6 replies to this topic

[Thomas Klitgaard]

Hi,

 

In this link, advice is provided wrt. how to adjust for unequal variances when doing BE testing in parallel design.

 

Adjusting for Unequal Variances in WNL Bioeq - Phoenix WNL basics - Certara Forums

 

There are two "parallel testing" examples in the PHX flow, and it is stated that both columns Repeated and Period may be used . However, the two columns are not similar - one has the all 1's (Period) and one has 1-2, suggesting that each subject gets the both treatments. 

 

I can see that results are the same for the approaches, but I dont understand why that is?

 

Secondly, I have used the approach in a study with 3 treatments Ref-Test1-Test2. Test 1 is not of relevance for me, however I get two different results if I analyse  BE of Test 2 vs. Ref on the data with and with Test 1 excluded from the datasheet. How do the data from Test 1 inform the analysis and should it be included or not? THX 

[Linda Hughes]

Hi Thomas,

 

For your first question (why you get the same results if you use, for the Repeated specification, the Repeated column that has all 1's or the Period column that has 1's and 2's), as noted in the referenced post, ideally you wouldn't include the Repeated specification at all, as is done in SAS PROC MIXED, and only the Subject and Group are needed to determine the R matrix in this case.  But since the Phoenix UI requires the Repeated specification in order to specify the Subject and Group, it is necessary to fill in with a column that is not going to affect the model given the structure of the R matrix determined by Subject and Group.  Group in this case is set to Formulation, so when using Subject and Formulation to structure the R matrix, the Period column doesn't add any information to the model, nor does the Repeated column of all 1’s.  You can think of this intuitively as, in the Parallel case, if Formulation is known, the Period is already known.  The Crossover case is less intuitive, and depends on Subject(Sequence) being used as the random model.

 

For your second question, I would need to create a project that shows this.  Or, if possible, you could upload your project here.

 

Regards,

Linda Hughes

Certara

[Linda Hughes]

Hi Thomas,

 

A colleague of mine provided a link where there is a discussion of your second question on the BEBAC forum:

https://forum.bebac....try.php?id=1082

 

Regards,

Linda Hughes

[Linda Hughes]

Hi Thomas,

 

Thanks for sending the project to Simon.  Where the Test1 data is included in the dataset, the Test1 data is still used in the Fixed Effects and Variance Structure models, so it affects the 'Diff_DF' parameter, which is then used to compute the confidence intervals.

 

In the link I posted above, it is noted that the EMA recommends:

In studies with more than two treatment arms, ... the analysis for each comparison should be conducted excluding the data from the treatments that are not relevant for the comparison in question.

[Thomas Klitgaard]

Hi Linda, Simon,

 

Thank you so much - this was very helpful. 

 

BR Thomas

[Helmut Schütz]

Hi Thomas,

 

maybe some background information helps,

 

[Thomas Klitgaard]

Hi Helmut,

This is very informative- thank you! Br Thomas