25 replies to this topic

[shikha rawat]

dear all forum members ,

 

have any one of you ever did analysis of "2treatment x 2 sequence x 4 period " (TRTR, RTRT )study on winnonlin . Firstly i did a aBE using the default settings given in the user manual for replicated study designs. i got the value of Var(per*form*sub)_21 = 0.058986 (which is the variance of the log transformed PK parameter for the reference product as per my knowledge) which further gives me the CV(%) as 24.6496% . The same data was analysed using PBE/IBE where i got Sigma WR(standard deviation within reference )= 0.2684 which gives me CV%= 27.3307%. although both the CV's are <30. , still i want to know which of these is the correct estimate .

thanks in advance

 

regards

stats06

[Simon Davis]

Dear Stats06,

   What do you want to use this IntraSubject cv% for? As that may have an impact on which one is most appropriate for you to use.

 

  Personally I think you should just use the AVERAGE BE as the IBE is obsolete now in my opinion, but there are better statisticians than me onthis board. You may find these related topics of interest.

 

http://pharsight.com/extranet/index.php?option=com_kunena&Itemid=55&func=view&catid=39&id=858#864

 

and on the BEBAC forum.

 

http://forum.bebac.at/mix_entry.php?id=7871#p7931

 

 Simon.

[shikha rawat]

Hi Simon ,

 

Thanks for your reply .. the links you mentioned were quite helful. My aim was just to chek the intrasubject CV(%) for the innovator data to confirm wether we need to apply SABE or not. i read the progestrone guidelines and calcuated the sigmaWR with my pocket calculator :-) which was equivalent to what i get from the IBE/PBE settings. The CV is below 30% so abe is the right approach mentioned by you as well.

 

thanks & regards

stats06

[Simon Davis]

SO are you looking to perform RSABE within a Phoenix workflow ?

 

We have this down as a 'to do' from the UI but you can, with a bit of work, construct the workflow template in Phoenix now. I think there is an example posted here or on the BEBAC forum, be sure to use 6.3 as it will give you a much neater workflow with the Data Wizard tool etc.

 

 Simon.

[Helmut Schütz]

Dear stats06!

Although you posted in the classical WNL section as Simon already suggested consider upgrading to PHX6.3. Settjng up the workflow according to FDA's code in WNL would be a nightmare.

Since I don't have your data in the following a comparison of CVwr obtained from different models (Examples folder, Data 2x4, reference=capsule, test=tablet):
PBE/IBE: SigmaWR 0.49050 (CVwr 52.15%). Note that SigmaWR is not the variance, but the standard deviation, therefore CVwr = sqrt(exp((SigmaWR)^2)-1)
unscaled ABE (to be honest, I never found out which variance represents s²WR…):
lambda(1,1)_11: 0.19910 (CVwr 46.94%)
Var(Period*Formulation*Subject)_21: 0.16726 (CVwr 42.67%)
FDA's RSABE code: s²WR 0.24059 (CVwr 52.15%)

You see, it's not that easy. ;-)

If you want to go for scaling, you have to follow FDA's code anyhow. For a quick look (without setting up the entire workflow) use SigmaWR from PBE/IBE and calculate CVwr as indicated above.

[shikha rawat]

Hi helmut ,

thanks for your suggestions . i had calculated the CV(%) from IBE/PBE and just to confirm that i was going in the right direction i calculated it manually as well. (Now i can go for the PBE/IBE approach in my next study just to have an estimate of CV% within reference but we still dont have within test variance so have to calculate that manually again ;-) )

I have phoenix 6.1 right now, will surely get the upgaraded version of 6.3 (i hope its free since we already have a license ).

##unscaled ABE (to be honest, I never found out which variance represents s²WR…):
lambda(1,1)_11: 0.19910 (CVwr 46.94%)##

Below is the information given in the parameter key in ABE
Ln(Cmax) ng/mL lambda(1,1)_11 Random No-Diag Factor Analytic 2 SUB
Ln(Cmax) ng/mL lambda(1,2)_11 Random No-Diag Factor Analytic 2 SUB
Ln(Cmax) ng/mL lambda(2,2)_11 Random No-Diag Factor Analytic 2 SUB
Ln(Cmax) ng/mL Var(PER*FORM*SUB)_21 Repeated Variance Components SUB FORM FORM_R
Ln(Cmax) ng/mL Var(PER*FORM*SUB)_22 Repeated Variance Components SUB FORM FORM_T
>>>>>>>>>>>>> the Results in the final variance sheet

Ln(Cmax) ng/mL lambda(1,1)_11 0.481713
Ln(Cmax) ng/mL lambda(1,2)_11 0.409761
Ln(Cmax) ng/mL lambda(2,2)_11 0.000000
Ln(Cmax) ng/mL Var(PER*FORM*SUB)_21 0.058986
Ln(Cmax) ng/mL Var(PER*FORM*SUB)_22 0.055046

Now as we can see that lambda(1,1)_lambda(1,2) lambda(2,2)_ here are the non diagonal analytic factors. so they can not be the varainces ...(i am not sure but as per my understanding this may be the cov's).

Can u guide me in setting the workflow for RSABE in phoenix as I am not able to understand this approach :-(
i saw the SAS codes given at BA/BE forum but still was not able to find the logic of this approach and the calculation of upper confidence bound.

thanks & regards
stats06

[Helmut Schütz]

Hi stats06!

I have phoenix 6.1 right now, will surely get the upgaraded version of 6.3 (i hope its free since we already have a license ).

Yes it is. Get it - makes life so much easier. ;-)

Can u guide me in setting the workflow for RSABE in phoenix as I am not able to understand this approach :-(
i saw the SAS codes given at BA/BE forum but still was not able to find the logic of this approach and the calculation of upper confidence bound.

I try to attach a project (you will need to upgrade to PHX6.3 in order to work with it) with my implementation of FDA's code (full and partial replicate, both RSABE and ABE). Note that the chi²-distribution is not available in PHX. I have included a table with critical values for 1-200 degrees of freedom (see the Information tab for the R-code I used).

Since I don't have SAS myself, everybody using my code does so on his/her own risk; I don't take any responsibility about its correctness!

I rely on the cross-validation we have performed in the BABE-Forum.

[file name=FDA_RSABE_PHX63.phxproj size=3852050]http://www.pharsight.com/extranet/media/kunena/attachments/legacy/files/FDA_RSABE_PHX63.phxproj[/file]

P.S. @mods: Can you move this thread to the PHX/WNL category?

[Simon Davis]

6.3 is the current version and can be downloaded freely by all currentl licensees from;

 

ftp://support_ftp:pt9JTmD9GdXQLtCgPkvL@ftp.certara.com/support/desktop/Phoenix/Phoenix_1.3.zip

 

I would suggest you speak with your internal IT contact to make sure he passes on information from Pharsight about these updates.

 

 Simon.

[Helmut Schütz]

Hi stats06,

 

I prepared a description of the workflow:

http://bebac.at/downloads/Replicate_Designs_for_SABE_according_to_FDA_with_Phoenix_v1.pdf

 

At the end there is a link to an updated project as well. Hope you find it useful.

[shikha rawat]

hi helmut & simon,

thanks a lot for your help :-)
Now i have the RSABE results using SAS as well ( same study was sent to a CRO ) so we can easily validate the results once i do it using Phoenix.the description of the workflow is really g8. it will be a great help for those who are doing SABE for the first time (just like me ;-) )


i try to attach a project (you will need to upgrade to PHX6.3 in order to work with it) with my implementation of FDA's code (full and partial replicate, both RSABE and ABE). Note that the chi²-distribution is not available in PHX. I have included a table with critical values for 1-200 degrees of freedom (see the Information tab for the R-code I used).

sorry ,but m not able to extract the files from it(tried using winzip but m asked for a password time nd again )
As soon as i upgrade phoenix 6.1 to 6.3 ,ill complete my project and post my results :-)

@simon I have asked the IT people to get the upgarded version of phoenix & keep me updated with all the future developements as well .Also thanks for the link

thanks & regards
stats06

[Simon Davis]

Stats06,
Which link are you having problems downloading? Which browser are you using?

PHX 6.3 s/w from the Pharsight FTP site?

Helmut's PDF from BEBAC site

the PHXPROJ referenced at the end of the PDF?

If the last I too had an issue in Firefox (13.0.1)

but it downloaded fine with IE. I have heard that some times the .PHXPROJ file extension is replaced with .ZIP when downloading via IE. If you have that problem, simply rename the file extension to ****.PHXPROJ and then open with Phoenix normally.

Simon.

[shikha rawat]

hi simon,

 

its the file attachment by helmut

File Name: FDA_RSABE_PHX63.phxproj

i am not able to extract the files , the message when i ltry to do so is " enter the password for ecrypted file :FDA RSABE.winnonlin" :-0

 

the rest two links are fine :-)

 

 

thanks

stats06

[shikha rawat]

hi simon,

 

u were right . i am able to load that project successfuly now.. thanks :-)

 

stats06

[Helmut Schütz]

Hi stats06 & Simon,

Oh, I see. Altered the .htaccess at my server. Now the file should be ready for download in any browser.

Since this happens here also sometimes, ask the IT people to add following line to .htaccess:

AddType application/octet-stream .phxproj

[shikha rawat]

hi helmut & simon,

Had a really hectic week but finally I have completed my analysis of the RSABE in phoenix ..... the results are given below,

critbound -0.04467813 CV(%) 33.16

Results using SAS

Critbound -0.04320959 CV(%) 33.20

although the results differ slightly but both show RSABE :-)

thanks a lot for your help and the document is extremely helpful in understanding the logic of RSABE .
also thanks for making me aware of the PHOENIX 6.3 it really makes our work easy and the workflow simplifies the process to some extent .
hopefully we will soon get the UI for the same

regards
stats06

[Helmut Schütz]

Hi stats06!

phoenix ..... the results are given below,
critbound -0.04467813 CV(%) 33.16

Results using SAS
Critbound -0.04320959 CV(%) 33.20

although the results differ slightly but both show RSABE :-)

That's a much larger difference that I would accept. Are you sure you haven’t made a minor mistake in setting up the workflow – or alternatively – in SAS? Please check. Consider posting your project here.

[shikha rawat]

dear helmut !

i did recheck the workflow in phoenix but dint find anything wrong.! The SAS results were given by some CRO (i dont have SAS here ) so i have asked them as well to recheck their results. i am waiting for their reply , in case i dont get a solution i'll post my project here

regards
stats06

[shikha rawat]

hi helmut ,

 

The results are still the same . i am not able to figure out the reason so m posting my project here .

 

thanks

stats06 [file name=fast_data.phxproj size=1345604]http://pharsight.com/extranet/media/kunena/attachments/legacy/files/fast_data.phxproj[/file]

[Helmut Schütz]

Hi stats06,

I checked your project and it seems to be OK. critbound is -0.040467813, whereas in a previous post you stated -0.04467813 for PHX; I guess this was just a typo in the 3 decimal place. Not clear why the result differs from SAS' -0.04320959.

Since I don't have SAS as well, can anybody at Pharsight please check that?

[shikha rawat]

hi helmut,

 

ooops ..u are right, that was a typo error ...

thanks for confirming my results.. u made my day :-)

Regarding the SAS results the problem is still open.

 

 

Regards

stats06