from which the variance could be explained by sampling error
alone.14,16 Although the Q-statistic underestimates the existing heterogeneity
in meta-analysis, the percentage of total variation across studies
was used by calculating I2
, which reveals a better measure of the
consistency between trials.17 When significant heterogeneity was found
on the Q-statistic (or I2 values "50%),17 a random effects model was
applied.18,19 Pearson correlation coefficients were calculated between
individual effect sizes, and additional time exercising in the experimental
groups, as well as the methodological quality of the studies. To
investigate the effects of differences in treatment contrast in selected
studies, a cumulative meta-analysis adjusted for treatment contrast was
applied for the ADL outcome, using a random effects model. Finally, a
sensitivity analysis was performed for studies with high and low
treatment contrasts. Effect sizes were calculated in Excel for Windows
XP, whereas SPSS 11.5 for Windows was used for statistical analysis.
For all outcome variables, the critical value for rejecting H0 (ie, there is
no evidence for augmented therapy time) was set at 0.05.