Statistical analysis

Descriptive statistics of the outcome variables will be reported by using means and standard deviations or median and interquartile ranges, depending on their data distribution. Normality will be checked with the Shapiro-Wilk test and histograms will be checked for symmetry. Mixed models will be used to study changes after the intervention over time. By using random effects, these models are able to correct for the dependency among repeated observations. Furthermore, these models deal with missing data offering valid inferences, assuming that missing observations are unrelated to unobserved outcomes [70]. Based on the data distribution, linear (parametric) or generalized (non-parametric) linear mixed models will be used. Changes over time will be tested between groups, by analysing treatment-time interactions. In case of such a significant treatment-time interaction, changes over time will be investigated separately in each group. Significant time trends will be further investigated with pairwise post hoc tests to compare time points. Additionally, the effect size will be calculated using the Cohen’s d formula (small, 0.2–0.5; medium, 0.5–0.8, and large > 0.8) [71]. Both clinical (age, baseline AHA score, sensory function, mirror movements) and neurological predictors (brain lesion characteristics, structural and functional connectivity and CST wiring) will be included as covariates in the models for the primary outcome measure, together with their interaction with time and treatment to evaluate their potential confounding factor. The two-sided 5% level of significance will be used. All statistical analyses will be performed using SAS version 9.2 (SAS Institute, Inc., Cary, NC) and SPSS Statistics for Windows version 24.0 (IBM Corp. Armonk, NY: IBM Corp.).