Measurement error associated with gait cycle selection in treadmill running at various speeds

**Background:** Biomechanical studies of running commonly average data from several gait cycles to compute a 'representative mean' of an individual's running technique. The number of gait cycles used varies widely across the literature, and the impact of gait cycle selection on kinematic outcome measures has not been systematically investigated. Previous work by Oliveira & Pirscoveanu (2021) found that >10 gait cycles are typically needed for stable biomechanical measures in overground running, but did not examine lower limb kinematics or treadmill running specifically.

**Methods:** The authors used the public dataset from Fukuchi, Fukuchi & Duarte (2017), which contains lower-extremity kinematics and kinetics of 28 recreational runners (27 male, 1 female; age = 34.8 ± 6.7 years; height = 1.76 ± 0.68 m; mass = 69.6 ± 7.7 kg; running experience = 8.5 ± 7.0 years; running pace = 4.1 ± 0.4 min/km). Participants ran on an instrumented treadmill at three speeds (2.5 m/s, 3.5 m/s, and 4.5 m/s) with a 3-minute accommodation period followed by a 30-second data collection period. Kinematics were processed using OpenSim 4.0 with a generic musculoskeletal model. Five kinematic variables were extracted: hip flexion/extension, hip adduction/abduction, hip internal/external rotation, knee flexion, and ankle plantarflexion/dorsiflexion. For aim 1, the authors calculated 'ground truth' values from all available gait cycles, then iteratively computed means using 5–30 consecutive gait cycles (1,000 random samples per condition). For aim 2, they compared means from two non-overlapping subsets of 5–15 consecutive gait cycles from different parts of the capture period. Error was quantified as absolute difference for 0D (peak) variables and peak absolute difference across the normalised gait cycle for 1D (waveform) variables.

**Key Results:** For 0D kinematic variables, mean error, variance, and range progressively reduced as the number of gait cycles increased. At 2.5 m/s and 3.5 m/s, maximum errors were less than 1° even with small numbers of gait cycles. At 4.5 m/s, maximum errors typically exceeded 1–2° for peak hip and knee joint angles when fewer gait cycles were used, requiring 25–30 cycles to achieve similar error magnitudes to slower speeds. A bimodal distribution was observed at 4.5 m/s, where some sampling iterations produced relatively higher versus lower errors. Peak ankle dorsiflexion showed consistently low errors (<0.5°) across all speeds. For 1D kinematic variables, near-identical patterns were observed. When sampling from different parts of the capture period, variation remained relatively consistent irrespective of the number of gait cycles used. At 2.5 m/s and 3.5 m/s, variation was always less than 1.5°. At 4.5 m/s, variation increased to 2–4° across most kinematic variables. The reduction in potential error plateaued above 15–20 gait cycles, indicating diminishing returns.

**Clinical Implications:** The findings suggest that using 5–10 gait cycles at slower running speeds (2.5–3.5 m/s) introduces acceptably small error (<1°) for most lower limb kinematic variables. At faster speeds (4.5 m/s), more cycles (25–30) are needed to achieve comparable precision. The error from gait cycle selection is small compared to other established sources of error in biomechanical analysis (e.g., soft-tissue artefact, different joint coordinate systems, marker placement variability). Researchers should consider the magnitude of potential error against the size of effects they aim to detect—using fewer than 10 cycles to explore effects <1–2° in hip or knee flexion would be unwise. The study supports using as many gait cycles as practical, but acknowledges diminishing returns beyond 15–20 cycles. These results apply specifically to consecutive gait cycles from a 30-second treadmill running bout and may not generalise to non-consecutive sampling, overground running, or other biomechanical outcome measures.