Subgroups and multiple outcomes:
key points
i)
Subgroup analyses
The analyses
described so far have compared all patients in one group with all patients in
another group. However, there may be
clinical relevance to comparisons which addressed the question of whether or
not the treatment effect differed between different sub-groups, e.g. between
males and females.
ii)
Methods for comparing and selecting
subgroups
It is
important that differences between subgroups should be assessed using the
correct test of interaction, not simply by comparing P-values. Also, if the subgroups to be compared are not
selected at the outset of the trial, care needs to be taken to ensure that you
can defend the method of selecting subgroups against accusations that you are
simply fishing for any significant difference you can lay your hands on.
iii)
Multiple outcomes
It is usual
in clinical trials to measure several outcomes.
How these are assessed needs care, especially from the point of view of
hypothesis tests. If you perform many
hypothesis tests, then a few will be significant by chance, even if the null
hypothesis is true.