From: "Stefan Wilde" <stefanwilde@hotmail.com>
Subject: 2-compartment model: V2 doesn't vary
Date: Fri, 28 Sep 2001 19:41:46 +0200

Dear nonmem users,

I'm analysing a dataset in a sparse data situation:
30 subjects, 3 visits per subject, 3 samples per visit (1st: 5-200 min post
infusion, 2nd sample 60 min. later, 3rd sample approx. 16h p.i.)

2 different drugs are studied, both of them known to follow 2-compartment
kinetics.

Analysing these data with NONMEM (NMTran, Advan3, Trans4) gives reasonable
estimates for all kinetic parameters and for variability of all parameters
but V2.
The peripheral Vd doesn't vary beetwen subjects
(Est. of eta = 1EXP-11).
Even when fixing intercompartment clearance V2 doesn't vary.

Estimating 2-compartment kinetics of the 2nd drug given (slightly different
post-infusion times) the results are the same: no variability of V2.
We are quite concerned about these findigs that don't sound reasonable to
us. Any explanations or suggestions of how to solve this problem are
appreciated.

Below you find excerpts of the nonmem control stream and the dataset.

Thanks in advance,

Stefan

Control stream excerpt:

$INPUT ID CY TIME AMT RATE DV MDV EVID
$SUBROUTINE ADVAN=ADVAN3 TRANS=TRANS4

$PK
CALLFL=1
CL = THETA(1)*EXP(ETA(1))
V1 = THETA(2)*EXP(ETA(2))
Q = THETA(3)*EXP(ETA(3))
V2 = THETA(4)*EXP(ETA(4))
S1 = V1/1000

$ERROR
Y = F*(1+ERR(1)) ;Multiplikatives Modell


$THETA
(0,1.3,10000)
(0,30,10000)
(0,.9,10000)
(0,500,10000)

$OMEGA DIAGONAL(4) .01 .1 .1 .01
$SIGMA .8
$ESTIMATION METHOD=COND NOABORT PRINT=5 MAXEVAL=9999 SIGDIGIT=7

Dataset excerpt:
ID Cycle Time Amt Rate DV MDV EVID
3 1 0 50 3.33333 0 1 4
3 1 15 0 0 0 1 0
3 1 130 0 0 8.41 0 0
3 1 345 0 0 6.98 0 0
3 1 1195 0 0 3.88 0 0
3 2 0 50 2.5 0 1 4
3 2 20 0 0 0 1 0
3 2 150 0 0 8.1 0 0
3 2 990 0 0 4.02 0 0
3 3 0 50 0.625 0 1 4
3 3 80 0 0 0 1 0
3 3 95 0 0 12.3 0 0
3 3 405 0 0 6.59 0 0
3 3 1245 0 0 4.59 0 0
9 1 0 48 1.29729 0 1 4
9 1 37 0 0 0 1 0
9 1 41 0 0 222 0 0
9 1 165 0 0 6.89 0 0
9 1 1277 0 0 2.24 0 0

--
Stefan Wilde
Institute of Pharmacology
Clinical Pharmacology
University of Cologne
Germany

 

 

*****

 

 

From: "bvatul" <bvatul@ufl.edu>
Subject: Re: 2-compartment model: V2 doesn't vary
Date: Fri, 28 Sep 2001 15:23:59 -0400

Hello Stefan
A couple of things:
In sparse situation this can happen. You can fix the eta to say about 30%
and run your analysis. Check your parameter estimates (calculate half-life
etc) and see if they make any meaning. Also I was wondering why you changed
the default value of NSIG to 7. Can this make things difficult, I am not
sure! Such a low value of eta might mean you might be either
overparameterising or the data structure might be doing so!! You might want
to try simpler models.

Hope this helps
Atul

 

 

*****

 

 

From: Joern.Loetsch@t-online.de (Joern Loetsch)
Subject: Re: 2-compartment model: V2 doesn't vary
Date: Fri, 28 Sep 2001 21:42:42 +0200

I wouldn't be concerned. The result tells you that the interindividuela
variability in your is sufficiently accounted for when assigning ETAs to the
other structural parameters. An IIV on V2 does not further improve your fit.

Jorn
_______________________________________________________
Jorn Lotsch, MD
pharmazentrum frankfurt, Department of Clinical Pharmacology
Johann Wolfgang Geothe-University Hospital
Theodor-Stern-Kai 7
D-60590 Frankfurt
Germany
Phone: +49-69-6301-4589
Fax: +49-69-6301-7636

 

 

*****

 

 

From: Lewis B Sheiner <lewis@c255.ucsf.edu>
Subject: Re: 2-compartment model: V2 doesn't vary
Date: Mon, 01 Oct 2001 07:02:06 -0700

This has been discussed many times on nmusers - see archive.
Basically, it is impossible to distinguish multiple components
of variance when data are sparse. Maximum Likelihood (not NONMEM)
prefers to drive certain variances to zero if they are unidentifiable.
This is a peculiarity of ML. It does not mean that there is no variability
in V2; it simply means that variability on V1, CL, etc., is adequate to
explain the variability in the data. If you are using FO, it may
be possible to estimate additional variance components by using FOCE, but that
is inevitable.

LBS.