How long is required to undertake step variability analysis during running? A pilot study

1. Introduction

Step variability seems to be related to both injuries [1, 2] and endurance performance [3]. Nevertheless, the accurate assessment of step variability remains problematic. In 1995, Belli et al. [4] indicated that step variability during running was difficult to estimate due to the lack of measurement devices. Today, many devices provide real-time feedback on spatiotemporal parameters while running (e.g., OptoGait™, Stryd™ or Myotest™). Therefore, the limitation is not how to collect the data but how long should last the data collection to obtain accurate assessments of step variability.

Belli et al. [4] suggested that 32–64 consecutive steps are required to assess step variability, which represents approximately 15–20 s when running at submaximal velocities. To the best of the authors’ knowledge, the evidence available about how many steps or how long should last the data collection to obtain accurate assessments of step variability is limited, and no more studies have reconsidered this topic adapted to the new devices. However, some studies have addressed this analysis during walking [5, 6, 7]. A previous work examined the minimum number of steps required to accurately estimate spatial and temporal step kinematic variability of subjects walking on a treadmill [5], concluding that at least 400 steps are required. To further examine the step variability during running, the aim of this study was to determine the minimum time required for assessing spatiotemporal variability during continuous running on an instrumented treadmill. The authors hypothesised that the variability in spatiotemporal gait parameters during running would be very similar in short recording intervals (e.g. 10 s, 20 s, 30 s, 60 s or 120 s) compared to a longer recording interval (180 s).

2. Methods

3. Results

The ANOVAs showed no significant differences in the magnitude of the spatiotemporal parameters between the recording intervals (CT: p = 0.454, F = 0.998, partial Eta 2 = 0.263; FT: p = 0.396, F = 1.117, partial Eta 2 = 0.285; SL: p = 0.433, F = 1.039, partial Eta 2 = 0.271; SF: p = 0.671, F = 0.643, partial Eta 2 = 0.187). An almost perfect association was also observed in the magnitude of the four spatiotemporal parameters between the recording intervals (ICC $>$ 0.90) (Table 3).

The ANOVAs conducted on the within-participants SD revealed significant differences for FT ( p = 0.007, F = 5.153, partial Eta 2 = 0.648) and SF ( p = 0.035, F = 3.320, partial Eta 2 = 0.543), while no significant differences were observed for CT ( p = 0.172, F = 1.825, partial Eta 2 = 0.395) and SL ( p = 0.167, F = 1.851, partial Eta 2 = 0.398). Regarding the ANOVAs conducted on CV% values, significant differences were observed for FT ( p = 0.025, F = 3.680, partial Eta 2 = 0.568) and SF ( p = 0.036, F = 3.283, partial Eta 2 = 0.540), but not for CT ( p = 0.155, F = 1.916, partial Eta 2 = 0.406) and SL ( p = 0.172, F = 1.825, partial Eta 2 = 0.395). Bonferroni pairwise comparisons never reached statistical significance. An almost perfect association between the recording intervals was observed for both the within-participants SD (ICC $>$ 0.90; Table 3) and the CV% (ICC $>$ 0.90; Table 3).

Bland-Altman plots revealed heteroscedasticity of error for the FT (0–10 s vs. 0–180 s, r 2 = 0.222; and 0–60 s vs. 0–180 s, r 2 = 0.199) and for the SL (0–10 s vs. 0–180 s, r 2 = 0.115), while heteroscedasticity of error was not observed for the rest of the recording intervals of those spatiotemporal parameters, and for CT and SF ( 0.1). Longer recording intervals yield smaller systematic bias, random errors, and narrower limits of agreement for each spatiotemporal parameter.

4. Discussion

The results demonstrated that the duration of the recording interval plays an important role in the accuracy of the measurement (i.e. variability in spatiotemporal gait parameters), with longer intervals (180 s) showing smaller systematic bias and narrower limits of agreement than shorter intervals (10 s, 20 s, 30 s, 60 s or 120 s). However, based on the analysis of the magnitude of the differences (i.e. ICC) between shorter and longer recording intervals, the authors suggest that in some contexts where the accuracy requirements are not maximum and time-efficient methods are needed (e.g. clinical setting or big groups of athletes), shorter intervals allow sufficient accuracy to assess step variability. Additionally, to correctly interpret these results it is important to note that might be restricted to trained endurance runners [3] and to a running protocol performed on a treadmill at a fixed submaximal velocity [11, 18].

Despite the importance of step variability, very few studies have focused on determining how many steps are required to accurately estimate spatial and temporal step variability during running. Indeed, the authors found just one study from 1995 [4], which examined step variability in terms of step duration and vertical body displacement. Belli et al. [4] concluded that 32–64 consecutive steps ( ∼ 20–25 s) are required to accurately estimate step variability during running at submaximal velocities. Since our findings are expressed in time intervals (s) and in order to make comparisons easier, it is worth noting that 10 s of recording interval equals 27–30 step/min, 20 s equals 54–60 step/min, 30 s equals 81–90 step/min, 60 s equals 163–177 step/min, 120 s equals 326–354 step/min and 180 s equals 489–531 step/min (i.e. on average, participants run at 12 km/h with a SF of 170 ± 7 step/min).

Despite methodological differences, the current results are partially in line with those reported by Belli et al. [4], highlighting that mean spatiotemporal gait parameters during running, and variability in those parameters, might be assessed through the data collected over a time period of only 10 s (i.e., ∼ 30 steps) or, as long as the accuracy requirements are not maximum. Nevertheless, the Bland-Altman method conducted in the current study showed that longer recording intervals yield smaller systematic bias, random errors, and narrower limits of agreement so, if maximum accuracy is required (e.g., scientific approach), longer recording periods (e.g. 180 s), with a greater number of steps (e.g. ∼ 500), must be used. These results seem to be consistent with the results reported by previous studies focused on the analysis of step variability during walking. Owings and Grabiner [5] indicated that at least 400 steps are required to accurately estimate step variability, König et al. [6] proposed the collection of at least 50 step cycles, whereas Bruijn et al. [7] concluded that longer data series ( $>$ 300 steps) led to more precise estimates of variables referred to step variability.

Finally, some limitations need to be considered. First, the footwear was not standardized, but all runners wore their own footwear to increase the ecological validity of the study. Second, the protocol itself, with participants running on a treadmill at a fixed submaximal velocity. Notwithstanding these limitations, the authors consider that this study provides an answer to the initial question and it leaves some unanswered questions about the influence of the running velocity or level of exhaustion on step variability and the time required to accurately measures it.

In summary, the data suggest that the duration of the recording interval plays an important role in the accuracy of the measurement (i.e. variability in spatiotemporal gait parameters during running at 12 km/h), with longer intervals (180 s) showing smaller systematic bias and narrower limits of agreement than shorter intervals (10 s, 20 s, 30 s, 60 s or 120 s).

Therefore, from a practical standpoint, if maximum accuracy is required (e.g., scientific approach) longer recording periods must be used, but shorter recording intervals might be a time-efficient option for clinicians or coaches working with big groups of athletes, or where logistical issues difficult long-lasting assessment protocols (e.g., athletes with pain during running).