Revisiting sources of variability in gait analysis

Abstract

Background

Gait analyses in clinical populations must be considered differently, as variation in measurements may be related to the clinical condition and not just factors of interest. However, measurements taken from gait also have natural variability and this variability is further compounded when multiple factors may be of clinical interest.

Research question

Do current methods properly assign and quantify the amount of variability in gait data?

Methods

Simulated data were utilized to identify subject and therapist effects using multiple gait trials; data were simulated with and without multiple sessions with therapists. Five different statistical designs were considered that allow within-subject, within-therapist, and between-therapist errors. These are (1) a series of nested models, (2) a single model with interaction effects and nested structure, (3) cross-sectional ANOVA with fixed effects, (4) cross-sectional ANOVA with random effects, and (5) nested ANOVA. All modeling considered different therapists, trials, and subjects, and considered models were identified from gait literature. Ratios between estimated variances and the overall statistical errors were calculated; ratios were averaged and considered correctly identified when the estimated variance or variance component was greater than the random errors.

Results

The series of nested models identified therapist and session effects for all simulated outcomes but failed to account for subject and interaction effects. Estimates from the single model with interaction effects and nested structure exhibited a broader range of averaged ratios. The cross-sectional ANOVA with fixed effects accurately identified the sources of variability and can better quantify the source of variation, compared to all other considered models.

Significance

Accurately identifying and assigning sources of variability is imperative to accurately interpret gait which may influence or change clinical interpretation or understanding. The appropriate statistical design allows one to partition variation to accomplish this purpose.