This is a general question about good practices in per protocol
analyses in a randomized clinical trial.
ICH recommends a series of analyses for describing baseline data and
patients demographic characteristics.
I'm wondering whether we need to re-do these analyses on the Per
Protocol Set of patients, and further, re-compare general patients
characteristics on this subset by tests to check that trt groups still
are homogeneous on that sub-population.
The concern that I have is that the benefit of the randomization
(which yields groups where general patients characteristics are
randomly balanced across trt groups and ensures that the observed
effect size can, at the end of the day, be attributed to treatment
only) is lost when the analysis is conducted on a subset of the ITT
To make it simple, that would mean that, if the primary analysis of a
RCT on the ITT population is a t-test, we would need, switching to the
Per Protocol Set, to replace the t-test by a multiple linear model,
controlling for baseline trt group differences.
What do you think ?
Many thanks in advance.
||4/6/2010 7:07:36 PM