Towards Establishing Ecological Validity of Robust Gait Variability Metrics for Walking in Real-World and Fall Risk Assessment

Introduction: Gait variability is a crucial indicator for dynamic stability and fall risk. However, inherent sensitivity to outliers can compromise measurement reliability of standard variability metrics like the Coefficient of Variation (CV), especially when applied to the "messy," non-normal data in free-living environments. The objective of the study is to validate a robust, non-parametric alternative — the Robust Coefficient of Variation using Median Absolute Deviation (RCV-MAD) — to ensure clinically reliable remote monitoring.

Methods: We analysed datasets collected in controlled lab settings (n=100) and real-world home environment 6-Minute Walk Tests from a cohort study (n=2,193). We compared the gold-standard standard deviation-based CV against RCV-MAD. We assessed the data distribution, test-retest reliability (intraclass correlation coefficient), temporal consistency, and clinical utility by comparing effect sizes in distinguishing older adults with lived fall experience versus those without.

Results: Real-world gait data proved inherently “heavy-tailed” (high kurtosis), violating statistical assumptions required for CV and rendering it sensitive to environmental artefacts. In contrast, the robust RCV-MAD is less affected by environmental noise, yielding significantly better temporal consistency and test-retest reliability in free-living conditions. Crucially, RCV-MAD of spatio-temporal gait parameters demonstrated a larger effect size in differentiating older adults with lived fall experience from those without compared to standard CV.

Conclusion and Impact: Conventional statistical approaches used in the lab may not be sufficient for unsupervised real-world monitoring. The RCV-MAD provides a more consistent and accurate measure of gait variability in daily living conditions. Adopting robust metrics like RCV-MAD is essential for minimising noise-induced false positives generated by wearables, enabling reliable longitudinal tracking and more accurate community-based fall risk assessment.