Bioequivalence and Bioavailability Forum • Specifying ANOVA model with covariates (WinNonlin)

PKPDFX
●

2006-10-05 09:20
(6786 d 08:05 ago)

Posting: # 274
Views: 9,641

Specifying ANOVA model with covariates (WinNonlin) [Software]

Greetings,

I am using the BEQ wizard in WNL v4.1. I am analyzing data from a crossover design study of two formulations that used both male and female subjects.

I used the default setting for the crossover study, average beq analysis, but I want to include both sex and body weight in the ANOVA model as covariates and cannot figure out how to specify their inclusion in the model.

I tried subject+period+sequence+sex+bw as the model, but got the same results as when I leave out the covariates. I know it is something simple, but I dont work with ANOVAs very often and do not know how to formulate the correct model. Can anyone help?

Helmut
★★★

Vienna, Austria,
2006-10-05 14:51
(6786 d 02:34 ago)

@ PKPDFX
Posting: # 276
Views: 8,065

Pharsight = Far Side?

Post reply

Hi PKPDFX, nice nick ;-)

OK, just to stay inline with WinNonlin’s nomenclature:

Classification variables or factors are categorical independent variables, e.g., formulation, treatment, and gender.
Regressor variables or covariates are continuous independent variables, e.g., temperature or body weight.

IMHO your model specification ‘subject+period+sequence+sex+bw’ is correct.
I just reproduced your results with both WinNonlin 4.1 and the current version, 5.0.1: no difference in results. :-(

I remember that I’ve tried to include covariates in BE a couple of years ago and failed as well.
I put it on my ‘to-do-list’, and left it there until today…

Sorry, no better news, consider contacting Pharsight’s support—and post the answer to the forum (I’m curious as well).

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

Helmut
★★★

Vienna, Austria,
2007-10-22 18:28
(6403 d 22:57 ago)

@ Helmut
Posting: # 1239
Views: 7,641

Solution from Pharsight's support

Post reply

Dear group members!

❝ I remember that I’ve tried to include covariates in BE a couple of years ago and failed as well.

❝ I put it on my ‘to-do-list’, and left it there until today…

❝

❝ Sorry, no better news, consider contacting Pharsight’s support—and post the answer to the forum (I’m curious as well).

After more than a year I finally satisfied my curiosity and requested a solution from Pharsight.
Starting with WinNonlin’s data-set ‘Data22.PWO’ in the ‘Examples’-folder, I added a variable ‘Gender’ which I assigned ‘male’ to the first 10 subjects and ‘female’ to the next 10 subjects.

Pharsight’s support suggested a model with fixed effects for ‘Gender’ and a ‘Treatment by Gender Interaction’.

Normal BE-assessment, reference='Capsule' (residual DF) gets for
Fixed effects: Sequence+Formulation+Period
Random1: Subject(Sequence)
Ratio 104.6588, 90% CI (92.2133, 118.7840)
Covariate included
Fixed effects: Sequence+Formulation+Period+Gender+Formulation*Gender
Random1: Subject(Sequence)
Ratio 104.6588, 90% CI (92.7438, 118.1045)
Partial tests for Gender p 0.3303, Formulation × Gender p 0.1082.

No large differences are to be expected even for drugs with large gender differences in BA, because BE is a within subject comparison.
A similar model may be used if the study was performed in two or more groups (e.g., for logistic reasons). Just use 'Group' instead of 'Gender' in the model. If the 'Treatment × Group Interaction' term is statistically not significant (p<0.05), the simple model (i.e., naive pooling of data) can be applied.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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