6 replies to this topic

[Julie Dumond]

Hi,

 

I am trying to use the bioequivalence wizard to generate GMRs/90% CI for a study with 4 PK assessments over the course of pregnancy. Dose was increased at the 3rd PK, but otherwise the treatments are the same and obviously there was no randomization/cross-over and the sequence is the same for everyone. I can make it work with the parallel design, but that's not quite right since the AUCs are not independent. Is there a way to change the set up in the cross-over set-up to allow me to do this with the same sequence and treatment for all subjects, but also account for the fact that the same subjects were sampled repeatedly?

 

Thanks,

 

Julie

[Helmut Schütz]

Dear Julie!

OK, you have a paired design. First you have to code your dataset in such a way, that the treatments are coded differently (I don't know your intentions, but one of the treatments must act as the reference and the others as different test treatments). In the BE setup. 'Model tab' select Type of study 'Parallel/Other', select your 'Reference Formulation'. Map 'subject' as a classification variable. In the 'Fixed Effects tab' add Subject to the Model Specification (should read 'Treatment+Subject'). Execute the workflow. Caution: it may be necessary to adjust your alpha-level if you compare three tests to one reference. Bonferroni would do the job - test at alpha 0.05/3 (otherwise the patients' risk will be inflated to 1-(1-alpha)³=0.1426. You can adjust the alpha-level in the "Options tab': instead of 90% use 0.967).
You find the adjusted CI at CI_User_Lower and CI_User_Upper.

@The developers. That's a flaw: It should be possible to paste values in full numeric precision. The correct value would be 0.96 repeating: 1 - 2 × 0.05/3 = 0.966666666666...

[Ana Henry]

A note to let you know that the defect reported above has been fixed in the next release of PHX WNL 6.2 (to be released at the end of this year) - the tracking number QC 9381. This was fixed in relation to a previous forum posting.

 

QC 9381: In Phoenix WNL 6.1, the "Percent of Reference to Detect" and "Confidence Level" fields on the Bioequivalence "Options" tab only allowed for 5 characters total, e.g., 20.55. This has been fixed for the PHX WNL 6.2 so that these fields will accept much more accuracy

 

 

Ana Henry

Pharsight

[Helmut Schütz]

Hi Ana,

 

great to hear. Nice!

[Emily Colby]

Dear Julie,

 

My name is Emily Colby, I work for Pharsight and my background is in biostatistics, so they assigned me to be the moderator for your post. Sorry for the late reply -- I have been on the road teaching courses on Phoenix. I'm out of town again next week, so if you reply to this message, I might not be able to respond until the week of Oct 11.

 

The study you described does not sound like a typical bioequivalence study. It seems to me that it would be best to model the data using a longitudinal linear mixed effects model, which can be done using the Linear Mixed Effects object found in "NCA and Toolbox". In fact, the "bioequivalence" model is a speical case of the linear mixed effects model, so it is possible to get the same results. However, it's important to think it through rather than relying on a standard tool in this case.

 

I can't say much more than that because I don't know the details of your dataset. If there is a statistician where you work, I suggest collaborating with them to determine how best to analyze the data.

 

Best regards,

Emily

[Emily Colby]

Julie,

 

I just noticed that you have a UNC email address. I suggest collaborating with the Biometrics Consulting Lab (where I used to work) in the Biostatistics Department in the School of Public Health. If you email bcl@unc.edu, you can schedule an appointment to meet with them. Usually, they ask for you to provide a list of research questions that you're trying to answer, along with the dataset, and several dates and times that you're available to meet with them.

 

Best regards,

Emily

[Zancong Shen]

I noticed if you don't add "subject" in the model specification but keep it in the classification, on next page, add subject to "random effect model" it still gives same correct results. However this approach sometimes generate "negative error" message, don't know why that happens.

I am using WNL 5.2 bioequivalence wizard and assume it works principally same as PHX.

PKbeginner