FW: Model building
Dear All,
For the sake of correctness, Residuals Analysis is one of the most
relevant topics in the field of Regression. The very notion of objective
function is nothing more than a single characteristic of a type of
residuals. So, raw residuals, standardized residuals, partial residuals,
studentized residuals and so on, are essential to assess serial
correlation, collinearity, leverage, hat-matrix, scedasticity,
transformations, and on, and on, particularly in multivariable and
nonlinear regression. The fact that Nonmem only provides some residuals
by default does not mean we shouldn't look at them and even calculate
others. Using just the value of the objective function and a predictive
check measure is like stirring a boat in the fog with the eyes just
glued on the bow. The picture is much broader. Please check the
extensive literature on Regression and model identification.
Cheers
Luis
-----------------------
Luis Pereira, PhD
Associate Professor
Childrens' Hospital Boston
Harvard Medical School
Boston MA02115
Quoted reply history
________________________________
From: [email protected] on behalf of Nick Holford
Sent: Fri 12/4/2009 10:14 PM
To: nmusers
Subject: Re: [NMusers] Model building
Leonid,
I rely on the objective function for model development. Note the word
objective.
I have never looked at a RES or WRES plot except to laugh at the
subjective foolishness one can imagine there. Note the word subjective.
Of course one can run into problems by looking only at the objective
function but that is when the VPC is most helpful. I like to use the VPC
to decide which model(s) are fit for purpose.
Thank you for recognizing my extreme views. I prefer to be an outlier
than lost among the pseudo-random residuals :-)
Nick
Leonid Gibiansky wrote:
Hi Nick,
As usual, you are very extreme. VPC could be more sensitive in some
cases but the first step is to get the model with good fits, RES, WRES
plots. The original question was whether to choose the model with
numerical problems but good WRES plots versus converged problem with bad
WRES plots. Your answer effectively means: do VPC fists, then decide.
Let me disagree and recommend the model with better WRES plots even if
this model does not converge.
Thanks
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com http://www.quantpharm.com/
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
Nick Holford wrote:
Indranil Bhattacharya wrote:
"So my question is whether the fits, RES, WRES plots and the ofv values
have meaning even when the minimization terminates"
I do not agree with Joachim that RES, WRES are useful. IMHO these have
almost no diagnostic merit except for the most extreme cases of a bad
model. Simulation based diagnostics (VPC, SPC) have better diagnostic
properties and are being actively evaluated by many groups interested in
modelling methodology.
See Karlsson MO, Savic RM. Diagnosing model diagnostics. Clin Pharmacol
Ther. 2007 Jul;82(1):17-20. for a demonstration of the problems.
Nick
Thanks Joachim, that is what I thought. I wanted to be sure that I
invest time building the right model and not just some model which works
(converges) but is biased.
Neil
On Fri, Dec 4, 2009 at 3:31 AM, Grevel, Joachim
<[email protected] <mailto:[email protected]>
<mailto:[email protected]> > wrote:
Hi Neil,
You ask:
"So my question is whether the fits, RES, WRES plots and the ofv
values have meaning even when the minimization terminates"
The answer: you bet they matter! Residual plots are the most
informative output NONMEM gives you. They should guide you when
you determine the basic structure of your model that is supported
by the data. Successful termination, covariance step, standard
errors, Eigen values, messages, warnings... are just icing on the
cake vis-a-vis the residual plots.
These are my two pennies worth of advice,
Joachim
*Joachim Grevel *
Senior Pharmacometrician
_____________________________________________________________________
*AstraZeneca R&D Charnwood*
Clinical Pharmacology & DMPK, Charnwood
Bakewell Road, Loughborough, Leics., LE11 5RH, England
Tel +44 (0) 1509 644035 Fax +44 (0) 1509 645576 Mobile +44 (0)
7920 285905
[email protected]
<mailto:[email protected]>
<mailto:[email protected]> _
_ _
P Please consider the environment before printing this e-mail
------------------------------------------------------------------------
AstraZeneca UK Limited is a company incorporated in England and
Wales with registered number: 03674842 and a registered office at
15 Stanhope Gate, London W1K 1LN.
*Confidentiality Notice: *This message is private and may contain
confidential, proprietary and legally privileged information. If
you have received this message in error, please notify us and
remove it from your system and note that you must not copy,
distribute or take any action in reliance on it. Any unauthorised
use or disclosure of the contents of this message is not permitted
and may be unlawful.
*Disclaimer:* Email messages may be subject to delays,
interception, non-delivery and unauthorised alterations.
Therefore, information expressed in this message is not given or
endorsed by AstraZeneca UK Limited unless otherwise notified by an
authorised representative independent of this message. No
contractual relationship is created by this message by any person
unless specifically indicated by agreement in writing other than
email.
*Monitoring: *AstraZeneca UK Limited may monitor email traffic
data and content for the purposes of the prevention and detection
of crime, ensuring the security of our computer systems and
checking compliance with our Code of Conduct and policies.
*From:* [email protected]
<mailto:[email protected]>
<mailto:[email protected]>
[mailto:[email protected]
<mailto:[email protected]>
<mailto:[email protected]> ] *On Behalf Of *Indranil
Bhattacharya
*Sent:* 03 December 2009 15:41
*To:* [email protected] <mailto:[email protected]>
<mailto:[email protected]>
*Subject:* [NMusers] Model building
Hi, I am in the process of developing a PK/PD model and have a
naive question regarding model building.
I currently do not have all the PD data and more data would be
available in the future. For the current data set, I have tried
models say A to D.
Now model A (cell kill) and B (cell kill +transduction) converges
using FOCE (no CV% but I am willing to live with that) but from
the RES and WRES plots we can clearly see that there is some bias.
The fit is OK but not great. The ofv values are around 400.
Now models B (cell cycle specific kill), C (cell + precursor cell
kill +transduction) and D (cell cycle specific + indirect response
model) do not converge using FO or FOCE methods but when I look at
the fits from the terminated runs, the fits are much better than
those obtained with Model A, and there seems very little bias.
Also the ofv values are between 160-250.
So my question is whether the fits, RES, WRES plots and the ofv
values have meaning even when the minimization terminates.
Regards
Neil
-- Indranil Bhattacharya
--
Indranil Bhattacharya
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology & Clinical Pharmacology
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: [email protected]
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology & Clinical Pharmacology
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: [email protected]
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford