From: "HUTMACHER, MATTHEW [Non-Pharmacia/1825]" <matthew.hutmacher@pharmacia.com>
Subject: Increasing the umber of sub-population in a mixture model
Date: Fri, 25 May 2001 13:07:28 -0500
Hello,
I have a very complicated data pattern. Many of you have given great advice on how to use a mixture model at the IOV stage and I appreciate it. I have found, that I need more sub-populations to characterize the lag-time distribution in my model (the distribution of me predicted etas are still very skewed). I have tried to increase the number of sub-populations and I get this error message:
AN ERROR WAS FOUND IN THE CONTROL STATEMENTS.
AN ERROR WAS FOUND ON LINE 43 AT THE APPROXIMATE POSITION NOTED:
P(5)=THETA(.)
X
THE CHARACTERS IN ERROR ARE: 5
332 $MIX: NSPOP AND INDICES I OF P(I) MAY ONLY BE 1, 2, 3, OR 4.
STOP 4 statement executed
Is there a way to increase the number of allowable sub-populations.
I have extended the code Dr Sheiner supplied to me for two sub-populations (at IOV level) to three:
$MIX
P(1)=THETA(.)
P(2)=1.-THETA(.)
P(3)=THETA(.)
P(4)=1.-THETA(.)
P(5)=THETA(.)
P(6)=1.-THETA(.)
NSPOP=6
Any help in this would be appreciated.
Thanks again.
Matt
Increasing the umber of sub-population in a mixture model
From: "David Nix, Pharm D." <nix@pharmacy.arizona.edu>
Subject: Re: Increasing the umber of sub-population in a mixture model
Date: Fri, 25 May 2001 13:26:18 -0700
I'm having the same problem with a data set involving administration of the same drug under four conditions to a group of 16 volunteers. The lag time ranges from 0 to 6 hours. Variability of lag time is high within subjects and is not explained by any covariates or by conditions of treatment. The lag time also does not appear to follow any particular statistical distribution.
My most recent idea was to provide a fixed and variable component to the lag time. I entered a variable LAG into the data set. LAG was defined as the last time for zero concentration in the inverval between the dose time and the time of first detected concentration minus 15 min. I substracted 15 min since this will allow some room to predict the Tlag without forcing. Values less than 0 were set equal to 0.
IF(LAG.EQ.0) ALAG1=0
IF(LAG.GT.0) ALAG1=LAG + THETA(.) * EXP(ETA(.))
I'm currently refining this approach. 1. whether ALAG1 should be fixed at 0 if there is no apparent lag or whether I need allow some variation. 2. whether the distribution for the the THETA portion is additive, proportional or exponential. The exponential error has worked best so far.