These
analyses, although prespecified, are
considered exploratory because the study
is not powered for such analyses, and
false results due to multiple comparisons
must be considered when interpreting
findings that are nominally statistically
significant. Furthermore, on-treatment
analyses lose the protection from confounding
by unmeasured variables that
is afforded by randomization.