Statistical analyses
Given our interest in the impact of 2 training interventions
on different outcomes in patients with iSCI, statistical
analyses were applied to quantify differences in
change scores between the two interventions for each
outcome. The use of a cross-over design was chosen to
reduce the impact of inter-individual variability by having
each participant act as its own control [45]. To allow
an upfront interpretation, we adopted parametric testing,
as only a few secondary outcome measures did not show
normally distributed within-subject change scores (tested
with the Shapiro-Wilk Test), namely SCIM, WISCI,
UEMS and LEMS. This approach allows including the
data in future meta-analyses. Figure 3 displays an overview
of the applied statistical analyses, accounting for the
characteristics of cross-over designs. As carry-over and
treatment by period interaction are unlikely to be separable
[45], we did one analysis for both combined.
To evaluate the longitudinal influence of the interventions
on pain intensity, we plotted the mean VAS-scores
before and after training against time for all 16 sessions
for each intervention and performed linear regression analyses.
We considered a longitudinal decrease in pain intensity
to be significant, when the regression coefficient was
significantly smaller than zero. Additionally, we investigated
the short-term effect of training on pain intensity by
subtracting the mean VAS-scores after training from those
before training for each intervention and compared these
change scores with a Paired-Samples T Test.
For all outcome measures, intention-to-treat analysis
was performed using the last observation carried forward
method to account for missing data. Only 1 participant refused
to take part in the follow-up measurement, and this
was treated as missing data. Alpha was set at 0.05.