Objectives: Develop algorithms to detect gait impairments remotely using data from freely worn devices during long-term monitoring. Identify statistical models that describe how gait performances are distributed over several weeks. Determine the data window required to reliably assess an increased propensity for falling.
Methods: 1085 days of walking data were collected from eighteen independent-living older people (mean age 83 years) using a freely worn pendant sensor (housing a triaxial accelerometer and pressure sensor). Statistical distributions from several accelerometer-derived gait features (encompassing quantity, exposure, intensity, and quality) were compared for those with and without a history of falling.
Results: Participants completed more short walks relative to long walks, as approximated by a power law. Walks less than 13.1 s comprised 50% of exposure to walking-related falls. Daily-life cadence was bimodal and step-time variability followed a log-normal distribution. Fallers took significantly fewer steps per walk and had relatively more exposure from short walks and greater mode of step-time variability.
Conclusions: Using a freely worn device and wavelet-based analysis tools allowed long-term monitoring of walks greater than or equal to three steps. In older people, short walks constitute a large proportion of exposure to falls. To identify fallers, mode of variability may be a better measure of central tendency than mean of variability. A week's monitoring is sufficient to reliably assess the long-term propensity for falling.
Significance: Statistical distributions of gait performances provide a reference for future wearable device development and research into the complex relationships between daily-life walking patterns, morbidity, and falls.