From: musor000@optonline.net
Subject: [NMusers] What are the best studies to combine?
Date: Thu, 30 Mar 2006 14:33:26 -0500
Hello NONMEM users,
We try to combine 2 studies of healthy volunteers (single IV and SC dose) and 1 study of
subjects who have a disease (multiple SC doses; concentrations are measured before dosing).
NONMEM code is running well. Will such diverse studies lead us to unbiased conclusions? Also,
will estimates of Ka from the study of diseased subjects be reliable assuming that concentrations
are measured before dosing?
Thank you!
Pavel
What are the best studies to combine?
From: Mark Sale - Next Level Solutions mark@nextlevelsolns.com
Subject: RE: [NMusers] What are the best studies to combine?
Date: Thu, 30 Mar 2006 13:22:18 -0700
IMHO, this is a perfectly fine combination, with (of course) a few caveots. First, one must
always have a set of objectives in mind when selecting data and a modeling approach. If you are
interested in defining differences between HV and patients, or IV and SC then clearly these data
will be helpful. If you have no interest in the SC absorption, and you have enough data (enough
individuals, rich enough) in the HV IV study, then don't include the SC etc, etc. But, it sounds
like your interest might include KA for SC administration, and if you have inadequate data in the
patient population alone, then might consider adding the HV with SC. If you have inadequate data
to define the distribution/elimination in the patient data and HV SC data, you might consider adding
the IV (which will likely better define the distribution, possibly the elimination).
Caveot #2, you MUST ALWAYS, at least consider that these different populations are different in
important ways. Might be safe (please note might) to assume that the 2 HV populations are the same
(especially if the entry criteria/demographics, formulation, assay method etc) are the same. But,
I'd suggest you are obligated to test whethe the patient population is different, in pretty much
every parameter (clearance, KA, lag time, volume - maybe even K12/K21). Probably should consider
even whether the structural model (# of compartments) is different - although if found to be, it may
be just as likely that sampling scheme difference results in the appearance of structural differences.
Finally, you should also consider whether the interindividual or (especially) residual error is different.
So, yes, by all means combine data if the objectives of the analysis warrant it. But, consider all the
possible (biologically plausible) ways that the different populations might be different - and test the
ones you can test.
Not sure what you mean by estimate of KA, given observation prior to dosing - is this an endogenous
compound? But, in general, yes it should be OK.
Mark Sale MD
Next Level Solutions, LLC
www.NextLevelSolns.com
From: "Abutarif, Malaz" malaz.abutarif@spcorp.com
Subject: RE: [NMusers] What are the best studies to combine?
Date: Thu, 30 Mar 2006 15:56:48 -0500
This is my take and I hope we'll get comments from others:
To test your question, I would look at the following:
1. Residual plots separating (or different colors for) the data points from the different studies (i.e., one
color for data points in study 1, another for study 2, ..etc). This way, you'll get a "rough idea" of any
inter-study differences or if one study is on one side of the plot while the other is on the other side.
2. If you feel you have enough data, you may add the study number as a covariate
3. Look at time point distribution (time post last dose) from the different studies and try to see which
study may "skew" the estimates of certain PK parameters (e.g., ka). Test for the importance of these parameters
using sensitivity analysis (again, the usefulness of this may depend on the type of data you have).
And before all that, look at the clinical relevance. See if it makes sense to have a different Ka for patients
and Healthy volunteers, if the exposures at comparable time points and doses are similar between studies (or
in the same ballpark), if there is a reason to believe that half life may be different, ...etc.
And the most important question in my mind is: Why are you combining the data (for what purpose?) i.e., are
you combining the data to have a bigger data set for covariate analysis or are you combining the data to have
better confidence in the PK parameters for the patient study to simulate/estimate individual exposures for
patients in that study? ... Depending on what the purpose of the modeling exercise is, it may or may not matter
how good/bad the outcome of the points above is.
This is how I personally think about the analysis when combining several data sets. I hope to hear from
others how they approach this issue.
Malaz
From: musor000@optonline.net
Subject: Re: RE: [NMusers] What are the best studies to combine?
Date: Thu, 30 Mar 2006 18:24:59 -0500
Hello Malaz and Mark,
Thank you for your responses. The primary objective is to look how demographic covariates affect
PK parameters. The problem/question is: Will such diverse studies provide reliable information on
covariates (gender, weight, and age)?
If SC study of diseased subjects provide reliable estimates of Ka (sparse data from the phase 3 study,
measures are taken before dosing), then I can estimate it for each subject. Otherwhise, I have to
assume that Ka is the same for all subjects within this study. Ka is not the most important parameter.
Probably, it is reasonable to do screening to find potential predictors. Then it is reasonable to add
potential covariates and study number to the population PK model.
I use different errors for each study. Otherwhise, the model does not converge well.
Thank you,
Pavel
From: Mark Sale - Next Level Solutions mark@nextlevelsolns.com
Subject: RE: RE: [NMusers] What are the best studies to combine?
Date: Thu, 30 Mar 2006 17:50:38 -0700
If the demographics/covariates are different between the groups, then you can't be sure if you have a study
effect of a covariate effect. Imagine if you have one study of all males, with CL = 1 l/min and another with
all females, CL = 2 l/min. Is this a study effect or a gender effect? No way to know.
Mark Sale MD
Next Level Solutions, LLC
www.NextLevelSolns.com
From: musor000@optonline.net
Subject: Re: RE: RE: [NMusers] What are the best studies to combine?
Date: Fri, 31 Mar 2006 07:12:38 -0500
Hello Mark,
I agree. Healthy vlunteers and diseased subjects have different demohraphics. Demographic parameters overlap,
but the differences are significant. Thus, we are in grey area. We may not be able to separate study effect
and effect of demographic variables completely. This should be an issue every time someone pools the data.
I assume there are few ways to deal with this:
1. Try to find similar studies;
2. Pool a lot of studies;
3. Compare pooled results and results by population (healthy vs diseased);
Thank you,
Pavel
From: Mark Sale - Next Level Solutions mark@nextlevelsolns.com
Subject: RE: RE: RE: [NMusers] What are the best studies to combine?
Date: Fri, 31 Mar 2006 05:27:37 -0700
In theory, one could put in both effect (demographic effect and study effect). Whether you can seperate
these two correlatated effects depends on how much overlap there is between the groups. In my experience,
usually there is considerable overlap (especially in weight and age), allowing you to identify, or at least
investigate, both effects.
Mark
Mark Sale MD
Next Level Solutions, LLC
www.NextLevelSolns.com
_______________________________________________________