Re: FW: Revisit parameterization question
Kuenhi,
The fact that in your particular data set you dont see a strong correlation between CL and V does not mean they are not correlated via weight and many other factors. The estimated correlation is determined by design (or lack of design) of your experiment. If you had a really big weight range e.g. 1 kg to 150 kg then you would clearly appreciate the weight effect. But in the usual adult weight range there are other sources of between subject variability that frequently hide the weight effect.
Dont fret over small correlation differences e.g. CL and V with weight. Correlation is the weakest form of science and should only be considered as a last resort to point in the direction of understanding something. The relation between CL, V and weight is rock solid truth. There is no need to fumble about with correlations and other statistically short-sighted metrics.
Monolix will estimate correlations only if you ask it to. Its very easy to specify with the visual interface. Just click on the off-diagonal elements of the variance-covariance matrix box. When you get a 1 it means that the covariance will be estimated.
Dont expect Monolix and NONMEM to give identical results. On average if all other things are the same you should trust Monolix over NONMEM because it uses a provably better estimation algorithm.
Nick
Tsai, Kuenhi wrote:
> Nick,
>
> Thank you SO much for your help. I read your discussion on this issue in
> 2001. Your and Steve Dullful's helpful reply confirmed me that I should
> redo my models now using TRANS4! And I need to think and learn about
> other issues (F1, protein binding,..) you mentioned in your email.
>
> One question... Although I believe CL and V is correlated to weight, the
> correlation (using MONOMIX) between V1 and weight is < 0.2. Should I
> still incorporate weight as a covariate into V1? Also my CL has higher
> correlation to height instead weight (0.35 vs. 0.2), shall I use height
> instead of weight as a covariate for CL? Do I stick too much on
> algebraic results?
>
> I am running models by both NONMEM and MONOLIX. Even I use the same
> models (perhaps, since I am not sure whether MONOLIX automatically
> incorporates all correlations of thetas in estimation), I get the
> different results of estimation. Did you have any experiences on this
> issue? Marc gave me some suggestions. I haven't done so yet, but like
> to hear your comments.
>
> Thank you very much again for your help.
>
> All the best
>
> Kuenhi
>
Quoted reply history
> -----Original Message-----
>
> From: Nick Holford [ mailto:[EMAIL PROTECTED] Sent: Monday, September 15, 2008 3:40 PM
>
> To: Tsai, Kuenhi
> Subject: Re: [NMusers] Revisit parameterization question
>
> Kuenhi,
>
> Parameterisation is important for intepretation and for estimation and should be distinguished from the algebraic convenience of performing some calculation.
>
> Interpretation of parameters is more useful when the parameters can be related to physiological or pharmacological mechanisms. Parameters estimated in this way can have their variability explained better by other covariates e.g. renal function will change clearance of renally cleared drugs but there is no physiological entity resembling a rate constant so parameterisation in terms of a rate constants will always require an empirical application of renal function.
>
> Differences in estimation can sometimes be observed with different parameterisations because of the dependence on numerical issues related to such matters as derivatives. But this is only a challenge for better computer hardware and software and not of fundamental importance.
>
> So my bottom line preference is to parameterise in terms of quantities that can be mapped to some mechanistic or physical reality. This means using volumes and clearances for distribution and mass transfer
>
> kinetics.
>
> Note also that one of the most important mechanistic causes for correlation between CL and V is weight. You should include weight on all
>
> CL, V1, Q and V2 because there is no sensible reason to believe they do not increase with weight. After that you might add an ETA to F1 in order
>
> to capture other correlations due to between subject variability in bioavailability and protein binding. I would definitely use TRANS4 always in preference to TRANS1.
>
> Nick
>
> Tsai, Kuenhi wrote:
>
> > Dear All,
> >
> > I am working on a data set that the estimates become quite different
> > under different model assumptions (such as fixed or not fixed ka,
> > different block structure) using ADVAN4 TRANS1. One suggests me that
>
> I
>
> > should use TRANS4 to avoid the problem of the high correlations among
> > ka, Cl, V, k, and k12. When I reviewed the previous NM user
>
> discussion
>
> > in "reparameterization", the highest "scored" discussion occurred in
> > 2001. I am wondering whether there are any updated discussion and
> > references on this topic. Any of your help are greatly appreciated.
> >
> > All the best
> >
> > Kuenhi Tsai
> > Notice: This e-mail message, together with any attachments, contains
> > information of Merck & Co., Inc. (One Merck Drive, Whitehouse Station,
> > New Jersey, USA 08889), and/or its affiliates (which may be known
> > outside the United States as Merck Frosst, Merck Sharp & Dohme or
> > MSD and in Japan, as Banyu - direct contact information for affiliates
>
> is
>
> > available at http://www.merck.com/contact/contacts.html) that may be
> > confidential, proprietary copyrighted and/or legally privileged. It is
> > intended solely for the use of the individual or entity named on this
> > message. If you are not the intended recipient, and have received this
> > message in error, please notify us immediately by reply e-mail and
> > then delete it from your system.
--
Nick Holford, 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
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford