From: Sandra.Visser@astrazeneca.com
Subject: [NMusers] no. of dummy points affect simulation and estimation results
Date: Mon., April 20, 2004 09:10
Dear Nonmem users,
I try to fit a set of 5 differential equations in the ADVAN6 or ADVAN9
subroutine to a dataset with observations every 20 min for two days. I
initialize each PD compartment via AMT=X with MDV=1 for each compartment at
time 0. It appeared that the time to steady state in the set of equations is
appr. 2 days. Therefore, I have added dummy time points (DV=0, MDV=1) for >2
days in order in order to allow sufficient time to steady state.
However, it appears to be that both simulation and estimation results depend
on the number and the particular placement of the dummy time points. Adding
dummy time points each 20 min for several days seems to give the resonable
results but calculation time rises substantially, which I would like to
avoid.
Does anyone know how to select the optimum number and times of the dummy
points? And could anyone give me some information on the integration
routines that are used in the ADVAN6, 8 and 9 subroutines and the selection
of the stepsize.
I would appreciate any ideas or suggestions on this matter.
Best regards,
Sandra Visser
no. of dummy points affect simulation and estimation results
From: Nick Holford n.holford@auckland.ac.nz
Subject: RE:[NMusers] no. of dummy points affect simulation and estimation results
Date: Tue, April 20, 2004 5:10 am
Sandra,
The inclusion of additional event records in the data file should not influence the
solution in theory. However, in practice the DE solver may be sensitive to the
requested output times. This is particularly likely if you have discontinuous DEs
e.g. if you are modelling some process with a conditional expression in $DES. There
are some special discontinuities (lag time and infusion duration) that the PREDPP
library 'knows about' and can usually solve.
I would suggest that you test your model by simulating it with a totally different
piece of software e.g. WinNONlin, ModelMaker, Stella, Trial Simulator, etc. Make
sure that your NONMEM simulated output is similar to the output of another program
so that you can rule out coding errors.
As far as I know NONMEM uses some kind of Gear method for ADVAN8. I cannot easily
find any information on how ADVAN6 or ADVAN9 solve DEs. The stepsize is chosen
automatically. All modern DE solvers do this. Only ancient methods let the user
specify a stepsize.
You might try increasing the value of TOL e.g. TOL=6 or higher to see if this helps
to stabilize the solution by increasing the precision of the local error in the
solutions.
Nick
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
email:n.holford@auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
From: GIRARD PASCAL PASCAL.GIRARD@adm.univ-lyon1.fr
Subject: RE: [NMusers] no. of dummy points affect simulation and estimation results
Date: Tue, April 20, 2004 5:37 am
Sandra,
Here are 2 more efficient ways to initialize your compartment, which avoid defining
dummy times:
1) either put a steady state (SS=1) dose at time 0 with a correct value for AMT
producing the initial correct value in the given compartment, or
2) put a dummy unit dose of 1 at time 0, as you did, and set F1, F2, ..., F5 the
bioavailability parameters for each compartment to their respective initial values.
Here is a simple illustration for an indirect action model.
$MODEL
COMP=(CENTRAL) ; Compartment 1
COMP=(RESP) ; Compartment 1
$PK
KIN=THETA(1)
KOUT=THETA(2)
BASE=KIN/KOUT ; Baseline is the ratio of Kin and Kout
; INITIALISATION OF RESPONSE COMPARTMENT
; NEED to have AMT=1 at TIME=0 in compartment 2
F2=BASE
Best regards,
Pascal Girard
From: phil.lowe@pharma.novartis.com
Subject: RE:[NMusers] no. of dummy points affect simulation and estimation results
Date: Tue, April 20, 2004 6:14 am
Dear Sandra,
Although interested in the responses from nmusers on the technical fixes for the
sampling times, with your PD system, is there a way to calculate what the steady-state
should be from time zero and set this explicitly? This would avoid having to allow
time for the system to come to equilibrium.
I copy below some code I was working on over a weekend to explore the possibility
of estimating PD turnover rates for a linked PD system. In this example, the starting
conditions for each of the PD variables is set using the fraction, F, and setting a
dose of 1 at t=0 in the data file. I have not got any further as yet to explore other
interconnection possibilities than a linear chain. A bit of algebra would be required
to calculate the initial conditions for more complex networks.
Best regards, Phil.
Philip Lowe PhD
Modelling & Systems Biology
Clinical Development & medical Affairs
Novartis Pharmaceuticals AG
4002 Basel
Switzerland
From: musor000@optonline.net
Subject: RE:[NMusers] no. of dummy points affect simulation and estimation results
Date: Tue, April 20, 2004 7:02 pm
Sandra,
It seems like NONMEM starts integration at time equal to the time in the firist data
record. If we do not specify dummy record with time=0 then NONMEM asumes that
patients get infusion at time of the first blood sample.
Take care,
Pavel
_______________________________________________________