2 replies to this topic

[Veronika Trushko]

Dear all,

 

I am a new user of Phoenix WNL and it is my first project so I look forward to your understanding.
The project represents bioequivalence study with 2х2 crossover design. The study protocol says that the aim is to determine 90%CI (with alpha=0.05 or p=0.95 and method power(1-beta)=0.80).

I need to computate a power for observed data to prove that it is bigger then 0.80. If I right that the power from average BE output I should use is Power_80_20? And according to the terms of protocol, does the confidence level I should put in the options tab equal to 90%?

 

 

Thanks,

Veronika.

Edited by Veronika Trushko, 25 September 2017 - 09:46 AM.

[Helmut Schütz]

Hi Veronika,

 

The study protocol says that the aim is to determine 90%CI (with alpha=0.05…

 

So far so good.

 

… and method power(1-beta)=0.80).

 

Sorry to say but that’s a stupid protocol. Post hoc power is meaningless and its computation futile. Either a BE study has demonstrated BE (the 90% CI lies entirely within the acceptance range) or not.

Power is only relevant in study planning. A sponsor is free to choose a desired Type II Error (β, the producer’s risk). The sample size depends on assumptions (about the CV and GMR) – which might be false…

 

I need to computate a power for observed data to prove that it is bigger then 0.80.

 

OK, but only to comply with the protocol. Avoid it in future ones. What will you conclude if “power” is less than 80% although the study showed that the products was BE? Regulators are should be only interested in the probability of a the Type I Error (α, the patient’s risk) which is ≤0.05 independent from power.

Since the sample size was based on assumptions (see above) nobody can be punished if the assumptions turned out to be wrong. If the CV was higher, the GMR more deviating from unity, or the dropout-rate higher than expected, power will be lower than desired. But this risk is on the producer’s side. In other words, the chances to show BE were lower, but you still succeeded. Open a bottle of champagne.

 

See this presentation.

 

If I right that the power from average BE output I should use is Power_80_20?

 

No. This is the power to show a 20% significant difference according to the FDA’s “80/20-Rule”. This method was abandoned in 1987 and is kept in the output of PHX/WNL just for backwards compatibility. It is of historic interest only.

I hope that it will removed in a future release of PHX/WNL. Causes a lot of confusion. It is possible to have a failed study with power for the 80/20-Rule close to 1. That’s bizarre.

 

If you want to calculate post hoc power, you have to take the observed intra-subject CV and GMR into account.

See Output Data > Average Bioequivalence > Power_TOST

and in the Core Output the line starting with Power_TOST

 

And according to the terms of protocol, does the confidence level I should put in the options tab equal to 90%?

 

Yes. That’s the default in PHX/WNL.

[Veronika Trushko]

Dear Helmut, thank you so much for your reply! It is very helpful for me!