zan
☆    

US,
2014-06-28 01:29
(3561 d 18:22 ago)

Posting: # 13161
Views: 27,288
 

 Phoenix WNL to handle 2x2 crossover study with multip group [Software]

Hi All,
I am working on a 2x2 crossover BE study with large N (N=54). Due to limited clinical capacity the site separated subjects into two dosing groups (N=27 each). I have come across several discussion threads in the forum regarding the use of SAS model to deal with this issue, however, I would like to see how is this modeled correctly in Phoenix Winnonlin 6.3 as I only have WNL available.
Per FDA guidance the model should reflect that the periods in the first group are different from the periods in the second group. I immediately have two thoughts about this:
  1. is it acceptable to just assign 4 periods to the 4 dosing days? If so,
    In the model specification: Treatment+Sequence+Period
    and in the variance structure--Random effect: Subject(Sequence).
    Is there any issue with this approach? It looks to me these periods are independent occasions, which might not really separate each two periods for the same group.
  2. Can we still keep these 2 periods and add a Group term to these subjects (eg. Group 1 and Group 2)?. If so, in the model specification:
    Treatment+Sequence+Period+Group
    but in the variance structure, I dont know what is appropriate to be added in the Random effect field. Should it be Subject(Sequence*Group).
I wanted to make sure I choose the correct procedure in PHX Winnonlin so that when FDA asks whether we properly deal with the issue with multiple dosing groups.

Really appreciate for any help.

Zan


Edit: Category changed. [Helmut]
ElMaestro
★★★

Denmark,
2014-06-29 01:23
(3560 d 18:27 ago)

@ zan
Posting: # 13166
Views: 25,527
 

 Phoenix WNL to handle 2x2 crossover study with multip group

Hello Zan,

I don't speak WNL but:

❝ 1. is it acceptable to just assign 4 periods to the 4 dosing days? If so,

❝ In the model specification: Treatment+Sequence+Period

❝ and in the variance structure--Random effect: Subject(Sequence).


That would be a little untraditional with four period levels. But it can be done and it is a little like working with period in group. What's the title of your study? (If it is something like "A two-treatment, two-period, two-sequence bioequivalence study blah blah" then you almost have the answer right there).

❝ 2. Can we still keep these 2 periods and add a Group term to these subjects (eg. Group 1 and Group 2)?. If so, in the model specification:

❝ Treatment+Sequence+Period+Group


Yes, much better, this is what I would do.

❝ but in the variance structure, I dont know what is appropriate to be added in the Random effect field. Should it be Subject(Sequence*Group).


Oh here we go again: Depending on your data listing you can probably just write Subject.

Pass or fail!
ElMaestro
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-06-30 18:55
(3559 d 00:56 ago)

@ zan
Posting: # 13174
Views: 25,919
 

 Group effect commonly not relevant

Hi Zan,

first I agree with what ElMaestro wrote. Two groups are no issue generally (same protocol, same site, not too large interval [months!] between groups).
Maybe you are interested in this thread and some quotes from the FDA. I don’t have Win­Nonlin any more (consider upgrading to Phoenix; release 6.4 is expected this summer). My setup in Phoenix / Win­Nonlin 6.3:
  • Group model I:
    fixed:  Group+Sequence+Sequence(Group)+Period(Group)+Treatment+Treatment×Group
    random: Subject(Sequence×Group)

    If the Treatment-by-Group term is not significant (p ≥0.1) this term can be dropped from the model, proceeding to Group Model II.
    If the Treatment-by-Group term is significant (p <0.1), FDA’s Division of Bioequivalence requests that equiva­lence be demonstrated in one of the groups, provided that the group meets minimum requirements for a com­plete BE study (i.e., 12).
  • Group model II:
    fixed:  Group+Sequence+Sequence(Group)+Period(Group)+Treatment
    random: Subject(Sequence×Group)

  • Conventional model:
    fixed:  Sequence+Period+Treatment
    random: Subject(Sequence)

    If all conditions stated in FDA’s letters are fulfilled (most cases), that’s the way to go.
Like any between-subject test the one for Treatment-by-Group is lacking power (therefore, testing at α 0.10). However, you could expect then a false-positive rate in 10% of studies!

❝ I am working on a 2x2 crossover BE study with large N (N=54). Due to limited clinical capacity the site separated subjects into two dosing groups (N=27 each).


IMHO, a 1:1 split is not optimal. Let’s assume the worst, i.e., for any wacky reasons FDA insist on model I and there is a significant Treatment-by-Group Interaction. No pooling; go with one of the groups. In your design both have a size of 27. Let’s assume a T/R of 0.95, CV 30.5%, and target power 90%. You end up with 54 subjects (power 90.38%). BTW, your 1:1 split results in imbalanced sequences in both groups. Let’s further assume that in both groups the T/R-ratio and CV come out exactly as planned and you have no drop-outs. Power will be just 60.97% in both. If both groups are equally sized the questions also arises on which one you will base the assessment of BE. The nicer one? :cherry picking:
I would suggest to split subjects in such a way that one of the groups is the maximum capacity of the clinical site. Which power can we expect in the groups?
G1  % power  G2  % power
────────────────────────
27   60.97   27   60.97 
28   63.13   26   58.63 
30   66.92   24   53.92
32   70.27   22   48.33 
34   73.25   20   41.98 
36   75.89   18   34.91 
────────────────────────

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
zan
☆    

US,
2014-06-30 21:18
(3558 d 22:33 ago)

@ Helmut
Posting: # 13175
Views: 25,449
 

 Group effect in 2x2 BE study

Thanks very much for the useful reply, ElMaestro and Helmut! Yes I am using Phoenix Winnonlin to work on this analysis. If there is a significant treatment*Group effect, is the overall (all subject) ratio (CI90%) still valid, or we have to adopt the ratio (CI90%) from separate groups?

Regarding FDA's requirement below (equiva­lence be demonstrated in one of the groups), do we have a link to the source of the letter? I am hoping to keep it as a reference for study files.

❝ If the Treatment-by-Group term is not significant (p ≥0.1) this term can be dropped from the model, proceeding to Group Model II.

❝ If the Treatment-by-Group term is significant (p <0.1), FDA’s Division of Bioequivalence requests that equiva­lence be demonstrated in one of the groups, provided that the group meets minimum requirements for a com­plete BE study (i.e., 12).


Best regards
zan
ElMaestro
★★★

Denmark,
2014-06-30 22:36
(3558 d 21:14 ago)

@ zan
Posting: # 13176
Views: 25,421
 

 Group effect in 2x2 BE study

Hi Zan,

❝ Thanks very much for the useful reply, ElMaestro and Helmut! Yes I am using Phoenix Winnonlin to work on this analysis. If there is a significant treatment*Group effect, is the overall (all subject) ratio (CI90%) still valid, or we have to adopt the ratio (CI90%) from separate groups?


I personally find it very difficult to address a significant group x treatment interaction. To help visual interpretation you could plot the four effects in a bar diagram. A low p-value for group x treatment means you think the four bars all not of equal height. Where does that leave us?
ABE is all about the treatment performance in itself and not treatment in summer versus treatment in winter (the latter just to make it a little extreme, hope you get the general idea). Sure, if there is a pronounced difference in stability of T and R, respectively, and the two groups are well separated in time then we could observe group x treatment, but this is speculation and besides if such stability differences would exist they should be caught and handled well before any study is kicked off cf. GMP and GCP.
I am inclined to say a significant group x treatment should be considered a mere nuisance just like Conchita Wurst or nail fungus.

Pass or fail!
ElMaestro
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-06-30 22:51
(3558 d 21:00 ago)

@ zan
Posting: # 13177
Views: 25,310
 

 To pool or not to pool

Hi Zan,

❝ […] If there is a significant treatment*Group effect, is the overall (all subject) ratio (CI90%) still valid, or we have to adopt the ratio (CI90%) from separate groups?


If you pool groups you’ll end up with common estimates of between- and within-subject variances. If groups are different in their responses to the treatments you will not get pooled unbiased esti­mates. AFAIK, there is not corrective method. Theoretically – if you would know (!) the correlations between intra-/inter-subject variances, take the group sizes into account, :blahblah: – maybe some fancy weighting is possible. But: Estimates of these effects are not accessible in a simple 2×2 cross-over. Too many questions asked to poor little data.

❝ Regarding FDA's requirement below (equiva­lence be demonstrated in one of the groups), do we have a link to the source of the letter? I am hoping to keep it as a reference for study files.


Nope. I have seen similar wordings in confidential letters (and Detlew as well). Most phrases were identical to the letter. Copy/paste at the FDA?

❝ ❝ If the Treatment-by-Group term is not significant (p ≥0.1) this term can be dropped from the model, proceeding to Group Model II.


At least a similar quote in this post. I’m not in the office; will check tomorrow whether the signi­fi­cance level (0.05 different to FDA’s 0.1) is not a typo.

Do you have any good reasons to walk this dirty road at all? Multi-site study or the like? A split due to logistic reasons is very (very) common in BE. I have limited experience with the FDA, but in 30+ years I have neither seen a single BE study in Europe including group term(s) nor heard of any as­ses­sor asking for one. All trouble I experienced came from MENA-countries & Gulf states.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
zan
☆    

US,
2014-06-30 23:02
(3558 d 20:49 ago)

@ Helmut
Posting: # 13178
Views: 25,344
 

 To pool or not to pool

Thank you for all good comments (ElMaestro and Helmut)!

Helmut,
I just found a link to an FDA reviewer letter to a BE study where multiple groups were conducted, and one of the groups demonstrated a significant group by treatment effect. The criteira the reviewer cited was 0.1.

http://www.accessdata.fda.gov/drugsatfda_docs/anda/2008/077570Orig1s000BioeqR.pdf

Best regards


Edit: Full quote removed. Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post! [Helmut]
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-06-30 23:38
(3558 d 20:13 ago)

@ zan
Posting: # 13179
Views: 25,628
 

 To pool or not to pool

Hi Zan,

THX for the reference! With one week between groups I would never ever thought a milli­second of setting up a group model. Obviously this was also the approach of the applicant. In Table 9 (page 19) we read:

┌──────────────────────────┬──────────────────────────────────────┐
│ Do you agree or disagree │ Do not agree. Firm analyzed the data │
│ with firm’s decision?    │ as one group, the reviewer analyzed  │
│                          │ data as two separate groups.         │
└──────────────────────────┴──────────────────────────────────────┘


Without knowing the original data I will not comment on the reviewers comments. Do you think that she is aware of “false positives”?

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
zan
☆    

US,
2014-06-30 23:46
(3558 d 20:05 ago)

@ Helmut
Posting: # 13180
Views: 25,220
 

 To pool or not to pool

Hi Helmut,
Was it due to significant effect observed for Group x Treatment that resulted in the reviewer's decision to run final analysis by separate groups (on Page 5)?

Best Regards
zan


Edit: Full quote removed (again!). Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post! [Helmut]
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-01 00:22
(3558 d 19:28 ago)

@ zan
Posting: # 13182
Views: 25,464
 

 To pool or not to pool

Hi Zan,

❝ Was it due to significant effect observed for Group x Treatment that resulted in the reviewer's decision to run final analysis by separate groups (on Page 5)?


Sure. The text contains a reference:

In control #98-392 (see attachment), it is stated that “if the Group-by-Treat­ment interaction is statistically significant (p<0.1), DBE requested that equi­va­lence be demonstrated in one of the groups, provided that the group meets mini­mum requirements for a complete bioequivalence study”.

However, #98-392 is not attached. I failed to dig it out at the FDA (controlled corres­pon­dence not covered by FOI?). Note the wording almost identical to my post above.

Please no more TOFU in future posts. THX.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
zan
☆    

US,
2014-07-01 03:15
(3558 d 16:36 ago)

@ Helmut
Posting: # 13183
Views: 25,290
 

 To pool or not to pool

Hi Helmut,
The document #98-392 is hidden in page 52 of the PDF file (took me a while to dig it out).
Also, in PHX winnonlin, should partial test or sequential test the best approach to look for all the stat test results (eg. Group, treatment, period effect)? I tend to go with partial test as it is tested independent of order but not sure if I am correct.
zan
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-01 14:26
(3558 d 05:25 ago)

@ zan
Posting: # 13187
Views: 25,495
 

 Partial vs. sequential tests in WNL

Hi Zan,

❝ The document #98-392 is hidden in page 52 of the PDF file […].


It took the FDA until Sep 1999 to answer a letter received in Oct 1998. Note that this was well before the 2001 Stat. Guidance. Amazingly the letter deals with a one-week (!) washout and the same interval between groups.
┌───────────────┬──────────┬──────────┬──────────┬──────────┐
│ Dosing Scheme │   11/1   │   11/8   │   11/15  │   11/22  │
├───────────────┼──────────┼──────────┼──────────┼──────────┤
│       1       │  Group 1 │  Group 1 │  Group 2 │  Group 2 │
│               │ Period 1 │ Period 2 │ Period 1 │ Period 2 │
├───────────────┼──────────┼──────────┼──────────┼──────────┘
│       2       │  Group 1 │  Group 1 │          │
│               │ Period 1 │ Period 2 │          │
│               │          │  Group 2 │  Group 2 │
│               │          │ Period 1 │ Period 2 │
└───────────────┴──────────┴──────────┴──────────┘

It’s interesting that the FDA stated:

1. Both Dosing Schemes are acceptable to the Division of Bioequiva-
    lence.

With the second scheme you’ll have the entire bunch of sub­jects at the cli­ni­cal site on the second day. That’s bizarre; I don’t get it. If the CRO is large enough, why deal with two groups at all?

I would not bother about #1–8 and rather con­cen­trate on

9. If ALL of the following criteria are met, it may not be necessary to
    include Group-by-Treatment in the statistical model:

  • the clinical study takes place at one site;
  • all study subjects have been recruited from the same enrollment pool,
  • all of the subjects have similar demographics;
  • all enrolled subjects are randomly assigned to treatment groups at study outset.

In this latter case, the appropriate statistical model need include only the factors Sequence, Period, Treatment, and Subject (nested within Sequence).


❝ Also, in PHX winnonlin, should partial test or sequential test the best approach to look for all the stat test results (eg. Group, treatment, period effect)? I tend to go with partial test as it is tested independent of order but not sure if I am correct.


‘Partial Tests’ are expected to agree with SAS’ Type III LSMs in most of cases (whereas ‘Sequential Tests’ ≡ SAS’ Type I). See the “Phoenix User’s Guide” Table 17-13 (p448–50 = p478–80 of the PDF).

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
zan
☆    

US,
2014-07-01 20:03
(3557 d 23:48 ago)

@ Helmut
Posting: # 13196
Views: 25,179
 

 Partial vs. sequential tests in WNL

❝ I would not bother about #1–8 and rather con­cen­trate on

9. If ALL of the following criteria are met, it may not be necessary to

   include Group-by-Treatment in the statistical model:

● the clinical study takes place at one site;

● all study subjects have been recruited from the same enrollment pool,

● all of the subjects have similar demographics;

● all enrolled subjects are randomly assigned to treatment groups at study outset.

In this latter case, the appropriate statistical model need include only the factors Sequence, Period, Treatment, and Subject (nested within Sequence).


Hi Helmut,
On page 6 of the document (http://www.accessdata.fda.gov/drugsatfda_docs/anda/2008/077570Orig1s000BioeqR.pdf) we discussed above, reviewer provided the justification why the two dosing groups needed to be separated (looks stringent):

Although the subjects in the study were from the same geographic region and have the same demographic profiles, they were not enrolled in the study at the same time. The subjects enrolled in group 2 were screened after the beginning of the fasting study for group 1. There were 15 days in between the dosing of the two groups.
Therefore, individual statistical analysis is performed for each group.

zan
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-01 20:23
(3557 d 23:27 ago)

@ zan
Posting: # 13197
Views: 25,172
 

 Loosing patience

Hi Zan,

the company asked the FDA in October 1998 for a comment and – if I inter­pret the table correctly – planned the study for November. I’m too lazy to browse through the document, but I guess that sooner or later they gave up waiting and performed the study anyhow. Almost one year later the company received FDA’s answer. Too bad that they didn’t have a crystal ball (which would have told them to go with option #9).
As one of my friends use to say:

Don’t worry, it’s too late.

But we know better now.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
ElMaestro
★★★

Denmark,
2014-07-01 00:11
(3558 d 19:40 ago)

@ Helmut
Posting: # 13181
Views: 25,213
 

 To pool or not to pool

Hi Helmut and Zan,

❝ Without knowing the original data I will not comment on the reviewers comments. Do you think that she is aware of “false positives”?


Considering that the reviewer is Bing Li, I'd say the dossier had a very competent review.
However, the review is from 2005. That was around the time when ordinary sequence effects were proposed to be tested with 10% alpha and not considered nuisance. It might be the same phenomenon applied here to group which is also a between-subject term.
Today I think the policy has informally changed. As usual I could be wrong; at least I am not aware of any formalised guidance.

Pass or fail!
ElMaestro
zan
☆    

US,
2014-07-01 03:18
(3558 d 16:33 ago)

@ ElMaestro
Posting: # 13184
Views: 25,241
 

 To pool or not to pool

In looking at the BE test results, should we all use 0.1 as significance level? Does FDA require level of 0.05 for these tests?
Thanks,
zan
zan
☆    

US,
2014-07-02 21:56
(3556 d 21:54 ago)

@ ElMaestro
Posting: # 13202
Views: 25,080
 

 To pool or not to pool

I am still wondering does FDA have any guidance regarding what significant level is recommended (0.05 or 0.1) for assessing all effects. What if I observe a period or sequence effect at 0.06.
zan
ElMaestro
★★★

Denmark,
2014-07-03 02:17
(3556 d 17:34 ago)

@ zan
Posting: # 13204
Views: 25,108
 

 To pool or not to pool

Hello Zan,

❝ I am still wondering does FDA have any guidance regarding what significant level is recommended (0.05 or 0.1) for assessing all effects. What if I observe a period or sequence effect at 0.06.


As mentioned above, I do not think there is any formal guidance. If you have a reason to analyse a factor at 0.05 then do that. If you have a reason to analyse it at 0.1 then do that. I can't give you the optimal reason or the right alpha as it could come down to the specifics associated with your product/API. It all makes little sense to discuss alpha if you are not going to react to significance anyway. So this discussion is more than just the choice of alpha, it is also about what to do with tests that are significant.
Do what you think is right, fair, ethical and decent and what your SOP tells you to do.
At the end I think the principal distinction of BE versus not BE is made on basis of the confidence interval so the decision is not directly associated to the P-values of the ANOVA (that's unless you are a Danish regulator but we'll leave it out of this discussion for now).

Bear in mind that in the ANOVA you are testing hypothesis of equality.
This stands in contrast to the null hypothesis we evaluate when we test for bioequivalence.
Therefore, to me most anova p-values (regardless of alphas) are of secondary importance. If your randomisation was not effective (all the fat guys ended up in seq TR and all the slim ones in RT) then you might well see a sequence effect. If your first period was in the summer and the second period was in the winter then you might see a period effect. Etc. It will not necessarily have much to do with product performance.

Pass or fail!
ElMaestro
zan
☆    

US,
2014-07-03 02:36
(3556 d 17:14 ago)

@ ElMaestro
Posting: # 13205
Views: 25,149
 

 To pool or not to pool

Thank you, ElMaestro. This makes sense to me. Looks like as long as the significant level is chosen consistently throughout the analysis.
zan
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-03 12:05
(3556 d 07:45 ago)

@ zan
Posting: # 13208
Views: 25,129
 

 sequence- / period effect #666

zan,

❝ What if I observe a period or sequence effect …


though actual number might be lower, it feels like we had similar questions a hun­dred times. Please search the forum. You will find not only answers but a lot of re­fe­rences as well.

❝ … at 0.06.


0.05 < 0.06 < 0.10…

<irony>

The “Box and whiskers plot” was introduced* by John W. Tukey. He sug­gested the quartiles ±1.5× the inter­quar­tile range (IQR) as warning limits (aka “inner fences”) for “mild” outliers (and 3×IQR aka “outer fences” for “se­vere” ones).
At a conference he was asked “Why 1.5, Mr Tukey?” and replied “Well, that’s larger than one and smaller than two.”

</irony>


  • JW Tukey
    Exploratory Data Analysis
    Addison-Wesley Publishing Co., Reading, MA (1977)

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
AngusMcLean
★★  

USA,
2014-07-06 18:19
(3553 d 01:31 ago)

@ Helmut
Posting: # 13219
Views: 25,087
 

 Group effect commonly not relevant

❝ […] (consider upgrading to Phoenix; release 6.4 is expected this summer). My setup in Phoenix / Win­Nonlin 6.3:

❝ ● Group model I:

fixed:  Group+Sequence+Sequence(Group)+Period(Group)+Treatment+Treatment×Group

random: Subject(Sequence×Group)


Helmut: I saw above post focusing on performing studies in a large number of subjects so it is convenient to split the subjects into 2 groups. I created a data set with two groups (2 way crossover study with 2 treatments and two periods and two sequences) to try out above code in Phoenix WinNonlin 6.3 (V6.4 will be available in August).

I tried to set up the model from above, but I found the program rejected the code I put it in (text turns red). I do not know what boxes to check to get the program to accept the code, which includes "Group".

Group+Sequence+Sequence(Group)+Period(Group)+Treatment+Treatment*Group

Please can you advise accordingly,

Angus
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-06 22:32
(3552 d 21:19 ago)

@ AngusMcLean
Posting: # 13221
Views: 25,072
 

 Group effect commonly not relevant

Hi Angus,

❝ I tried to set up the model from above, but I found the program rejected the code I put it in (text turns red). I do not know what boxes to check to get the program to accept the code, which includes "Group".


Sorry; Phoenix insists in Formulation, not Treatment. In the setup you have to map Group to Classification. Try:
fixed:  Group+Sequence+Sequence(Group)+Period(Group)+Formulation+Formulation*Group
random: Subject(Sequence*Group)

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
AngusMcLean
★★  

USA,
2014-07-07 00:52
(3552 d 18:59 ago)

@ Helmut
Posting: # 13222
Views: 25,020
 

 Group effect commonly not relevant

Hello Helmut: Yes; I did see that Formulation not Treatment later after lunch, but it still did not work. Then following your direction about Group being a classification variable (and including subject in model) I did get it to work.

Dependent hyothesis     NumDF Denomin     Fstat       p value
Ln(AUC)   int               1   64    11226.673       0
Ln(AUC)   Group             1   64        0.006028186 0.93835554
Ln(AUC)   Sequence          1   64        3.4513852   0.067800018
Ln(AUC)   Group*Sequence    1   64        0.000321727 0.9857451
Ln(AUC)   Group*Period      2   64        0.064946799 0.937179
Ln(AUC)   Formulation       1   64        0.36503556  0.54785867
Ln(AUC)   Formulation*Group 1   64        0.002630007 0.95925927

The formulation by group interaction p value was 0.95 for this composed data set.

Many thanks and also your sample size program in R works well: I have almost zero experience in R, but the reference you provided was easy to follow.


Here is the setup I used in Phoenix: Subject has to be mapped to get it to work (see link below)
http://screencast.com/t/EpJVIvdBr

Angus


Edit: BBCoded and a later post merged. [Helmut]
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-07 01:57
(3552 d 17:53 ago)

@ AngusMcLean
Posting: # 13224
Views: 24,962
 

 Group effect commonly not relevant

Hi Angus,

❝ I did see that Formulation not Treatment later after lunch,


For years I try to convince Pharsight to switch from “Formulation” to the more general “Treatment”.

❝ Then following your direction about Group being a classification variable…


Yep, in the BE-module standard mappings are Subject, Sequence, Period, and Formulation. Any additional Classification variables (aka factors = categorical in­de­pen­dent variables) have to be manually mapped. The same is true for regressors (aka con­ti­nu­ous covariates like body weight).

❝ … (and including subject in model) I did get it to work.


Hhm, I think that Subject should have been mapped automatically. :confused:

❝ Many thanks and also your sample size program in R works well: I have almost zero experience in R, but the reference you provided was easy to follow.


The flowers should go to Detlew. I wrote maybe <1‰ of the code…

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
AngusMcLean
★★  

USA,
2014-07-07 17:04
(3552 d 02:46 ago)

@ Helmut
Posting: # 13226
Views: 24,855
 

 PHX: Parameter mapping

Helmut: Below comment is of interest.

❝ Hhm, I think that Subject should have been mapped automatically


Yes; I thought so too. :confused: I spent quite some time on this but found that I had to check the subject box to map it.

I tried to run the program on a number of occasions and it did not run; it told me to map the subject. So at that point I included in the model by clicking on the orange box. I wonder if there is a sequence of events that must be followed to cause subject to map automatidcally

Angus


Edit: Standard quotes restored and subject line changed. [Helmut]
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-07 17:45
(3552 d 02:06 ago)

@ AngusMcLean
Posting: # 13227
Views: 24,864
 

 PHX: Parameter mapping

Angus,

can you try the following:
  • Create a new Phoenix-project.
  • Navigate to Data > right-click > Import
  • Navigate to {your program files}\Pharsight\Phoenix\application\Examples > Select Data 2x2.csv > Open
  • Keep all default options > Finish and Open
  • Add a new column > Column Name: Group
  • Input 1 for subjects PHST-01 to PHST-10 and 2 for PHST-11 to PHST-20.
  • Right-click Data 2x2 > Send To > NCA and Toolbox > Bioequivalence
  • In my installation Subject is automatically mapped indeed. See below:
    [image]
    Manual mapping is only needed for AUClast > Dependent and Group > Classification.
Can you check?

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
Shuanghe
★★  

Spain,
2014-07-07 19:35
(3552 d 00:16 ago)

@ Helmut
Posting: # 13228
Views: 24,858
 

 PHX: Parameter mapping

Hi Angus, Helmut,

I would suggest to add the usual PK parameters (Cmax/AUCt/AUCinf or whatever you'd usually written in your data set such as logAUCt, ... in Edit>Preferences>Mapping Contexts>Global Contexts and then In ...>Mapping Contexts>Context Associations>Bioequivalence Model mapping them to Dependent so you don't have to manually map them every time you run the BE model.

I would suggest to spend some time to create all PK parameters and depending on the necessity mapped them to different settings such as in BE model, different plots etc. As these are prety routine work, once everything is set, you'll seldom need to adjust them and almost all mappings (be it BE model or different plot model) will be automatically done. :-D

All the best,
Shuanghe
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-07-07 19:54
(3551 d 23:57 ago)

@ Shuanghe
Posting: # 13229
Views: 24,930
 

 Great hint!

Hi Shuanghe,

wow, I didn’t know this one! Now I have automatic mapping… It is not necessary to create all possible variables. I added AUC and in the right window of Global Contexts at Potential Column Names I added to AUC the usual sus­pects (AUClast, AUCinf, AUCtau). Simply great – should have RTFM!

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
AngusMcLean
★★  

USA,
2014-07-07 22:49
(3551 d 21:02 ago)

@ Shuanghe
Posting: # 13233
Views: 24,805
 

 PHX: Parameter mapping

Shuanghe: Thank you for your suggestion. I tried your approach and it worked well for mapping out the variables including the "group". However when you look at the corresponding code for "Fixed effects" and "random effects" in the tabs you will see that it does not have the correct code (from Helmut's original post) so you will run the incorrect model if you do not provide it.

So you have to enter the these effect manually.

Angus.
AngusMcLean
★★  

USA,
2014-07-07 20:46
(3551 d 23:04 ago)

@ Helmut
Posting: # 13230
Views: 24,885
 

 PHX: Parameter mapping

Thank you; I will try it out. First I just verified that presently it does automatically map.....definitely not.


HELMUT: NO mapping whatsoever!

You have to mp it yourself,


Helmut: I was able to repeat the work by reimporting the example file (2x2CSV) data file: yes; it is only necessary to add the dependent variable AUC plus group variable.

However when add the group variable you have to edit the model and this necessitates virtually rewriting the model to include the new terms outlined in your original post. If you just clicked on group for mapping and the model is automatically organized for you is what you would like top happen,

The need to add "GROUP" as classification variable on mapping is the fly in the ointment.


Angus


Edit: Please don’t open new posts if you want to make only a minor correction or want to add something. I merged two of your follow-ups. You can edit your posts for 24 hours (see also the Forum’s Policy). [Helmut]
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
90 visitors (0 registered, 90 guests [including 6 identified bots]).
Forum time: 18:51 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5