All analyses were based on intention-to-treat, with the last value carried forward for missing values. The χ2 test, t test, or Wilcoxon rank sum test were used, as appropriate, to compare baseline characteristics. Changes from baseline to 6 months were summarized as means and standard deviations. Estimates of between-group mean differences, adjusted for baseline values, were computed, together with 95% confidence intervals (CIs). Analyses of covariance were performed to examine differences in changes across groups, with baseline value as covariate. Because gait velocity may represent a potential confounder when evaluating gait variability,38 additional analyses with adjustments for changes in gait velocity were performed for all variability measures. Data were also analyzed at a normalized velocity of 100 cm/s (ie, within the individual range of velocities and close to the mean usual gait velocity) using an interpolation procedure, pooling the 3 walking conditions under a single task (ie, self-selected slow, usual, and fast speeds), as described elsewhere.39,40 Six- to 12-month changes were summarized as means and standard deviations. Crossover analyses were performed using a specific Stata procedure (command pkcross) that enables analysis of crossover experiments and provides significance values for sequence, period, intervention, and carryover effects. The data were reanalyzed (1) without adjustment for baseline covariates, (2) using per-protocol analysis of study completers without imputation, and (3) using linear mixed-effects regression models (with Stata “xtmixed” command) to predict the primary end point, with visit, intervention, and visit by intervention interaction as exploratory variables.