Date: Fri, 17 Sep 1999 10:25:00 -0400

From: "Kevin Dykstra" <KDykstra@genetics.com>

Subject: Interoccasion variability

All,

I have data for which PK was measured on multiple occasions for a number of patients. The drug in question obeys simple one-compartment kinetics and is administered as an IV infusion. I would like to obtain estimates of interoccasion variability as well as inter patient variability. In other words, we have

Pij = Phat + Ei + Kij

where Phat is the population typical parameter value, Ei is the interpatient error and Kij is the error for the jth occasion for the ith patient.

It is not clear to me how to code this directly in NONMEM, although I have seen the problem addressed a couple times in the literature. I can imagine a two-stage approach where posthoc parameter estimates for each occasion are obtained and these then subjected to a linear mixed model analysis. Any other suggestions?

Thanks in advance for your help.

Kevin Dykstra

Date: Fri, 17 Sep 1999 11:13:44 -0400 (EDT)

From: Jogarao Gobburu 301-594-5661 FAX 301-480-8329 <GOBBURUJ@cder.fda.gov>

Subject: Re: Interoccasion variability

Kevin,

You can model IOC as follows using NONMEM:

Lets say, you are interested in estimating IOC of CL across three occassions (OCC1,OCC2,OCC3) coded using dummy variables, then

...

$PK

TVCL = ...

CL = TVCL*EXP(ETA(1)+ETA(2)*OCC1+ETA(3)*OCC2+ETA(3)*OCC3)

...

$OMEGA BLOCK(1) 0.09 ; interindividual variability (IIV)

$OMEGA BLOCK(1) 0.05 ; OCC1

$OMEGA BLOCK(1) SAME ; OCC2

$OMEGA BLOCK(1) SAME ; OCC3

You need to use the option SAME for OCC2,OCC3 as you want to preserve the distribution but sample ETAs for each occassion. There is an example in the Help Guide which might be useful.

Regards,

Joga Gobburu

Pharmacometrics,

CDER, FDA

From: "BRUNO, Rene" <Rene.BRUNO@RP-RORER.FR>

Subject: RE: Interoccasion variability

Date: Wed, 22 Sep 1999 10:46:59 +0200

The two-stage approach also works fine, provided you get good individual parameters estimates. We have a couple of exemples where we got very similar estimates using both approaches. It can be used either for a rapid screening of covariates effects (as an alternative to standard regression approaches e.g. MLR, GAM ... when you have multiple occasions) or when you have models that may be too complex to be implemented in NONMEM with multiple occasions and covariate effects e.g. parent + metabolite(s).