Back to the Top
Inter-occasion variability can easily be modeled in NONMEM.
Could anybody tell me how IOV can be coded in WinNonMix?
Back to the Top
The following message was posted to: PharmPK
As far as I know, IOV is not built-in in WinNonMix. I will be very
interested to know how to implement it in WNM too.
Sam Liao, Ph.D.
Address: 20 second street,
Jersey City, NJ 07302
Back to the Top
Concerning interoccasional variability (IOV). This is a very
useful idea. However. it is useful in addition to use a method of this
type to track the behavior of a drug in acutely ill, highly unstable
patients, from day to day and from dose to dose, and to estimate the
most probable Bayesian posterior state of the patient after the most
recent serum concentration measurement, for example.
Using nonparametric population PK/PD models, one can use
sequential Bayesian modeling. However, if one uses the conventional
maximum aposteriori probability (MAP) Bayesian technique, in which the
Bayesian posterior after the most recent measurement becomes the prior
for the next measurement, one will perceive a set of changes in the
patent's posteriors throughout the sequence of serum measurements.
However, the problem with this approach is that after the last
measurement, one winds up with the same posterior that one would obtain
if one had fitted all the data in a batch mode.
The reason for this is that the MAP Bayesian fitting procedure
is always estimating the parameter values that best fit all the
available data. Because of this, even if the patient has been having
changing parameter values during the data set, they will not be
perceived by this approach. It has the built-in hypothesis that there
is one and only one set of parameter values that best fit the data.
There is no opportunity for the model to change during the fitting of
the data set.
Another approach does better. It is the interacting multiple
model (IMM) sequential Bayesian approach. It comes from the aerospace
community which is interested in tracking hostile targets which are
taking evasive action. If you fit all the data as in the MAP Bayesian
approach above, the most recent evasive maneuver may well be missed by
the preponderance of all the other data. This IMM approach is based on
multiple models of the system, based here on the nonparametric (NP)
population PK/PD models, in contrast to parametric models, which have
only means and variances for the parameters. The IMM approach is not
limited by a fixed parameter distribution (a fixed set of parameter
values and their fixed probabilities. One can specify a certain
probability of a random change in the "true patient" from one support
point to another whenever a new dose is given, for example, or a new
serum concentration measurement is obtained. If it is more likely that
the "true patient" may change from one to another, based on the most
recent data, the "true patient" will change from one model in the set
of population parameter support points to another.
The NP population modeling approach, upon which the IMM
approach is based, estimates the entire most likely distribution of
parameter values in the population. In addition, since it is a
mathematically consistent method, its estimates of parameter means and
variances are usually better that those obtained by current parametric
modeling methods which use the FOCE approximation of the likelihood,
which is not consistent. For more information about this, go to our web
site www.lapk.org, and click on New Advances in PK/PD Modeling, to get
the paper presented by Bob Leary at the Population Analysis Group in
Europe meeting in Paris last June.
Using the IMM approach, the behavior of a drug in a carefully
simulated changing patient (based on real data) is tracked with about
half the total error seen with the MAP Bayesian approach, or with the
multiple model Bayesian approach, when both carry the built in
hypothesis that there is only 1 set of parameter values (or
distributions) that best fit the data.
For more information, go to our web site www.lapk.org, go to
teaching topics, and click on material in section 9, for the IMM, and
section 7, for the NP population modeling approaches. We think this IMM
approach is a good way to take clinical advantage of information about
IOV directly in improving the precision of drug dosing regimens for
optimal patient care. Our new MM-USCPACK clinical software uses
multiple model (MM) dosage design to achieve desired target goals
specifically with maximum precision, and either MM or IMM, as desired,
to get the Bayesian posteriors for individual patients (other ways of
using Bayes' theorem), and once again, the maximally precise dosage
regimen for the future. This clinical, Windows-based software is now in
beta phase. If you would like to try it, please let us know.
All the best holiday wishes,
Roger W. Jelliffe, M.D. Professor of Medicine,
Division of Geriatric Medicine,
Laboratory of Applied Pharmacokinetics,
USC Keck School of Medicine
2250 Alcazar St, Los Angeles CA 90033, USA
Our web site= http://www.lapk.org
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (firstname.lastname@example.org)