Different models for different age groups? Infinite objective functions
From: "James Bailey" <James_Bailey@Emory.org>
Subject: Different models for different age groups? Infinite objective functions
Date: Wed, 18 Jul 2001 11:04:41 -0500
To all:
Two questions.
#1
I am analyzing data from a study of a vasoactive intravenous drug in
pediatric patients. The majority of patients are neonates or infants.
Two previous, limited studies of this drug in pediatric patients used
two or three compartment models. In these earlier studies loading doses
were given, very quickly in the study that found a three compartment
model to be optimal and less quickly in the study using a two
compartment model. In the study I am analyzing, a loading dose was
given very slowly, followed by a constant rate infusion. For this data,
I find that the optimal model is one compartment. But when I analyze
subsets (neonates, infants, or chidren) I find that a two compartment
model works best for children and the one compartment model is optimal
for neonates and infants.
I am not satisfied using different models for somewhat arbitrarily
defined age groups. Does anyone know of a method to develop a unified
model? I have tried
using age as a covariate, with the following CONTROL file
(the model parameters are volumes and clearances)
TVV1=WT*THETA(1)
V1=TVV1*EXP(ETA(1))
TVV2 = WT*THETA(2)*EXP(THETA(5)*(AGE-8))
V2=TVV2*EXP(ETA(2))
TVCL=WT*THETA(3)
CL=TVCL*EXP(ETA(3))
TVQ=WT*THETA(4)
Q=TVQ*EXP(ETA(4))
(8 is the median age, in months, in the study). I have done something
similar for distribution clearance, Q. Any suggestions?
A second question relates to an error message I get frequently when
using conditional estimation. The error message is "Minimization
terminated due to proximity of last iteration estimate to a value at
which the objective function is infinite". I know the NONMEM Users
Manual says estimates from a minimization which is terminated due to
rounding errors may be useful, depending on the number of significant
digits, etc. Is there any useful information fom a minimization which
is terminated for the above reason?
Any help will be appreciated.
Jim Bailey