Bioequivalence and Bioavailability Forum

Main page Policy/Terms of Use Abbreviations Latest Posts

 Log-in |  Register |  Search

Back to the forum  Query: 2017-05-29 18:58 CEST (UTC+2h)
 
pa1kumar.anna
Junior

India,
2008-12-01 09:51
(edited by pa1kumar.anna on 2008-12-01 13:14)

Posting: # 2843
Views: 14,400
 

 SAS code of partial replicate design [Software]

DEAR ALL,

please provide the sas code of partial replicate design.

--
with regards
APK
Ohlbe
Hero

France,
2008-12-01 10:45
(edited by Ohlbe on 2008-12-02 09:40)

@ pa1kumar.anna
Posting: # 2844
Views: 13,017
 

 Etiquette

Dear pa1kumar.anna,

Please read the Forum policy and instructions. For instance:
Be polite!
It's nice to start your post with a salutation, and include a signature as well.
You may save a signature with your User's data; it will be automatically attached to your posts.



This will greatly improve your chances of getting an answer.

Regards
Ohlbe

--
Edit: initial message edited by APK to add salutation and signature :-)
d_labes
Hero

Berlin, Germany,
2008-12-02 09:17

@ pa1kumar.anna
Posting: # 2858
Views: 12,947
 

 SAS code for what?

Dear APK,

» please provide the sas code of partial replicate design.

Be a little bit more specific.
What do you mean with partial replicate design? A 3-period replicate design? What SAS code do you need? Average bioequivalence? Or what else?

BTW: Hope your user name is not your password :-D ;

Regards,

Detlew
Helmut
Hero
Homepage
Vienna, Austria,
2008-12-03 15:42

@ d_labes
Posting: # 2873
Views: 13,070
 

 partial replicate = 3-period replicate

Dear DLabes,

» Be a little bit more specific.
» What do you mean with partial replicate design? A 3-period replicate design?

I've heard this term for the first time at the recent Workshop at Ahmedabad. Yes, it's a 3-period replicate design.

[image]All the best,
Helmut Schütz 
[image]

The quality of responses received is directly proportional to the quality of the question asked. ☼
Science Quotes
MGR
Regular

India,
2008-12-04 11:19

@ Helmut
Posting: # 2876
Views: 13,136
 

 partial replicate = 3-period replicate

Dear Helmut,

We too heard this type of design and we are also planning for the study as per the sponsor's request.

Here we got the type like Partial scaled average Replicate Bioequivalence study. In this study, we had two treatments forming 3 period 3 sequence design as follows:

1) T R R
2) R T R
3) R R T


Now my question is:

Then what is the difference between an Scaled average Replicate 3-way cross over study with the above stated one?

Please clarify me regarding this.

Thanks in advance.

Regards,
MGR
d_labes
Hero

Berlin, Germany,
2008-12-04 14:17

@ Helmut
Posting: # 2877
Views: 14,496
 

 3-period replicate designs and SAS

Dear Helmut, dear all,

» I've heard this term for the first time at the recent Workshop at Ahmedabad. Yes, it's a 3-period replicate design.

Taking the Holy Bible[1] of cross-over designs the best known 3-period replicate design has the name 3-period dual design with the two sequences

TRR
RTT

In the FDA Guidance an other 3-period replicate design is recommended:

TRT
RTR

But nowhere it is named "partial replicate design".

I have found this term in discussions (FDA and others) about scaled average bioequivalence. But here it denominates a one-sequence design :ponder: with replication of the reference only, for instance

TRR.


To answer the original question:
If the design is a 3-period design with more then one sequence, to my knowledge the SAS code for the evaluation of average bioequivalence does not depend on the specific design used in replicate studies, provided you will go with restricted maximum likelihood estimation (Proc MIXED in "The power to know").

The code recommended in the FDA guidance was discussed already in breadth on this forum. See for instance this thread and others (use search!).
Let me give it here again for your convienience.

Proc MIXED data=YourData method=REML alpha=0.1;
  class treatment period sequence subject;

  model Y= treatment period sequence / ddfm=satterth;     
  random treatment /subject=subject type=FA0(2) G;
  repeated /group=treatment subject=subject type=simple;

  ODS output LSMeans=_LSMeans Diffs=_CI;

  lsmeans treatment/diff cl alpha=0.1;
run;


Y is the pharmacokinetic target (eventually log transformed f.i. for AUC, Cmax). The ODS output statement saves you the least square means and the 90% confidence intervals in SAS datasets for further processing, f.i. to back transform them into the original scale if your target was log-transformed.
Be aware that SAS always gives you the difference (and 90% confidence interval) in least square means in lexicographic order, namely R-T if you code your treatments as R(eference) and T(est).
So do not forget to change the sign in subsequent processing!

But there are variants of this code (covariance structure other than FA0(2), other ddfm=denominator degrees of freedom) and there are complete other models to choose from!
See for instance [2] and [3].

If it is an ominous one-sequence design I think we have no sequence and period effects in the model.?

  1. B Jones and MG Kenward
    Design and Analysis of Cross-over Trials
    Chapman & Hall/CRC, Boca Raton, Chapter 4 (2nd ed. 2003)
  2. Patterson and Jones
    Replicate Designs and Average, Individual, and Population Bioequivalence
    GSK BDS Technical Report 2002 – 01 (part I)
    GSK BDS Technical Report 2002 – 05 (part II)
  3. S.A. Willavize and E.A. Morgenthin
    Comparison of models for average bioequivalence in replicate crossover designs
    Pharmaceut. Stat. Vol.5 (3), p. 201 - 211 (2008) published online 24 May 2006


Edit: Updated URLs. [Helmut]

Regards,

Detlew
d_labes
Hero

Berlin, Germany,
2008-12-12 10:51

@ d_labes
Posting: # 2912
Views: 13,487
 

 Extra-reference or partial replicate design

Dear all,

FYI:
The 3-sequence-3-period replicate design with the sequences

TRR
RTR
RRT

mentioned in MGR's post above is handled in the framework of the so-called Method of moments in

[1] R.J. McNally
Tests for Individual and Population Bioequivalence Using 3-Period Crossover Designs

and the 2-sequence variant using only the two first sequences in

[2] S.-C. Chow, J. Shao and H. Wang
Individual bioequivalence testing under 2×3 designs
Statist. Med. 2002; 21:629–648

which can be found [1] here and [2] there.

These papers deal with individual BE but the parts relevant for average BE can easily extracted.

Interesting enough these papers state, that the intra-individual variance component for T(est) is not identifiable/estimable, due to the "partial" replicate nature of these designs (replicates only for R).
Thus one would expect some difficulties with the FDA SAS code, which has a covariance parameter for that in the model. But my little experimentation with it and some artificial data for an extra-reference design show that the code is able to deliver a value for that variance parameter sWT.
Black magic :confused: or some sort of perpetuum mobile of information?

Because the intra-individual variance for the Reference is identifiable (regardless which method, REML or method of moments) one could go with Reference scaled ABE as described in
Haidar et al.
Evaluation of a scaling approach for the bioequivalence of highly variable drugs
AAPS J. 2008 Sep;10(3):450-4. Epub 2008 Aug 26

Regards,

Detlew
Helmut
Hero
Homepage
Vienna, Austria,
2008-12-12 11:34

@ d_labes
Posting: # 2913
Views: 13,049
 

 Extra-reference or partial replicate design

Dear D. Labes,

thanks for your references!

» Black magic :confused: or some sort of perpetuum mobile of information?

Neither nor - just
[image] :-D
For my simple mind it seems to be quite strange to get the variance of a measurement we have performed only once.
:ponder:

[image]All the best,
Helmut Schütz 
[image]

The quality of responses received is directly proportional to the quality of the question asked. ☼
Science Quotes
d_labes
Hero

Berlin, Germany,
2008-12-12 13:28

@ Helmut
Posting: # 2914
Views: 13,035
 

 Simple minded

Dear Helmut,

» » Black magic :confused: or some sort of perpetuum mobile of information?
»
» Neither nor - just the 'power to know'!
» For my simple mind it seems to be quite strange to get the variance of a measurement we have performed only once.

I am the last to tout "SAS all over the world" (although I have to deal with the 'power to know' most time of my working day). But here I must defend it.

I guess this is not a problem of SAS. It may be a problem of what we (I) do with it. Fitting an inappropriate? or over-specified? model.

Eventually anybody can fit the underlying model in WINNONLIN or R using REML to see what happens?
I have taken my data to play with from example 4.2 of Patterson/Jones "Bioequivalence and Statistics in Clinical Pharmacology" (C5300.zip with data and SAS code), changed the original sequences to TRR/RTR.

I place a bet that then also a fitted value for s2WT is obtained :-P . Maybe it is even correct!?
(I confabulate: total variability of T: estimable; variability of T-R: estimable and sum of intra-subject variabilities + treatment*subject interaction; intra-subject variability of R: estimable; so ......)

BTW: Simple minds are more then 30 years successful :-D .

Regards,

Detlew
Helmut
Hero
Homepage
Vienna, Austria,
2008-12-12 14:58

@ d_labes
Posting: # 2916
Views: 13,102
 

 Simple minded

Dear D. Labes,

» I guess this is not a problem of SAS. It may be a problem of what we (I) do with it. Fitting an inappropriate? or over-specified? model.
»
» Eventually anybody can fit the underlying model in WINNONLIN or R using REML to see what happens?

» I place a bet that then also a fitted value for s2WT is obtained :-P . Maybe it is even correct!?

I fitted Cmax with your modified sequences (because I didn't want to start with incomplete AUC data) in WinNonlin to PBE/IBE and obtained:
SigmaR  0.9999185
SigmaWR 0.4480995
(no SigmaWT!)

» (I confabulate: total variability of T: estimable; variability of T-R: estimable and sum of intra-subject variabilities + treatment*subject interaction; intra-subject variability of R: estimable; so ......)

If I run ABE in WinNonlin I'm lost in the options; choosing the defaults (no time to dig into the manual right now...)
fixed:    sequence+treatment+period
random:   subject(treatment) Type: Banded No-Diagonal Factor Analytic(f=2)
repeated: subject(period)   Group: treatment Type: Variance components

I get
Final variance parameter estimates:
           lambda(1,1)_11         0.885348
           lambda(1,2)_11         0.922848
           lambda(2,2)_11         0.213429
Var(period*treatment*subject)_21  0.224827
Var(period*treatment*subject)_22  0.0853385

where according to this post lambda(1,1)_11, lambda(1,2)_11 and lambda(2,2)_11 correspond to SAS' FA(1,1), FA(2,1) and FA(2,2)...

[image]All the best,
Helmut Schütz 
[image]

The quality of responses received is directly proportional to the quality of the question asked. ☼
Science Quotes
d_labes
Hero

Berlin, Germany,
2008-12-12 15:44

@ Helmut
Posting: # 2918
Views: 12,985
 

 SAS and WinNonlin

Dear Helmut,

thanks for your very quick replay.

» I fitted Cmax with your modified sequences (because I didn't want to start with incomplete AUC data) in WinNonlin to PBE/IBE and obtained:
» SigmaR  0.9999185
» SigmaWR 0.4480995 (no SigmaWT!)

Seems 'Method of moments' is used here?

» If I run ABE in WinNonlin I'm lost in the options; choosing the defaults [...]» I get
» Final variance parameter estimates:
»            lambda(1,1)_11         0.885348
»            lambda(1,2)_11         0.922848
»            lambda(2,2)_11         0.213429
» Var(period*treatment*subject)_21  0.224827
» Var(period*treatment*subject)_22  0.0853385

Here are the covariance parameters from Proc MIXED (FDA code)

                        Covariance Parameter Estimates

  Cov Parm   Subject   Group          Estimate   Alpha     Lower     Upper

  FA(1,1)    subject                    0.8865     0.1    0.7644    1.0425
  FA(2,1)    subject                    0.9554     0.1    0.8036    1.1073
  FA(2,2)    subject                  3.06E-17       .         .         .
  Residual   subject   treatment R      0.1931     0.1    0.1500    0.2597
  Residual   subject   treatment T      0.1277     0.1    0.07846   0.2518


With the exception of lambda(1,1) there seems no match :-( !
Now we need someone with knowledge of relationship of SAS and WINNONLIN parameters.
Is var(period*treatment*subject) the intra-subject variability?

But enough to now. Have a nice weekend.

Regards,

Detlew
Helmut
Hero
Homepage
Vienna, Austria,
2008-12-12 16:33

@ d_labes
Posting: # 2919
Views: 12,799
 

 SAS and WinNonlin

Dear D. Labes,

thanks for SAS' output!

» Seems 'Method of moments' is used here?

Yes – but only to get initial variance estimates.

» With the exception of lambda(1,1) there seems no match :-( !

Yes. :confused:

» Now we need someone with knowledge of relationship of SAS and WINNONLIN parameters.

We should wait for Simon Davis coming by (he has also experience with SAS)...

» Is var(period*treatment*subject) the intra-subject variability?

Yes, well the variance (and you get the CV in the usual way; not built-in in WinNonlin: pocket-calculator business):
Var(period*treatment*subject)_21 sigma²reference
Var(period*treatment*subject)_22 sigma²test

[image]All the best,
Helmut Schütz 
[image]

The quality of responses received is directly proportional to the quality of the question asked. ☼
Science Quotes
MGR
Regular

India,
2008-12-30 10:12

@ Helmut
Posting: # 2985
Views: 12,759
 

 Switchability and Prescribability in WinNonlin

Dear Helmut,

In the previous reply, you have given the output of WinNonlin like,
» Final variance parameter estimates:
»            lambda(1,1)_11         0.885348
»            lambda(1,2)_11         0.922848
»            lambda(2,2)_11         0.213429
» Var(period*treatment*subject)_21  0.224827
» Var(period*treatment*subject)_22  0.0853385

But my doubt is that, Can we find the Switchability, Prescribability and Global Variance from the above data in WinNonlin? If so can you please give me the formulas to calculate these parameters? As we are dealing a project of replicate, in the protocol they had mentioned these parameters.

Thanks in advance.

Regards,
MGR
Helmut
Hero
Homepage
Vienna, Austria,
2008-12-30 16:04

@ MGR
Posting: # 2991
Views: 12,891
 

 PBE and IBE in WinNonlin

Dear MGR!

» In the previous reply, you have given the output of WinNonlin like […],

This was the evaluation for Average Bioequivalence (ABE). In WinNonlin’s BE Wizard entry window:
  Type of Bioequivalence                     
    ⦿ Average       ○ Population/Individual


» […] Can we find the Switchability, Prescribability and Global Variance from the above data in Winnonlin?

Prescribability is assessed by Population Bioequivalence (PBE); Switchability by Individual Bioequivalence (IBE). In WinNonlin's BE Wizard entry window:
  Type of Bioequivalence                     
    ○ Average       ⦿ Population/Individual


» If so can you please give me the formulas to calculate these parameters?

For a basic reference see

WW Hauck and S. Anderson
Measuring Switchability and Prescribability: When Is Average Bioequivalence Sufficient?
J Pharmacokin Biopharm 22(6), 551–64 (1994)


Just run the BE Wizard. For lnCmax of the example data set we get:
Population Bioequivalence Statistics

    Difference (Delta) : -0.129898
           Ratio(%Ref) : 87.818513
                SigmaR : 0.999918
 Reference-scaling eta : -1.875611,    Upper limit : -1.149187
  Constant-scaling eta : -0.118141,    Upper limit : 0.294625
      sigmaR > sigmaP? : yes, mixed scaling uses reference-scaling eta

           Conclusions : BE shown for ratio test.
                         Population BE shown for refnc-scaling CI test.
                         Population BE not shown for const-scaling CI test.
                         Population BE shown for mixed-scaling CI test.

Individual Bioequivalence Statistics

    Difference (Delta) : -0.129898
           Ratio(%Ref) : 87.818513
               SigmaWR : 0.448099
 Reference-scaling eta : -0.516803,    Upper limit : -0.234932
  Constant-scaling eta : -0.115652,    Upper limit : 0.041727
     sigmaWR > sigmaI? : yes, mixed scaling uses reference-scaling eta

           Conclusions : BE shown for ratio test.
                         Individual BE shown for refnc-scaling CI test.
                         Individual BE not shown for const-scaling CI test.
                         Individual BE shown for mixed-scaling CI test.


» […] in the protocol they had mentioned these parameters.

Not knowing how to evaluate the study? :lol3:
IMHO both PBE and IBE are of historic interest only.

[image]All the best,
Helmut Schütz 
[image]

The quality of responses received is directly proportional to the quality of the question asked. ☼
Science Quotes
MGR
Regular

India,
2008-12-31 07:52

@ Helmut
Posting: # 2994
Views: 12,705
 

 PBE and IBE in WinNonlin

Dear Helmut,

Thanks for the Reply.

The exact doubt is that these three (Switchability,-----) terms are mentioned in the average bioequivalence study protocol but not in the Ind/Pop BE study.

So i got confused with the protocol :confused:

So i want to find these terms according to the Average BE Study not by using Ind/Pop BE study.

Thank you,

Regards,
MGR
NPavan
Junior

India,
2009-01-12 13:10

@ MGR
Posting: # 3021
Views: 12,582
 

 Switchability

Dear all,
I am new to this field. I am a bio statistician. These forums are very helpful to beginners like me. I have seen in a protocol regarding switchability concept in average bioequivalence. Can anybody please give me formula/reference/website for calculating switchability.
Thanks in advance.

Regards
Pavan


See the post above. [Helmut]

Regards,
Pavan
Back to the forum Activity
 Thread view
Bioequivalence and Bioavailability Forum | Admin contact
16,931 Posts in 3,629 Threads, 1,030 registered users;
12 users online (0 registered, 12 guests).

No effect that requires more than ten percent accuracy in measurement
is worth investigating.    Walther Nernst

The BIOEQUIVALENCE / BIOAVAILABILITY FORUM is hosted by
BEBAC Ing. Helmut Schütz
XHTML/CSS RSS Feed