Process stationarity and reliability of trunk postural stability.

BACKGROUND: Empirical assessments of torso stability can be estimated from postural variability and nonlinear analyses of seated balance tasks. However, processing methods require sufficient signal duration and test-retest experiments require the assessment must be reliable. Our goal was to characterize the reliability and establish the trial duration for torso stability assessment.
METHODS: Kinetic and kinematic data were recorded while subjects maintained a seated posture on a wobbly seat pan. Stability was evaluated from dynamic variability and nonlinear stability analyses. Process stationarity of the measured signals characterized the minimum necessary trial duration. Intra-class correlations measured within-session and between-session reliability.
FINDINGS: Trial duration necessary to achieve process stationarity was 30.2 s. Shorter time to stationarity was observed with measures that included multi-dimensional movement behavior. Summary statistics of movement variability demonstrated moderate intra-session reliability, intra-class correlation=0.64 (range 0.38-0.87). Inter-session reliability for movement variance was moderate, intra-class correlation=0.42 (range 0.22-0.64). Nonlinear stability measures typically performed better than estimates of variability with inter-session reliability as high as intra-class correlation=0.83. Process stationarity and reliability were improved in more difficult balance conditions.
INTERPRETATION: To adequately capture torso dynamics during the stability assessment the trial duration should be at least 30 s. Moderate to excellent test-retest reliability can be achieved in intra-session analyses, but more repeated measurements are required for inter-session comparisons. Stability diffusion exponents, H(S), and the Lyapunov exponents provide excellent measures for intra-session analyses, while H(S) provides excellent inter-session comparisons of torso stability.