Back to the Top
Hello,
I am working on writing a transduction cascade model (indirect model 1
with 1/tau to account for the lag using 3 differentials) for
pharmacodynamic parameters that were used to measure sedation and
analgesia following a single IV bolus of an alpha-2 adrenergic
receptor agonist. Although I can get the model to fit the data set
well and my WSSR, AIC and SBC are all in acceptable range, my CV%
range from single digits to 1000 depending on the data set and PD
parameter I am looking at (e.g. heart rate) for estimated parameters
such as Kin, Kout, IC50, etc. I am new to writing models and I was
wondering if someone might have advice on what I can do to try and
minimize my CV%?
Thanks,
Kristin Grimsrud
Department of Pharmacology and Toxicology
School of Veterinary Medicine
University of California, Davis
Back to the Top
Hi Kristin,
There are a lot of factors that could lead to inflated %CV, for
example: you could have identifiability issues with one of the
parameters, thus it would be poorly estimated and your %CV will be
large. So the question becomes do the observations uniquely determine
the parameters?
Another reason is experimental design (no. of points per subject and
time of sampling) which could lead to model distinguishability. So,
maybe your model could be simplified or you could fix one of the
parameters to a value from the literature or you could use a prior for
some of the parameters if you doing Bayesian/ population analysis
Hopefully this would help,
Best,
Nidal
Nidal AL-Huniti, PhD
Associate Director, Modeling and Simulations
ICON Development Solutions SM
Back to the Top
The following message was posted to: PharmPK
I am looking forward to seeing answers to Kristin's questions. This
isn't
my particular field, but I'm interested to see the role that biological
variability, parameter observability (for this particular model and
data)
and the particular choice of CV% have in the answer people provide.
Other ideas come to mind as well, e.g. the possibility that the Kin,
Kout
and IC50 may have distributions that are better behaved in logarithmic
space, and this can have impact on measures like CV%. These are all
very
general ideas, that do not depend on her particular model, modeling
approach, etc.
Again, I'm looking forward to seeing the answers. I always learn
something
from this group.
G. Scott Lett, Ph.D.
The BioAnalytics Group LLC
241 Forsgate Drive, Suite 209
Jamesburg, NJ 08831
slett.-a-.bioanalyticsgroup.com
Back to the Top
Hi,Kristin Grimsrud
I just doubt that when apply K-PD model, is there also some hypothesis
like one-compartment or two- compartment model, for example, if the
patients has system infection or inflammatory.
And hope the below paper could be a little helpful when you
condisering some details.
http://www.springerlink.com/content/5j13157243439820/fulltext.pdf
[not free -db]
Modelling Response Time Profiles in the Absence of
Drug Concentrations: Definition and Performance
Evaluation of the K-PD Model
P. Jacqmin,
1,6 E. Snoeck,1 E.A. van Schaick,1 R. Gieschke,2 P. Pillai,3
J.-L. Steimer,
3 and P. Girard4,5
Received April 21, 2006--Final August 23, 2006--Published Online
October 19, 2006
Currently, I am also try to establish a K-PD model of a anti-
inflammatory drug. And I am still on the process of collecting PD data
at different time points.I know little about this topic, but still try
to learn some knowledge from this disscussion group.
Thanks all
Back to the Top
Hello All,
Since my question has generated some good discussion (and I am glad it
has), let me expand on what I am working with since I feel I was a bit
to vague in the last note. I focus on veterinary pharmacology. I
administered a single IV bolus to 8 horses and collected blood and PD
data which included heart rates, respiratory rates, gastric motility,
parameters to represent sedation, analgesia and ataxia, blood for drug
concentrations, pack cell volumes, blood glucose since my drug of
interest inhibits insulin release, as well as some other. This drug is
rapidly cleared with a half life of less than 30m so we have about 20
We so excessive sampling with about 20 just in the first 6 hours and
12 of them being within the first hour for drug conc and about 7
within the first hour for PD data. So I have run the PK and generated
my constants, however what is interesting is that 2 animals clearly
present as a one compartment and the other 6 as a 2. Although one
would generally choose a single model that best characterizes the
drugs behavior with the emphasis of choosing the simpler model, in
this situation I am going to present both compartments in the
manuscript because I feel that it needs to be discussed.
For the issues with the transduction model, for parameters such as
glucose, heart rate, etc. not only am I dealing with a large degree of
individual variation (which is not so much of a problem since I am
modeling each horse individually with its own kinetics being used) but
there is a great deal of intra-individual variation throughout the
day. So my data sets are choppy and when my baseline heart rate starts
off between 32-40 and I am dropping it down to 22-28 following admin,
it makes it hard to model. The horse tends to be a much larger
challenge with many of these parameters compared to lab animals or
humans who have much larger values. However, I can take a lot of
samples often (I have 40L of blood to work with) and the PD data as
well. So I don't think my issue with the CV% is a lack of samples, but
possibly the timing.
My thought following some of this discussion is to work with this data
similar to that of PopnPK. I hate to pool data but as a last result it
provides an alternative. If I pull the data and model all of it
together, and use the averages from the PK data, then this may provide
some insight to whether my issue is not enough points. Since I am
dealing with choppy data right now, having more data for the model to
work with my allow it to decrease the CV%. If it doesn't work then at
least I can have some idea that maybe my problem is not related to the
number of data points. Any thoughts on this approach???
Kristin Grimsrud
Equine Analytical Chemistry Lab
Department of Pharmacology/Toxicology
School of Veterinary Medicine
University of California Davis
Back to the Top
Kristin,
Good project. I wonder if some of your "choppy" data is because your
drug is not the primary effector of the physiological variable you are
monitoring. Feeding time, circadian rhythms, external stimuli, etc may
be more important particularly at your later time points. As you are
at Davis have you asked Art Craigmill for assistance, this type of
physiological and popPK modeling is right up his alley.
Cindy Cole
Novartis Animal Health
Want to post a follow-up message on this topic?
If this link does not work with your browser send a follow-up message to PharmPK@boomer.org with "Modeling transduction cascades" as the subject | Support PharmPK by using the |
Copyright 1995-2011 David W. A. Bourne (david@boomer.org)