Online Program

Clustering Analyses Based on Gait Qualities of Community-Dwelling Older Adults (304174)

Jennifer Brach, School of Health and Rehabilitation Sciences, University of Pittsburgh 
Nemin Chen, School of Public Health, University of Pittsburgh 
Leslie Marie Coffman, School of Health and Rehabilitation Sciences, University of Pittsburgh 
Mark Redfern, Swanson School of Engineering, University of Pittsburgh 
Andrea Rosso, School of Public Health, University of Pittsburgh 
Ervin Sejdic, Swanson School of Engineering, University of Pittsburgh 
*Anisha Suri, Swanson School of Engineering, University of Pittsburgh 
Jessie VanSwearingen, School of Health and Rehabilitation Sciences, University of Pittsburgh 

Keywords: Gait accelerometry, gait analysis, older adults, Bayesian information criterion (BIC Score), Gaussian mixture model (GMM), unsupervised learning, life space assessment

Gait quality defined by spatiotemporal variability and acceleration may indicate risk for mobility problems and inform rehabilitation. We hypothesize that unsupervised clustering model of gait measures will distinguish subgroups of older adults by mobility.

In healthy participants (N=232, age 77.5±6.6, 65% females, Life Space Assessment (LSA) 75±19) during walking, we used tri-axial accelerometers to derive smoothness (harmonic ratio), similarity (cross correlations), regularity (entropy rate) and power (peak frequency). An instrumented walkway measured gait speed and step time variability (CoV). Clusters were defined using BIC and Gaussian mixed model (GMM); Gait quality and LSA (not originally in model) compared between groups using t-tests.

GMM indicated two clusters. Group 1 (N=189) has better gait quality (p<0.006) than Group 2: smoothness AP 2.9±.8 vs 2.3±.7; similarity AP-V .6±.1 vs .4±.2; regularity V .8±.1 vs .9±.1; power V 1.9±.2 vs 1±1.8; speed 1.1±.2 vs 1±.2 m/s; step time CoV 3.7±1.1 vs 5.1±2.4 and better LSA (76±18 vs 67±18).

Accelerometer measures cluster older adults by gait quality. The group with better gait has more LSA.