I am currently considering the design for a Phase III clinical trial.
The primary end-point will be binary, and I propose to use a
Cochran-Mantel-Haenszel analysis, adjusting for strata defined using one
or a few baseline variables that are thought to be prognostic of the
risk of the event.
As I see it, the value of a stratified primary analysis is:
(a) to adjust in case there is by chance an imbalance between the active
and control treatments in their baseline characteristics.
(b) to improve power and precision in estimating the size of the effect
Some very limited simulations have suggested that stratification has
only a very modest effect on power and precision, so stratification is
principally to ensure the primary analysis is as convincing as
possible. An unstratified analysis will also be performed as supporting
data.
My question is this - how finely to stratify, ie how many strata to
defne, what considerations should apply to choosing the stratification
variables, or the cut-points for many-levelled variables. I am sure
others have considered this question before, but I do not recall having
seen a discussion of these particular issues, so some pointers to the
best literature would be appreciated.
I anticipate that in this study, a simple randomisation will be used.
Would it make a difference if randomisation was stratified by some of
the possible stratification variables.
Responses to me please and I will summarize to the list.
Thanks for your interest
Tim Auton
--
The views, opinions and judgements expressed in this message are solely
those of the author. The message contents have not been reviewed or
approved by Protherics.
T R Auton PhD MSc C.Math
Head of Biomedical Statistics
Protherics Molecular Design Ltd
The Heath Business and Technical Park
Runcorn
Cheshire
WA7 4QF
UK
email: [log in to unmask]
|