Bioequivalence and Bioavailability Forum

Hi ElMaestro & Naveen,

❝ 1. There are no publications about two-stage BE trials combined with replicated designs.

Correct.

❝ You might be asked to prove by your own simulations that type I errors are controlled at max 5% or so.

Correct again. But – unless one has access to a supercomputer or grid-computing – I would say that’s science fiction. The problem lies in the intermediate power calculation and sample size esti­ma­tion for the second stage. Contrary to 2,2,2 crossovers and parallel designs there is no explicit formula… Although the convergence is pretty fast, we still would need 10⁵ sim’s (the 10⁴ the two Lászlós used in their paper was too low). With Power2Stage (or your famous C-code) on a rea­son­ably fast machine 2,2,2 sim’s run for ~two days. If we add sim’s within the sim’s that would mean ~2,700 years. I don’t believe in immortality.
If you start tomorrow, ask a fortune teller for the adjusted α to validate, buy every year the latest hard­ware, and believe in Moore’s law then you’ll be done in twenty years.