Limitations of study
Factors that need to be considered in the interpretation of
our findings include insufficient power and potential biases.
The actual power to detect a clinically important change (i.e. four points or greater) in the SGRQ was low (13%). Our failure to detect statistically significant between-group differences in the SGRQ-domain impact and activity was thus likely limited by sample sizes, since the direction and magnitude of treatment effects consistently approached values that are considered to be clinically important. Moreover, because of the low occurrence of emergency department utilization and hospitalizations, the power to detect differences between the groups for these outcomes was even lower. Although this was
a randomized trial, a few baseline characteristics were not
equally distributed among the experimental groups. For
this reason, we used multivariate analysis to adjust for
these baseline differences. In addition, our study did not
report measures on determinants of behavioral change. As
such, more attention should be paid to the patterns of
physical activity with process indicators (i.e. knowledge,
psychosocial beliefs and self-efficacy) e since they
determine behavioral change, which in turn determines
clinical and quality of life outcomes. Moreover, these
indicators may be more sensitive in capturing relevant
changes specific to self-management, and reduce the risk
of false-negative results.29 Finally, we cannot exclude the
presence of an attrition bias since seven patients allocated
in each group were not available for the evaluation at 1
year. However, this potential bias is minimized given their
comparable baseline characteristics with patients who
completed the follow-up. Moreover, the frequency and the
causes of dropping out were comparable between the
intervention groups.