Quantifying Variation in Gait Features from Wearable Inertial Sensors Using Mixed Effects Models

Area of Science:

• Biomechanics
• Wearable technology
• Statistical modeling

Background:

• Wearable inertial sensors enable continuous, longitudinal gait data collection outside laboratory settings.
• Free-living gait data is susceptible to confounding factors like gait speed and sensor mounting uncertainty.
• Existing methods struggle to control or estimate these variations, impacting data fidelity.

Purpose of the Study:

• To develop a statistical framework for analyzing wearable sensor-based gait data in free-living environments.
• To address challenges of data variation and confounding factors in wearable gait analysis.
• To facilitate more informative statistical inference for free-living gait.

Main Methods:

• Utilized statistical modeling to account for inherent uncertainties in wearable sensor data.
• Collected gait data from one healthy, non-elderly subject across 48 full-factorial trials.
• Employed random effects and fixed effects models to quantify variation impact.

Main Results:

• Identified four major sources of variation in wearable sensor-based gait data.
• Quantified the impact of these variations on the gait outcome 'range per cycle'.
• Demonstrated a statistical approach to model, rather than correct, data variations.

Conclusions:

• The developed methodology provides a foundational statistical framework for wearable gait data analysis.
• This approach facilitates robust statistical inference by accounting for real-world data variations.
• Enables more reliable interpretation of gait in free-living conditions using wearable sensors.