A total sample of 30 endurance runners (37.0 ± 9.7 years) participated in this investigation. The Clinical Research Ethics Committee of the Local Institutional Review Board (PI/012-20) approved this study according to the Declaration of Helsinki. All participants signed the informed consent as required. This investigation consisted of two-part complementary studies performed within a month.

Acceleration and orientation signals were recorded and analyzed with a proprietary software (Movalsys, Navarre, Spain). Data transformation and further analyses were implemented with MatLab R2016a (MathWorks Inc.; Natick, MA, USA). A fusion algorithm computes a 3D space orientation of the sensor-fixed reference frame S with respect to an Earth-fixed reference frame E. The analysis requires the acceleration component in the forward direction of the run, rotating frame E so that the y-axis points heading direction. The sampling rate was set at 120 Hz, and a 25 Hz low-pass filter with a zero-lag Butterworth was applied. Twenty seconds at each speed was analyzed to ensure a minimum of 10 strides per speed (36). The data of participants on the first day from “Study I” were used for “Study II.”

Mean and standard deviation (SD) were calculated for each measure. The data were screened for normality of distribution and homogeneity of variances using the Shapiro–Wilk and Levene's tests, respectively, and when appropriate sphericity (Mauchly's test). Paired t-test and mean bias ± limits of agreement (LoAs) based on t distribution with n-1 degrees of freedom (Bland and Altman plots) were used to analyze differences between the two testing sessions (“Study I”). Linear regressions from Bland and Altman plots (Appendix I) were calculated to study the systematic error in the distribution of the data points. One-way analysis of variance (ANOVA) with Bonferroni's adjustment was performed to detect differences among speeds. The reliability analyses included the following calculations: The intraclass correlation coefficient (ICC) (2,1), two-way mixed-effects model and absolute agreement with 95% confidence intervals (CI); the standard error of measurement (SEM) of paired data (SEM = SD_diff/√2), the coefficient of variation [CV = (SEM/mean)*100]; and the minimal detectable change (MDC = SEM × 1.65 ×√ 2). ICC values between 0.5 and 0.75 indicated moderate reliability, between 0.75 and 0.9 indicated good reliability, and >0.9 indicated excellent reliability (37). CV was classified as trivial (CV ≤ 5%), small (5%<CV ≤ 10%), and moderate (CV > 10%) (38). Means, CV, and MDC were calculated for 9, 15, and 21 km·h⁻¹ to ensure the inclusion of all 18 participants. In “Study II,” the differences were evaluated by two-factor repeated-measures ANOVA with Bonferroni’s corrections. The differences (e.g., ESs) were interpreted using previously suggested thresholds: small (0.2), moderate (0.6), large (1.2), and very large (2.0) (39). Significance was set at p < 0.05 for analyses not requiring post hoc adjustment. A statistical analysis was performed with SPSS version 20.0 (SPSS Inc. Chicago, USA).

During the maximal incremental running test, no significant differences were found between both testing days in HR_max [183 ± 13 vs. 181 ± 13 beats·min⁻¹; p = 0.16; effect size (ES) = 0.15, small] or MAS (18 ± 1.9 vs. 17.8 ± 1.7 km·h⁻¹; p = 0.26; ES = 0.11, small). The mean absolute values of the IMU variables studied for the selected speeds at each of the test days are shown in Table 1. No significant differences were observed between the first and second tests in terms of trunk displacements and RMS-RES (p = 0.22; trunk displacements, ES < 0.14, small; RMS-RES, ES < 0.49, small). Bland and Altman analyses between both tests for each variable at any speed showed the presence of non-significant systematic errors (p = 0.08–0.99; R²< 0.07; mean bias, −0.33–0.21) (Appendix I). ANOVA analyses showed the presence of random errors in VT displacement (0.43 cm), ML displacement (0.22 cm), RMS ML (0.47 m·s⁻²), RMS anteroposterior (AP) (0.49 m·s⁻²), RMS VT (0.63 m·s⁻²), and RES (0.74 m·s⁻²). Good to excellent ICC values were observed both for the trunk displacement variables (0.85–0.96) and the RMS-RES (0.71–0.94), while all the CVs at 15 and 21 km·h⁻¹ were trivial in the trunk displacements. The CVs in RMS ML and AP were small to trivial and were trivial in RMS VT. RES showed the lowest CV among all the measured variables. ANOVA analyses indicated that ICCs increased between 9 and 21 km·h⁻¹ (p = 0.001; ES > 3.1, very large), while CVs remained similar (p = 0.09) (Table 2).

The ANOVA results showed that as the speed increased, there was a decrease in VT displacement (p = 0.007; r = −0.74; ES = 1.0–2.5, moderate to very large) and an increase in ML displacement (p < 0.001; r = 0.68; ES = 1.8–2.3, large to very large). RMS increased with increasing speed in the ML and AP axes (p < 0.001; r = 0.79 and 0.90, respectively; ES = 1.3–4.6, large to very large) but remained unchanged in the VT axis (p = 0.23; r = 0.24; ES < 0.54, small). RES values increased with increasing speed (p = 0.002; r = 0.77; ES = 1.4–2.6, large to very large).

The male runners showed greater body mass (p < 0.001; ES = 2.8, very large), height (p < 0.001; ES = 3.4; very large), and MAS (p = 0.015; ES = 1.1, moderate) than those of the female runners. Absolute IMUs trunk both displacement and acceleration for both male and female runners as shown in Figure 1. The male runners had greater VT displacements at both 15 and 21 km·h⁻¹ speeds (Δ% = 11.1–14.6%; p = 0.024–0.037; ES = 0.90–1.0, moderate). In contrast, the female runners showed greater ML displacement at 9 km·h⁻¹ (Δ% = 33.4%; p = 0.001; ES = 1.6, large), as well as both RMS ML and RMS AP at both 15 and 21 km·h⁻¹ speeds (Δ% = 15.–6.1%; p = 0.005–0.048; ES = 0.83–1.9, moderate to large), and greater RES at 21 km·h⁻¹ (Δ% = 16.3%; p = 0.006; ES = 1.2, very large). No significant differences were found in RMS VT between the male and female runners (p = 0.14–0.96). The percentages of differences between sexes were all greater than the MDC reported (Table 2).

VT and ML displacements showed good to excellent reliability according to the high ICCs (range, 0.85–0.96) and relatively small CVs (<6.5%). The reliability was slightly higher in VT (CV < 4.3%) than in ML (CV < 6.5%) and was greater at higher speeds. Highly regular and reliable VT displacement could partially be explained by the fact that the VT direction of run is continuously constrained to gravity (19), and thus, less subjected to system noise than the ML displacement. According to the MDC (Table 2), a runner's change after an intervention should be considered significant if it exceeds 10% (VT) or 15% (ML) of the baseline values at 9 km·h⁻¹ and 8% (VT) or 10% (ML) at 21 km·h⁻¹.

The RMS-RES variables showed moderate to excellent reliability and a small error (ICCs range, 0.74–0.94; CV < 6.1%). These values were similar in magnitude to those observed in the trunk displacements and were slightly more reliable in RES (CV < 4%) and RMS VT (CV < 5%) than in RMS AP or ML (CV < 6%). The reliability values presented here are inferior to those previously reported for male endurance athletes running on a treadmill (17). However, these authors calculated a single CV for all the speeds tested together (between 2 and 16 km·h⁻¹). This differs from the present study in that reliability was calculated at each speed and found to vary across the different speeds. Moreover, a previous study showed higher reliability values than our study in a mixed-sex group of college-aged subjects running on a treadmill (28). However, their subjects performed both tests on the same day, separated by 60–180 s, whereas in our study both tests were separated by 7–10 days. In our study, a longer test–retest delay may increase within-subject random or systematic variation, potentially worsening reliability values compared to the within-day variability assessed (28). Based on the MDC (Table 2), the smallest amount of change that may reflect a true or relevant change in the economy-based variables after an intervention varies from 9 to 14% at 9 km·h⁻¹ and from 6 to 11% at 21 km·h⁻¹. RES needs the lowest amount of change to detect relevant changes (<6%). The results indicate that the MTw IMU is as reliable as other accelerometry-based devices and it could be used to analyze changes over time, considering the error reported in this study.

In the present study, VT values significantly decreased with higher speeds, from 10.2 cm at 9 km·h⁻¹ to 6.8 cm at 21 km·h⁻¹. These VT values agree with other studies conducted in mixed-sex groups and male recreational runners (12, 19, 22, 31) and international-level male athletes during treadmill or overground running (12, 19, 20, 22, 42), in which VT values were measured using optoelectronic motion analysis, sacral accelerometers, or a single digital video camera (12, 19, 20, 22, 42). The lower VT displacement in female runners could indicate lower VT loading rates and mechanical power per step but also per minute of running (20). This lower VT displacement is suggestive of decreased VT peak force (43), lower extremity joint energy absorption (22), impact shock when the foot hits the ground, and increased running economy (19, 22, 44).

ML displacement increased with increasing speeds in both groups, from average values of 2.1 cm at 9 km·h⁻¹ to 5.3 cm at 21 km·h⁻¹. However, females had higher ML displacement than males at the lowest speed (9 km·h⁻¹). The higher ML displacement, combined with the lack of differences observed in VT displacements between sexes at this speed (Figure 1), and the similar stride frequencies between sexes found in previous studies during running at low speeds (45–47) suggest that females exhibited greater lateral oscillations and greater trunk motion per step and per time unit compared to males. However, at 15 and 21 km·h⁻¹, the ML displacements were similar between sexes as shown in Figure 1, but females had lower VT displacements than males. It is suggested that contrary to what was found at the slowest speed, females exhibited less trunk motion per step than males at higher speeds.

On the other hand, RES values significantly increased with higher speeds, ranging from 13.2 m·s⁻² at 9 km·h⁻¹ to 22.2 m·s⁻² at 21 km·h⁻¹. The RES values agree with a study conducted on healthy college-aged individuals of mixed sex who run on a treadmill while wearing an accelerometer placed in line with the top iliac crest (28). However, they disagree with the values previously reported by the same research group in a study performed with male runners measured during treadmill running using the same accelerometer (17). A potential explanation for this inconsistency could be attributed to the methodology. Their formula did not include the number of samples in the dataset inside the square root to calculate RMS. The formula was rectified in their subsequent study which is consistent with our findings (28). In addition, the observed increase in absolute RES with higher speeds has also been observed by others (42). This increase is predominantly due to an increase in the AP (143%) and the ML (145%) axes, particularly in females, rather than in the VT (9%) axis (Figure 1), which suggests an increase in energy expenditure (17). This may explain why the female runners showed higher RES (Figure 1) than male runners. Previous studies also found higher RMS ML in female runners during treadmill running at different speeds between sexes (11.1 vs. 9.6 km·h⁻¹) (31). The higher RMS ML may be translated into higher ML loading rates during ground contact which are not useful for propulsion (25) and may result in both an increase in energy expenditure and a decline in running economy. This is also supported by another study (48), which hypothesized that female runners present a reduced neuromuscular control in the musculature of the sagittal plane to decelerate the body during the stance phase and rely more on the frontal plane. However, the novel results of this study suggest the presence of acceleration components that increase the signal power with increasing speed but do not contribute to a higher range of trunk motion (e.g., ML displacements at 15–21 km·h⁻¹) (Figure 1). This would mean that trunk displacements alone could not allow general conclusions between sexes in running gait and either some spatiotemporal and/or kinetic variables should be brought into the equation. Thus, caution should be taken when runners are not assessed using all the spectrum variables measured by the IMUs.

Some limitations need to be addressed. First, the incremental running protocol considered speeds up to 21 km·h⁻¹ during treadmill running. Thus, the present findings are likely to apply to the speeds and running surface used. Second, accelerations were analyzed for 20-s at each speed. These 20-s were considered sufficient to ensure the analysis of a minimum of strides in all participants. However, the higher average number of steps analyzed at 21 km·h⁻¹ (−70) compared with 9 km·h⁻¹ (−54) could somehow explain the better reliability found at higher speeds. Third, the implication of changes in signal power which do not imply changes in the trunk motion at a certain axis needs further study in terms of both performance and running-related injuries.