Back to the Top
Hi Matt
Some time ago (1994! - time flies!) I did some work with Nancy Sambol and
Stuart Beal on the interaction between the structural (one vs. two
compartment), statistical (variability) and the covariate model in a
situation very similar to what you describe. I found that if I fit a one
compartment model to sparse data simulated from a two compartment model then
I got artifactual covariate effects and a more complex statistical model in
the one compartment model (JPB, Vol 22 p165-177). I also tried a couple of
real data examples where the data were so sparse as to not support using the
a priori known two compartment model and again found covariates in the one
compartment model that were not included when a two compartment model was
fit to the data. The situation with real data is hard to pin down to
spurious effects since we never really know the truth. In my case the
covariate effects in the one comp model did seem reasonable. One of our
conclusions was that any covariate effect that is to be included in the
model should be biologically plausible. This would seem to be good advice
in your situation since including creatinine clearance for a drug that is
eliminated metabolically does not seem to be biologically plausible (unless
that covariate is a surrogate for something else, age for example). I would
suggest that you rerun your model using a two compartment structural model
(fix some parameter values to a priori values if necessary) and retest to
see if the covariates you found with the one compartment model are still
significant. A last thing to bear in mind is to look at the size of the
effect that your covariates have in the one compartment situation, if the
effect is not clinically relevant then maybe you don't want to retain those
covariates in the model anyway (but that does depend on what you want to do
with the results of your analysis).
I hope this is of some help,
Janet Wade
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)