Barry,
Thanks for this information. It is good to know that one can ignore this
limitation. I never understood why it was there - especially in NONMEM 7
which was supposed to be a complete rewrite with more flexible structure.
The inability to write out a label for THETA and ETA after 70 is just
one of those odd things about this program...
Nick
Barry Weatherley wrote:
Nick, only occasionally it is worth while to forget to read the manual!
In this instance (using NONMEM V, not tried for NONMEM 6), I needed
more than the allotted ration of THETAs and ETAs. I had to increase
the variables within *SIZES to allocate bigger LTH, LVR etc. Bill
Bachman gave me a spreadsheet to get the exact sizes of all the array
variables.
The only problem was that the output file could not count the THETAs
and ETAs beyond 70 for labelling them. So above this number the
labels for THETAs and ETAs were hieroglyphics but the values were fine.
Barry
In message <[email protected]>, Nick Holford
<[email protected]> writes
Phylinda,
Thanks for the explanation about the impracticability of using the
'complex flexible input' model. However, I would have thought the
problem was not the run time but the upper limit on number of THETAs
of 70 and on OMEGA+SIGMA of 70 in NONMEM (still there in NONMEM 7!).
"/III.2.9.1. Changing the Number of Theta’s, Eta’s, and Epsilon’s
LTH gives the maximum number of theta’s allowable. It must be between
1 and 70.
LVR gives the maximum number of eta’s plus epsilon’s allowable. It
must be between 1 and 70/" NONMEM VI User Guide III
Where would you get the ultra-big NONMEM version with 97 THETAs and
87 OMEGAs?
Nick
Chan, Phylinda wrote:
Hi Nick,
There are 97 thetas and 87 omegas in the complex flexible input model.
Despite of the run time, it is impractical to apply such model for
covariates searching in the meta-analysis.
Phylinda.
-----Original Message-----
From: [email protected]
[mailto:[email protected]]
On Behalf Of Nick Holford
Sent: 30 September 2009 04:31
To: nmusers
Subject: Re: [NMusers] time-dependent residual error models
Phylinda,
Thanks for the explanation -- it seems that the more usual approach
of complex structure+simple residual error model had already been
done by Barry Weatherley.
Your simple structure+complex residual error is an interesting
alternative but apart from your feelings ("We felt ...") was there
any reason not to use Barry's structural model?
Nick
Chan, Phylinda wrote:
Hi Nick,
Being a substrate of P-gp and CYP3A4, the compound itself has a very
complex absorption profile including dose non-linearity, double peaks,
food effects as well as high between individual and within individual
variability. Barry Weatherley has spent a substantial amount of time
and effort in understanding the dose non-linearity and some covariate
effects on the PK of this compound, including development of a very
complex flexible input model which was presented at PKUK in 2004.
More
details of some of this modelling work can be found in a recent
publication.
http://www3.interscience.wiley.com/journal/122386172/abstract
The main objective of the meta-analysis was to develop a compartmental
model which would be useful in identifying significant covariates
explaining inter-individual variability and was simple enough to be
used
in the later modelling of sparsely sampled PK in phase 2b/3 studies
where a full time profile and samples were likely to be clustered in
the
elimination phase of the PK. We felt the first-order input with dose
and food effects on Ka in addition to the time-dependent residual
error
model was adequate for this purpose.
For those who interested in the coding of the time-dependent residual
error model: $ERROR
IPRED = F+.00001
LPRED = 0
IF(IPRED.GT.0) LPRED = LOG(IPRED)
PMAX=THETA(7) TMAX=THETA(8) K=THETA(9)
BASE=THETA(10)
P=K*TMAX A=EXP(P)/TMAX**P
W= PMAX*A*(TAD+.01)**P*EXP(-K*(TAD+.01))+BASE
IRES= DV-LPRED
IWRES= IRES/W
Y= LPRED+EPS(1) * W
Note:
i) $SIGMA (1 FIX)
ii) TAD=time after dose
Phylinda.
-----Original Message-----
From: [email protected]
[mailto:[email protected]]
On Behalf Of Nick Holford
Sent: 24 September 2009 08:42
To: nmusers
Subject: Re: [NMusers] time-dependent residual error models
Mats,
I agree with your general idea but in this particular case there is no
description in the paper of efforts made to test structural models for
absorption apart from first order input with dose and food effects
on Ka. There seems to be quite a lot of time related structure in
the residual error model function that Phylinda reported and I
would have thought that at least some of this could have been
explored via
another
structural model e.g. involving parallel or sequential zero-order
inputs. It struck me as a rather unusual approach and I wondered
what reasons for it were.
It does not really bother me which approach is used when describing
absorption (fancy structure+vanilla residual or vanilla
structure+fancy
residual) because the details of the rate of absorption rarely have
any
clinical relevance (Justin Wilkins may want to disagree <grin>). Of
course, as you point out the errors may often arise from poorly
reproducible fixed effects such as timing errors etc. and thus the
goal
may be to describe the error adequately and not the structure
because the structure is not really fixed or of any interest.
Nick
Mats Karlsson wrote:
Hi Nick,
I can't answer for Phylinda, but the general idea is to build the
most
appropriate structural model that is supported by data. However,
after
that
is done, if there still is variation in residual error magnitude
should
take that into account and not ignore it. All models are wrong, and
would
say that in general our models for absorption are more wrong than
models
for disposition. That is not just because we have focused more on
latter, but because the underlying processes governing absorption are
of a
different nature (e.g. with discrete events like food intake, gastric
emptying, bile release and formulation disintegration and movement).
Further
often part of the error magnitude is from timing errors. Such errors
are
more pronounced when concentrations are changing fast (normally
fastest
changes in absorption phase). We wrote on time-varying residual
errors
(and
alternatives such as residual error magnitude related to rate of
change) in
these publications: J Pharmacokinet Biopharm. 1995 Dec;23(6):651-72.
J Pharmacokinet Biopharm. 1998 Apr;26(2):207-46
Best regards,
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Dept of Pharmaceutical Biosciences
Uppsala University
Box 591
751 24 Uppsala Sweden
phone: +46 18 4714105
fax: +46 18 471 4003
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On
Behalf Of Nick Holford
Sent: Thursday, September 24, 2009 7:46 AM
To: nmusers
Subject: Re: [NMusers] time-dependent residual error models
Hi,
If Phylinda reads this I'd be interested to hear why she choose to
use
a
plain vanilla first-order absorption model and a fancy time-dependent
residual error model rather than trying to model a fancy
absorption process with a plain vanilla residual error model?
Nick
Joseph Standing wrote:
Xiang,
There is a rather elegant time-dependent residual error model
described by Phylinda Chan et al in:
BJCP, 2008;65(S1):76-85.
BW,
Joe
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
[email protected] tel:+64(9)923-6730 fax:+64(9)373-7090
mobile: +64 21 46 23 53
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
