Principal components analysis of postural sway in persons with unilateral lower limb amputation: A wearable sensor approach

Abstract

Standing sway assessments can detect sensory imbalances which compromise postural control. Persons with lower limb amputation (LLA) often demonstrate impaired postural control, increasing fall risk. Here, principal features of postural sway were identified in persons with unilateral LLA using a single, commercially available wearable sensor. Sixty-one persons with LLA (n = 44 transtibial; n = 17 transfemoral) stood on a firm surface with eyes open/closed while wearing a single accelerometer mounted over the sacrum. Common parameters quantified spatiotemporal and spectral features of sway in anterior-posterior (AP) and mediolateral (ML) directions. Principal component (PC) dimensionality reduction was applied and loadings inspected to identify a reduced, non-redundant set among 14 original variables capturing 90 % variance. Six PCs described ≥ 90 % variance, with the first 3 explaining 75 %. With eyes open and closed, PC1 was loaded by variables characterizing trajectory planar size: area, jerk (i.e., sway smoothness), AP/ML RMS path distance, and AP/ML path range. With eyes open, PC2 was loaded by variables characterizing direction and spectral features: ellipse rotation, AP centroidal frequency, and ML jerk. With eyes closed, PC2 spectral loadings increased: ML centroidal frequency, ML frequency dispersion, and AP centroidal frequency. With eyes open, PC3 was loaded by ellipse rotation, jerk, ML velocity, ML centroidal frequency. With eyes closed, PC3 was loaded by ellipse rotation, ML centroidal frequency, ML frequency dispersion, and AP path velocity, characterizing off-axis error/corrections. RMS of path distance, ellipse rotation, centroidal frequency, frequency dispersion, path velocity, and jerk are a concise parameter set, derived from an accelerometer, to capture principal sway features in persons with LLA during standing balance with visual perturbations.