My question refers to a situation where the number of participants in some strata is unexpectedly low. Should randomization proceed in the same way as declared in the Protocol, i.e. by performing randomizing independently in each strata, or should strata be combined to prevent blindness? Let's call the two alternative treatments A and B, and let's consider the extreme case where only 2 people are randomized. In that case, it would be obvious that the only possible options out of 4 would be:
- first patient treated with A, second patient treated with B;
- first patient treated with B, second patient treated with A.
This would create two problems:
A) Assessors suspecting to be able to guess the treatment status of one participants would automatically suspect the treatment status of the other one as well;
B) There would be a complete dependence between the treatment status of the first and of the second participant.
At the same time, combining strata in order to determine the randomization set could be seen as a violation of the randomization protocol.
Thus, how should one behave in such situation?
