Astea
★

Russia,
2019-06-06 01:24

Posting: # 20319
Views: 1,434

## Two-Stage Design for FDC [Two-Stage / GS Designs]

Dear Smart People!

I'm wondering what is an appropriate way to use Two-Stage Designs for FDC (two analytes, A and B, for example)? It seems we are facing several questions:
1). Initial sample size calculation - best guess of CV for two CV?
2). Initial sample size calculation - power (see this thread). So if we would not expect independent hypothesis we should use the adjusted power for calculation. Correspondingly this will lead to neccesity of using additional simulations cause we will be automatically driven from validated values of 80 and 90%.
3). Interim analyses: for 2 analytes it leads to different possibilities: pass, fail, pass for A but need the second stage for B...
4). Sample size for the next stage:
- additional subjects could cause TIE inflation (as in the example with extra drop-outs, see this thread
- is it regulatory addopted - ignoring data for the second analyte?

"Being in minority, even a minority of one, did not make you mad"
mittyri
★★

Russia,
2019-06-08 15:44

@ Astea
Posting: # 20320
Views: 1,120

## No simple way out

Dear Astea,

as far as I can see you made a lot of forum research already

I don't see any other method except direct simulations, but as Detlew mentioned in the link you provided: too many what if's!!!
Remember that you need to take into account not only 2 analytes, but 2 PK metrics for both of them.

» 1). Initial sample size calculation - best guess of CV for two CV?

see above - not for 2 but for 4!
Guestimation is our friend. For that particular protocol I think you can prove almost any reliable numbers. (n1 is low: well, that's a 2 Stage design, I know nothing about CV!
n1 is high: I think CV for Cmax of that analyte should go to the sky!)

» 2). Initial sample size calculation - power (see this thread: ). So if we would not expect independent hypothesis we should use the adjusted power for calculation. Correspondingly this will lead to neccesity of using additional simulations cause we will be automatically driven from validated values of 80 and 90%.

If you don't know CVs how would you estimate rho?
May you want to build a correlation matrix for all of 4 pk metrics?

» 3). Interim analyses: for 2 analytes it leads to different possibilities: pass, fail, pass for A but need the second stage for B...

Yes, end of story (again, 4 metrics!). The framework becomes absolutely crazy. So I don't see any option except independent PK metrics analysis as you did for simple analytes in 2 stage designs. Forced BE? Yes, we're gonna live with that for now...

» 4). Sample size for the next stage:
» - additional subjects could cause TIE inflation (as in the example with extra drop-outs, see this thread.

From the link provided I don't see TIE inflation (using Potvin B and Detlew's function)

» - is it regulatory addopted - ignoring data for the second analyte?

No, as Helmut mentioned in the same link. I don't think experts be happy trying to dive into so complicated framework.

Kind regards,
Mittyri
Astea
★

Russia,
2019-06-09 22:56

@ mittyri
Posting: # 20321
Views: 1,088

## the more complicated the more interesting

Dear mittyri!

As for ρ (that is two PK metrics of the same analyte) I think that it is worth to expect correlation while for two different analytes correlation of the same PK metrics is not very likely.

» From the link provided I don't see TIE inflation (using Potvin B and Detlew's function)

Sorry, I've assumed the linked with this theme thread. Of course the theme is too complicated, but any possible affection on the TIE should be investigated, shouldn't it?

"Being in minority, even a minority of one, did not make you mad"