topic Re: How to simulate a replicate design bioequivalence study in SAS/IML Software and Matrix Computations

How to simulate a replicate design bioequivalence study

jacksonan123 — Wed, 17 Jun 2015 07:54:18 GMT

I have data from a two-way cross over Bioequivalence study for a test product and a reference product and I fit the data to obtain the individual subject parameters. I then can use SAS PROC CORR to obtain the variance/covariance matrix which would allow me to do a simulation by changing the test parameter for rate constant of absorption and proceed to simulate a two-way crossover Bioequivalence study. I can then take the Cholesky of the var/cov matrix using SAS IML to get the lower triangular matrix to conduct simulations usinf SAS IML.

I wanted to know is there any way to introduce random error into the parameters of the two-way crossover study to have those parameters be similar to what one might see if the two-way crossover study were replicated?

Two-way Design

Seq1 AB

Seq1 BA

Replicate Design

Seq1 A B A B

Seq2 B A B A

Re: How to simulate a replicate design bioequivalence study

Rick_SAS — Wed, 17 Jun 2015 13:38:43 GMT

Yes. However, I am not a biostatistician and I do not understand your jargon. Can you explain in simple English what you are trying to do and include some data? Are the data multivariate normal? Or are they multinomial (correlated categorical)? If so, the RANDNORMAL function and the RANDMULTINOMIAL functions might help you.

Re: How to simulate a replicate design bioequivalence study

JE_HamerMa — Mon, 22 Mar 2021 19:20:10 GMT

I have tried using proc simnormal where you use the covariance matrix from proc corr. The only issue is that you will need to know the what the correlation is for reference 1 to reference 2 and test 1 to test 2 which can not be obtained from your 2x2 trial.

Re: How to simulate a replicate design bioequivalence study

Rick_SAS — Sat, 10 Apr 2021 09:51:14 GMT

If you post sample data and code you are using, we might be able to help.

> you will need to know what the correlation is for reference 1 to reference 2 and

> test 1 to test 2 which can not be obtained from your 2x2 trial.

In simulation studies, the parameters for the simulation have to come from somewhere. In clinical trials, they sometimes come from a small pilot study that estimates the parameters. Another option is to use parameters from other similar studies. A third option is to use an expert's opinion as the best guess.

You can also choose a discrete set of parameter values and run a family of simulations. For example, you can choose R1 = {0.2, 0.4, 0.6} as possible values for the correlation of "reference 1 to reference 2" and R2 = {0.0, 0.1, 0.2} as possible values for the correlation of "test1 to test 2". You can then run the 3x3=9 simulation studies. If you have some way to compare the simulation studies to reality, you might be able to decide which one matches reality the best.