Bioequivalence and Bioavailability Forum

Hi Oiinkie,

welcome to the party! This is one of the questions I hear most often after one of my presentations on two-stage designs. First: Nothing is published so far. Stop searching. The GL tells us:

[…] appropriate steps must be taken to preserve the overall type I error of the experiment and the stopping criteria should be clearly defined prior to the study.
[…] the choice of how much alpha to spend at the interim analysis is at the company’s discretion. The plan to use a two-stage approach must be pre-specified in the protocol along with the adjusted significance levels to be used for each of the analyses.

Essentially you would have to set up a framework similar to the published papers (Potvin et al. 2008, Montague et al. 2011, Fuglsang 2013; forget Karalis & Macheras 2013) and demonstrate by simulations (sorry!) that the overall α-level is maintained. Points to consider:

• In the 3-treatment part with the usual model you have a common variance – which might be larger/smaller then the one in the second part.
  
• An option would be to use EMA’s approach of “leaving one out”; you would get only the pooled variance of the remaining pair.
  
• You should introduce at least a statistical futility rule based on the ratio. If T₁/R and T₂/R are similar (whatever that means), consider dropping the one with higher variability.

Just some desultory thoughts.

P.S.: Expect a couple of weeks running the sim’s 24/7 on a single machine.