low (0.512), and large effect sizes (Cohen’s d = 0.8) are required for sufficient power (>0.8). This suggests a likely large risk of type 2 errors in this study, indicating that moderate effect sizes do not reach significance due to the sample size. Accordingly, observed non-significant changes with moderate effect sizes should be interpreted with caution before conclusions are drawn. However, previous studies (Candau et al., 1998; Hunter and Smith, 2007; Kyröläinen et al., 2000; Nicol et al., 2007; Sabater Pastor et al., 2021; Lazzer et al., 2015) investigating the effect of fatigue caused by running on RE had sample sizes of 7–26. Another limitation in the sample size was that the group included one elite runner who was an outlier. When this elite runner is excluded, the regression model’s explained variance decreases from 91.8% to 84.9%, indicating that the presence of this outlier has an impact on our analyses. The group otherwise included runners ranging from amateur to well-trained, which limits the conclusions that can be drawn regarding the effects of fatigue in elite runners and the differences in responses between elite and well-trained athletes. However, the observed relationships did not differ much from previous research, and the participants’ results were logical and therefore included. It is also likely that adding more data points from a broader range of athletes would yield similar results. In our sample, there was only one female athlete; therefore, the results may not be representative of the female population, although the exclusion of this subject did not alter the variance explained by the model, which remained at 91.8%. Finally, the lack of a control group restricts the ability to establish causal relationships. Future studies should include larger sample sizes to increase statistical power, address sex differences, and explore the potential deterioration of running economy at different running speeds.