Minimisation And Stratification By Centre

Dear all,

 

I wonder if I could trouble any of you who could spare a few minutes to comment on the following issue.

 

We have a new 3 arm surgery trial in the planning stages.  As it is a surgery trial and so outcomes could be dependent on the surgeon and standard practice at the particular centre, we have been asked to include centre as a stratification variable along with 4 other variables.  We have always avoided stratifying by centre in the past due to the increased risks of doctors being able to guess the next allocated treatment.  Our statisticians have asked us to include a random element in with the usual minimisation algorithm, such that the first X patients will get a purely random allocation, and subsequently there is a 1 in 5 chance of being randomly allocated, and 4 in 5 of being allocated using the minimisation algorithm.

 

The following problems with this occur to me:

·         Centre come on line slowly, usually around 2 a month.  Also some centres often lead the pack in recruitment by a long way, meaning that the first X random patients will  be used up a long time before even half the centres come on line.  This means that if a doctor kept a list of allocations at a site, they could have a reasonable chance of success in guessing upcoming allocations, although the fact that this trial is 3 arm and not 2 arm does make the job of subverting it tougher.

·         If we changed the ‘first X patients in the trial given random treatment’ to ‘first Y patients at each centre given random treatment’ then we also could have problems:  Many centres do not recruit well, and many may only recruit fewer than ‘Y’ patients, meaning that they may never be remotely balanced in allocations by the trials end.  Also some that do recruit well, by an accident of timing of having 1 in 5 patients in the trial randomly allocated could end up always getting random allocations themselves and also not balancing their treatments.  We could perhaps change the 1 in 5 rule to apply on a per centre basis rather than total patients entered into the trial.

 

Do any of you run trials that are stratified by centre?  If so would you be able to share the principles (if not the details) about how to maintain allocation concealment in them, and still achieve an acceptable balance in the treatment arms?

 

Many thanks in anticipation!

 

Nick. Hilken

[Author]

Posts