From: "Liu, Qi"
Subject: [NMusers] Sample size requirement for POP PK analysis to identify drug interactions
Date: Thu, 23 Feb 2006 15:55:14 -0500

Dear NONMEM users,

I have a question regarding the sample size requirement for the application of POP PK analysis to
identify possible drug-drug interactions. Very often, people use some cutoff number or
percentage to decide whether we need to explore a concomitant drug (or a group of drugs)
as a potential covariate. For example, they might specify in their data analysis plan:” for
any specific drug interaction, a minimum of 20 patients (or 10% of the total population in
this trial) on the concomitant medication in question will be required for the analysis “. It
is important to have this kind of cutoff, particularly to avoid false negatives (inadequate power)
and to mitigate the impact of possible bias (lack of randomization). It is also a matter of
cost-benefit, since the analysis time increases almost exponentially with the number of
covariates to explore. It doesn’t make sense to waste time on some comedication if there
isn’t enough patients taking it to support a reliable conclusion. The question is – how do
we decide on this cutoff? I can imagine extensive simulations can give us some information
on this, but the answer will vary from one case to another and it doesn’t seem very practical
in the industry environment due to the usually aggressive timelines. Is there a general rule
of thumb? I would really appreciate if other NONMEM users could share your experience. Also,
it will be very beneficial if FDA and other regulatory agencies could share their view on this.
Particularly, for the claiming of “lack of interaction”,  how do the agencies decide whether
there is sufficient number of patients taking the comedication in the trial to support the claim?
I noticed that some agency will also see the confidence interval associated with the lack of
effect, but again, how should we decide whether the confidence interval for the lack of effect
is acceptable or not?

Thanks very much for your help,


Qi Liu, Ph.D.
Merck & Co., Inc.
P.O. Box 4
West Point PA 19486
Tel 215 652 4096
Fax 215 993 1265

From: "Stephen Duffull"
Subject: RE: [NMusers] Sample size requirement for POP PK analysis to identify drug inte ractions
Date: Sat, 25 Feb 2006 07:57:18 +1000

Hi Qi

Accurate identification of covariates is not a trivial task in any setting.
Determining sample size should be considered on a case by case setting.  For
general consideration you could have a look at Ribbing and Jonsson JPKPD
2004;31:109-134.  Essentially the size of the covariate effect, the
distribution of the covariate across the study population, the complexity of
the structural model (both PK and covariate) etc will influence findings.
One option, as you suggest, is to perform simulations to assess the design
for estimation of covariate effects.  However, this process can be a little
slow and requires the user to select designs.  You could try assessing the
design in an information theoretic approach (e.g. by assessing the
population information matrix) and choosing a design that maximises your
ability to identify your covariate effect (should one exist).  In the latter
setting there are various software that can be used to do this, e.g. POPED
(which allows you to have a prior on your parameters), PFIM/PFIM_OPT (splus
version) or POPT (Matlab version).

Good luck.

Stephen Duffull
School of Pharmacy, University of Queensland, Brisbane 4072, Australia
Tel +61 7 3365 8808, Fax +61 7 3365 1688, Email:
University Provider Number: 00025B