17 replies to this topic

[Naresh]

Dear All,

We are doing partial replicate study 3 period, 3 sequence, 2 treatment (reference repeated) for highly variable drug, can anybody share with me how to do partial replicate analysis in WinNonlin ver. 5.3., as we are interested to find intrasubject variability of reference product.

Used design
T R1 R2
R1 T R2
R1 R2 T

in this R1= R2 i.e. R1 and R2 are exactly same reference formulations.

Kindly suggest me how to do the analysis in WinNonlin5.3


Thanks in advance,

Naresh

(Note - moved to WinNonlin Classic general forum since this is not an M & S question, Simon.)

[pktastic]

Hi Naresh, a very similar question was asked on the BEBAC forum yesterday; perhaps you saw that already;

http://forum.bebac.a...p?id=6061#p6063

if not I'll post the essentials of Helmut's response;

"the coding within WinNonlin (or Phoenix) for any replicate design is the same. SigmaWR is given in the standard output; you have to calculate CV_WR = sqrt(e^sigma²WR-1) 'manually'. In WinNonlin you may write the formula directly in a cell, in Phoenix apply a user transformation."

I would add that you could equally use a Transform or Multi-transform in WNL 5.x to make it easier to review and refresh.

[Helmut Schütz]

Dear Naresh!

Used design
T R1 R2
R1 T R2
R1 R2 T

in this R1= R2 i.e. R1 and R2 are exactly same reference formulations.

I guess you have seen my reply at the BABE-Forum as suggested by Simon already. You should use a different coding (like in my linked presentation, slide 39). Your coding will not work - there's no replication of R! There's no distinction between 'R1' and 'R2' - simply use 'R' for both occasions of the reference treatment. Your coding should include 1-3 sequences, 1-3 periods, T or R treatments, 1-n subjects.

[Naresh]

Dear HS/Simon,

Thanks for your reply,

I have gone through the slide,

If I used= T R R
R T R
R R T

then I am able to get PK output in WinNonlin, but further I got stuck in
Bioequivalence Variance structure (in Bioequivalence Wizard)
Random effect Model - ????
Repeated specification - ????
Variance Blocking variable (Subject) - ????
Group -????

What I have to provide at the Question mark(????) place?
I am interested to get Bio equivalence output(90% CI and Ratio) also.


Kindly suggest me how to tackle with this problem.

Thanks and Regards,

Naresh

[Helmut Schütz]

Dear Naresh!

 

Just use WinNonlin's defaults – they are correct if you used the given coding. You should get the PE and CI from the BE wizard.

[Naresh]

Dear HS,

Thanks for reply,

yes, I am using same CI and PE but drug is highly variable and If I need to prove it is highly variable then Intrasubject variability of reference product should be >30% .

In Final variance parameter (BE output of WinNonlin) I am getting Variance of residual , using this I can calculate Intrasubject CV , Is this calculated Intrasubject CV of reference product or any other way I have to calculate Intrasubject CV for reference product??

kindly comment on my query.

Thanks and Regards,

Naresh

[Helmut Schütz]

Dear Naresh,

excuse me, but have you read the quote Simon gave you and my slides? Sigma_WR is the standard deviation of the reference treatment in a replicate design. It's a pitty that no clear explanation is given in WinNonlin's User Guide pp. 383...

[Naresh]

Dear HS,

Thanks for your reply,

I have read slides, but in this slide (slide 40 of 66) shows output from Population/ Individual Bioeq ,
do you wants to say that for getting SigmaWR I have to used Population/Individual BE method. and for 90% CI and PE used Average BE method.

I may be confuse somewhere, and I hope you have understand my problem.

Please comment on this..

Thanks and Regards,

Naresh Shirsath

[Helmut Schütz]

Dear Naresh,

sorry, but my WinNonlin5.3 is not working right now (see here). In Phoenix you get the value also for ABE. I will check it when I have resolved the software issue on my machines.

EDIT by Mod...
Naresh, I am pretty sure this was not changed with Phoenix so you should also be able to do this in version 5.x Simon.

[Helmut Schütz]

Well, cough... I guess you are right here! I was always wondering why there's a difference in variance estimates in WNL5x depending on whether ABE or PBE/IBE was selected.

According to a hidden document (somewhere at Pharsight's site) in WNL's ABE:
Var(period*treatment*subject)_21 = sigma²_WR
Var(period*treatment*subject)_22 = sigma²_WT

For my example dataset we get 0.214840 and sigma_WR (pocket calculator) 0.463508

If I run PBE/IBE in v5.3 (yes I have installed WNL again and it's running with some hicups), I get Sigma_WR = 0.46277 (that's the value in my slides).

Obviously 0.463508 # 0.46277, but why?

@Simon: yes, PHX6.1 gives exactly the same results as WNL5.3

Go to the support site and look for the document "How does WinNonlin Bioequivalence Wizard calculate Intrasubject and Intersubject CV?"
At the bottom you find:
"After the analysis is run, the Population/Individual Bioequivalence tab contains these parameters. SigmaWR is the within-subject variation for the reference formulation. SigmaR is the total variance of the reference formulation, i.e. the sum of within- and between-subject variance. To get the between-subject variation, subtract SigmaWR from SigmaR." My emphases in bold.
Took me ages to find out that this is a standard deviation and not a variance. BTW, what is a "variation"? Not only in my sample dataset (tested quite a lot of others as well) the between-subject variance is smaller than the within-subject variance, which is counterintuitive. To be honest, I'm not happy with both the results and the documentation. When it comes to a replicate design I would go for SAS or R-package bear.

See also this lenghty thread.

IMHO Pharsight should validate the results against FDA's SAS-code. Unfortunatelly FDA does not give an example dataset. EMA is preparing a document, where both SAS-code and a dataset will be published. Compliance will be mandatory...

[Naresh]

Dear HS,

 

Thanks for providing reference,

 

I have gone through it, and its quite lenghty.

now for CI and Ratio I can use ABE but problem is about SigmaWR .

so what you think , finally what I have to do for getting SigmaWR ???

 

kindly comment...

 

Thanks and Regards,

 

Naresh

[Helmut Schütz]

Dear Naresh!

Thanks for providing reference, I have gone through it, and its quite lenghty.

Yes, I know - it's quite complicated. We could not resolve the discrepancies between SAS and WinNonlin; although with another dataset we got a quite close match. IMHO it's scary that for one dataset we failed and with another one we succeeded...

See this even more lengthy thread. At the bottom of my post you find a comparison of results. If you are not confused afterwards, search the BABE-forum with keywords like "RSABE", "Scaling", or "HVD" for more chaos.

now for CI and Ratio I can use ABE but problem is about SigmaWR .
so what you think , finally what I have to do for getting SigmaWR ???

Reading all that stuff again, I would opt for the output of the ABE-module, where
Var(period*treatment*subject)_21 = sigma²WR (note: already the variance, not the standard deviation). But:
[ul][li]Var(period*treatment*subject)_21 = sigma²WR is undocumented (at least I could not find it)[/li][li]The discrepancy between the result of the ABE and the PBE/IBE module is unclear (to me, of course)[/li][li]This is an open forum - not Pharsight's support. Don't rely on my opinions - right now I would not dare to use PHX/WNL for sABE (which is a pity).[/li][li]Use FDA's SAS-code to get sigma²_WR - even for a European submission (but perform the scaling according to EMA's rules).[/li][li]If you want to be on the safe side (highly recommended!) wait for EMA's SAS-code and example dataset. Last response from my side - I'm off for a conference in Munich next week; maybe I can provide some news concerning this document by the end of next week.[/li][/ul]

[Helmut Schütz]

Dear Naresh!

 

Latest news: EMA's document (SAS-code + dataset) was recently finalized and will be published soon.

[Naresh]

Dear HS,

 

Thanks for your response,

 

As per your suggestions, I have started to to work on SAS code provided in the USFDA guideline (http://www.fda.gov/d...s/UCM209294.pdf).

I will try to use this code for SABE and getting CV for reference product.

I will discuss result of the same with you.

 

also waiting for the EMA's document (SAS-code + dataset).

 

 

Thanks again,

 

 

Thanks and Regards,

 

Naresh Shirsath

[Naresh]

Dear HS,

I have used reference (www.fda.gov/downloads/Drugs/GuidanceComp...dances/UCM209294.pdf).
and Herewith I have attached word file which includes the SAS code with dummy data and output obtained [file name=Partial_replicate_ScABE.doc size=37376]http://www.pharsight.com/extranet/media/kunena/attachments/legacy/files/Partial_replicate_ScABE.doc[/file] .


Output:

Obs id dfd s2wr theta y boundy sWR x boundx critbound
1 1 9 0.25856 0.79669 -0.20599 -0.10958 0.50849 0.087543 0.44085 0.24778

from above result, my conclusions are,
s2wr(=0.25856) = sigma²WR and should be used for calculating Intra CV for reference product in the formula CV_WR = sqrt(e^sigma²WR-1)

critbound (=0.24778) is greater than 0 , ( as per Step 3 condition a on page 3 of www.fda.gov/downloads/Drugs/GuidanceComp...dances/UCM209294.pdf )
Test product is not bio equivalent to the reference product,


Please go through the attached SAS code and results, and kindly let me know whether my result are correct or not?

Thanks and Regards,

Naresh

[Helmut Schütz]

Dear Naresh!

my conclusions are,
s2wr(=0.25856) = sigma²WR and should be used for calculating Intra CV for reference product in the formula CV_WR = sqrt(e^sigma²WR-1)

Right!

Please go through the attached SAS code and results, and kindly let me know whether my result are correct or not?

Looks OK to me – but I'm not a SAS-user.

Now let's have a look at WinNonlin 5.3: Interestingly WNL does not 'recognize' the replicated structure (why?) in the ABE-setting; therefore you have to adjust the model. In the BE-wizard keep the model as seq+trt+per, select Variable Transformation 'Already ln-transformed' Next>
change in Random 1-tab to Random Effects Model 'trt', Variance Blocking Variables (Subject) 'subj', Type 'Banded No-Diagonal factor Analytic(f)', Number of factors (f) = 2; Repeated-tab to Repeated Specification 'per', Variance Blocking Variables (Subject) 'subj', Group 'trt'

Var(per*trt*subj)_21 = 0.350924 is not the expected variance of the reference (FDA's SAS-output 0.25856). Therefore we can't use this value. Maybe my model coding is wrong?

Now let's select PBE/IBE:in the entry window of the BE-wizard. We get SigmaWR = 0.508488654276 – matching FDA's SAS-output (0.50849). Since CV_WR = 54.3% you are allowed to scale. No idea whether it's possible to set up WNL in such a way that you fulfill FDA's requirements.

If you want to go for a European submission: Your data are AUC – no scaling is allowed. ;-) Let's assume they are C_max; CV>30% scaling allowed. CV>50% must be treated as if CV=50%.

In both regulations the point estimate (GMR) is restricted to 80.00%–125.00%; bad luck for your dataset with T/R=141%.

But let's try to ignore all this and let's have a look how the procedure for EMA would be set up. The regulatory constant k is 0.760. The upper/lower limits are calcultated according to exp(±k*s_WR) or 0.67946–1.4717. In WNL's terminology 'difference to detect' = 1-0.67946=32.0536%. Only one decimal place is allowed: always round down in order to get a conservative estimate: 32.0% instead of 32.1% (will be corrected in the next release of Phoenix, but presumably not in WNL). AH-limits from above. You should get Ratio% 141.0056, CI 90% (103.9705, 191.2328), Failed to show average bioequivalence for confidence=90.00 and percent=32.0.

@Pharsight: Please check.

[Helmut Schütz]

Dear all,

see also this post in the BEBA-forum.

[Simon Davis]

Helmut,

 

  Apologies for the delay, are going to look at that presentation, specifically slide 15, next week and hopefully I/they can post some clarification shortly; thanks for bringing it to our attention.

 

Have a good week-end,

                                       Simon.