Bioequivalence and Bioavailability Forum

Hi Peregrin,

❝ Does this imply that the study should be conducted simultaneously in the whole study population?

Not at all. It means that you shouldn’t replace dropouts. You estimate the sample size (based one the expected T/R-ratio & CV, desired power) and take the expected dropout-rate into account. See this post. If the actual dropout-rate is lower than expected, fine. You get a reward (lower producer’s risk). If it is higher, you will loose some power. Example for 2×2 crossover, CV 25%, T/R 0.95, desired power 80%, expected dropout-rate 10%: You estimate the desired sample size with 28 (power 80.7%), and dose 32 (uprounded):

nel  doact (%)  power (%)
 32     0.0      85.7
 31     3.1      84.6
 30     6.3      83.4
 29     9.4      82.1
 28    12.5      80.7
 27    15.6      79.2
 26    18.8      77.6

n_el are the eligible subjects and do_act the actual dropout-rates.

The GL is talking about this idea: You desire 28, dose 32, have just 2 dropouts (say #27/28), and analyze #1–#26/#29/#30 (you don’t analyze #31/#32). Though that would safe some money in bioanalytics, there were regulatory concerns about this approach in the past:

• It is not ethical to potentially have data which will not be used.
  
• Doubts about the transparency of the procedure. The easiest rule would be “analyze the next ID”. Some people thought to be very clever and fiddled around with a “ran­dom” pick. Now doubts may rise: Is it possible that all ”spare” subjects were analyzed (undocumented, of course) and the “best” are picked? Yes, that’s plain fraud, but…

It is perfectly fine to perform a study in multiple groups (say for logistic reasons). Whether or not you have to modify the statistical model is a gray zone. Search the forum for “group effect”.
See also FDA’s 2001 guidance, Section VII.A.