From:"Emily"
Subject:[NMusers] T matrix
Date:Fri, March 8, 2002 9:39 am
Dear nmusers:
First, I'd like to thank Nick. After gave smaller Vm and bigger k12,
I still got the same error message. What esle can I do ?
See 99mar072002.html for previous related discussion.
And I'd like to ask a few questions:
Regarding the questions of T-matrix as well as R and S matrixes,
I am still confused by the following questions. I hope you can give me some inputs:
(1). When a T matrix is in output, does this mean
either R matrix or S matrix is sigular (but not both)?
(2) Does minimization appears only if
R matrix is not singular (no matter what S matrix is)?
(3). How a T matrix could be used
to estimate the confidence region of parameters?
(4) As you can see in my previous message, my control
file indicates that over-parameterization may not exist.
If so, then why I still get a message of R matrix algorithmically singular?
By the way, in nonmem user guide part VIII, page 114, it mentioned that,
for nonlinear model, ADVAN8 may be a preference to ADVAN6 when the
differential equations are stiff. What does the stiff mean here?
Thaks a lot !!
T matrix
From:Nick Holford
Subject:Re: [NMusers] T matrix
Date:Fri, March 8, 2002 4:36 pm
> From:
> "Emily"
> First, I'd like to thank Nick. After gave smaller Vm and bigger k12, I still got
> the same error message. What esle
> can I do ?
You should consider using ADVAN10 which is a special case for the one compartment
Michaelis-Menten elimination model. This may be more numerically stable than ADVAN6
(or 8 or 9) because it does not rely on numerical integration to solve the PK model.
Have you looked at graphs of time vs obs and individual (posthoc) predictions of
conc? You need to be sure you have a good fit before worrying about the small print
standard error issues.
You should understand that the SEs provided by NONMEM are not worth much. If you
want to use them as measure of the confidence interval for your parameters then they
rely on at least two untenable asssumptions 1) you have an infinite amount of data
2) the confidence intervals are symmetrical around the parameter estimate.
Alternative empirical methods e.g. bootstrap or log likelihood profile are better
way of understanding the confidence you have in your parameters.
Its unlikely you have a stiff system especially if you now have more realistic
estimates of Vmax and Ka. You can easily try using ADVAN9 (just change ADVAN6 to
ADVAN9 and re-run). Sometimes ADVAN9 is more robust. I have not had much luck with
ADVAN8 which is designed for systems that really are stiff ie. different rate
constants in the model vary by several orders of magnitude e.g having an absorption
half-life of a few minutes and an elimination half life of several days.
Nick
--
Nick Holford, Divn Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
email:n.holford@auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/