Bioequivalence and Bioavailability Forum

Ὀδυσσεύς

I the meantime I found a PDF.

❝ […] this add-on gives me some headaches. […] Does it have to be proposed in the initial protocol? Guess: yes, otherwise no way to preserve the consumer risk.

I think so. At least in 8.5.1.1 we read about sample size planning:

“[…] el error tipo I debe ser menor o igual al 5%.”

Correct. Even if we test at a nominal α 0.05, the actual risk might be <0.05. TOST is con­servative, especially for CVs >20–30% (and is to a minor degree dependent on the sample size) – see this plot. This conservatism explains the curves in the other thread. Below an updated version with identical TIE-scales of the approaches;
A unadjusted, B Bonferroni:


Dotted line: 0.05, dashed line: asymptotic maximum inflation for the chosen n₂-approach, solid line: maximum inflation.

• Lower inflation for higher CVs.
  
• Lower inflation with smaller total sample sizes.

In others words: Even if you don’t adjust, theoretically the type I error might not reach 5%. The Mexican requirement allowing to continue a study based on power <80% should add some conservatism. Practically this “works” only for HVDPs and sample sizes which’s power you wouldn’t like (i.e., forget this idea). As soon as you add enough subjects for a reasonable total sample size, you are history.
Example: You assume a CV of 30% and plan the study with 40 subjects. It turns out that the CV is 40%. “Power” is 55% (may pass by luck; let’s assume you fail). You perform the second stage with 26 subjects. If you don’t adjust, the patient’s risk will be ~7.7%.
OK; we have to comply with 8.5.1.1. It simply doesn’t work to keep 0.05 in the first stage. さようなら

❝ Which statistics are accepted to keep the consumers risk at 5%?

If you want to be on the safe side, follow Signore Bonferroni. Copy these equations to the protocol. The sample size penalty (to pass already in the first stage) is moderate (T/R 0.95, 80% power, α 0.05/0.025):
CV%   n0.05  n0.025
──────────────────────
15    12     16 (+33%)
20    20     24 (+20%)
25    28     36 (+29%)
30    40     50 (+25%)
40    66     82 (+24%)
50    98    124 (+27%)
Since for Mexico we need an intermediate power estimation (which could stop studies in the first stage), I guess it should be possible to use an α >0.025. Simulations required; talk to a com­petent statistician.
Note: Nothing is said both in the Japanese and Mexican guidance/regulation about the sample size estimation of the second stage. It seems that full adjustment (i.e., based on both the CV and T/R-ratio) is possible.

• Might screw up power if you use a futility criterion.
  
• Might lead to extreme sample sizes if you don’t.