- On 30 Jun 2003 at 15:01:44, "Nathan Lack" (nlack.at.anormed.com) sent the message

Back to the Top

The following message was posted to: PharmPK

I am wishing to replicate the study performed by Drusano et al. in

Antimicrobial agents and Chemotherapy 2002, Vol. 46 No. 3, p. 913-916.

In this, and other studies I have read, they predominately use the

program NPEM2 of the USC*PACK to generate population PK parameters.

What would the advantage of such a program have over the program ADAPT

II, WinNonMix or NonMem. In addition, could you still use the

parameters generated in ADAPT II to run an accurate Monte Carlo

simulation. Thank you for your assistance.

Regards,

Nathan Lack

Pharm/Tox

AnorMed Inc.

T: (604) 530-1057 x.795

F: (604) 530-0976

W: www.anormed.com

AnorMED Inc.

#200-20353 64th Ave

Langley, BC

Canada V2Y 1N5 - On 1 Jul 2003 at 11:39:44, Nick Holford (n.holford.-at-.auckland.ac.nz) sent the message

Back to the Top

The following message was posted to: PharmPK

Nathan,

Nathan Lack wrote:

>

> PharmPK - Discussions about Pharmacokinetics

> Pharmacodynamics and related topics

>

> The following message was posted to: PharmPK

>

> I am wishing to replicate the study performed by Drusano et al. in

> Antimicrobial agents and Chemotherapy 2002, Vol. 46 No. 3, p. 913-916.

> In this, and other studies I have read, they predominately use the

> program NPEM2 of the USC*PACK to generate population PK parameters.

> What would the advantage of such a program have over the program ADAPT

> II, WinNonMix or NonMem. In addition, could you still use the

> parameters generated in ADAPT II to run an accurate Monte Carlo

> simulation. Thank you for your assistance.

My understanding is that NPEM is an *estimation* rather than a

*simulation* program. It can be used to estimate pop PK parameters

(e.g. Clearance and its variability). WinNonMix and NONMEM are also

population estimation programs. WinNonMix cannot be used for Monte

Carlo simulation but NONMEM can. I don't know about ADAPT II for

simulation but it is not a true population estimation program. You

could use the standard two stage procedure to estimate population

parameters from individual estimates obtained with ADAPT. You can do a

stochastic (Monte Carlo) simulation using parameters from any source

you like. The "accuracy" of the simulation depends on the questions you

are trying to ask. There will always be some uncertainty in the

parameters and models no matter where they come from which may or may

not have an important impact on the objective of the simulation.

Nick

Nick Holford, Dept Pharmacology & Clinical Pharmacology

University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New

Zealand

email:n.holford.aaa.auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556

http://www.health.auckland.ac.nz/pharmacology/staff/nholford/ - On 1 Jul 2003 at 09:51:17, "Zhang, Jianping [CNTUS]" (JZhang9.at.CNTUS.JNJ.COM) sent the message

Back to the Top

Nick,

As I recall WinNonMix is able to do Monte Carlo simulation.

JP

JP Zhang, Ph.D.

Centocor, Inc.

[Boomer can ;-) - db] - On 1 Jul 2003 at 10:46:50, Bob Leary (leary.aaa.sdsc.edu) sent the message

Back to the Top

The following message was posted to: PharmPK

Nick Holford is correct that NPEM is an estimation program

rather than a simulation program. In the Drusano work referenced it

appears that NPEM was used to estimate PK population distribution

parameters, while ADAPT II was used for the Monte Carlo simulation.

The use of NPEM has several advantages, both generically and specific to

this context:

1. NPEM (and the follow-on NPAG program, which I believe is what was

used in the Drusano paper ), use an estimation technique based on

the maximization of an exact, nonparametric likelihood function and

produce statistically consistent population estimators. Parametric

programs such as NONMEM are based on the maximization of an approximate

(using such approximations as FO, FOCE, and Laplace) parametric

likelihood

function. The approximation, at least in the case of FO and FOCE,

involves the linearization of the model function that is the

solution to the kinetics equations, and can be arbitrarily bad. The

consequences of using

such approximations are loss of statistical consistency and

degradation of statistical efficiency (assuming that statistical

efficiency is a meaningful concept in the absence of

consistency, which is debatable).

Consistency has a price in terms of the amount of computation required,

and until several years ago NPEM required a supercomputer for the best

results. However, recent improvements in the algorithm, now implemented

as NPAG, allow most computations to be performed in reasonable times on

a PC.

2. The estimators produced by NPEM/NPAG are nonparametric distributions.

No initial specification of an assumed parametric form is necessary.

However, the NPEM/NPAG estimator can then be converted into any

particular

parametric estimator desired, preserving consistency. Thus a single

NPEM/NPAG esimator can be used to explore the viability of many

different

parametric assumptions, which is useful in the Monte Carlo simulation

context.

3. The NPEM and ADAPT series of programs were both developed at USC and

share common data file formats, and to this extent may be interoperable.

Nathan - if you wish to replicate George Drusano's study using the same

analytic methodology, I suggest you get in contact with the USC folks

(www.lapk.edu/hsc/lap_apk) who can help you get started with NPEM.

Bob Leary

Bob Leary

Senior Staff Scientist

San Diego Supercomputer Center

858-534-5123 - On 2 Jul 2003 at 10:17:42, Nick Holford (n.holford.-at-.auckland.ac.nz) sent the message

Back to the Top

The following message was posted to: PharmPK

JP,

"Zhang, Jianping [CNTUS]" wrote:

>

> PharmPK - Discussions about Pharmacokinetics

> Pharmacodynamics and related topics

>

> Nick,

>

> As I recall WinNonMix is able to do Monte Carlo simulation.

>

> JP

>

> JP Zhang, Ph.D.

> Centocor, Inc.

It is my recollection of WinNonMix that it had no built in mechanism

for automated replication of simulation data sets. This means that one

must simulate and estimate one data set at a time which makes it

impractical for Monte Carlo simulation experiments which typically

involve at least 100 and sometimes over 1000 replications in order to

obtain adequate precision of statistics of interest.

I have not used WinNonMix for some years for this and other reasons.

Perhaps some one from Pharsight can comment on this issue.

Nick

Nick Holford, Dept Pharmacology & Clinical Pharmacology

University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New

Zealand

email:n.holford.-at-.auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556

http://www.health.auckland.ac.nz/pharmacology/staff/nholford/ - On 2 Jul 2003 at 14:37:43, "Sima Sadray" (sadrai.-a-.sina.tums.ac.ir) sent the message

Back to the Top

The following message was posted to: PharmPK

Dear All,

PKBUGs under WINBUGs also can run Monte Carlo Simulations.

http://www.med.ic.ac.uk/divisions/60/pkbugs_web/home.html

Sadray - On 2 Jul 2003 at 17:51:51, Roger Jelliffe (jelliffe.-at-.usc.edu) sent the message

Back to the Top

Dear Nathan:

You might talk this over with George Drusano. In general,

though, parametric population modeling programs like IT2B and NONMEM do

not have the same consistent behavior as the nonparametric (NP) ones.

There is no guarantee that the more subjects you study in the

population, the closer the results will be to the truth. The reason for

this is the approximations that are made in computing the likelihood in

the parametric software using the FO or FOCE approximations, for

example. Bob Leary has examined this very carefully in material he

presented at the PAGE meeting in Paris in 2002. It is on our web site

www.lapk.org, under new advances in PK/PD modeling.

In contrast, since there is nothing to integrate in the NP

approach, the likelihood is exact and the behavior is consistent.

Because of this, even if the true population distribution is Gaussian,

the means and variances obtained with the NP methods are usually of

better quality than those with parametric methods.

The literature has shown that the population parameter CV's are

often smaller with FOCE methods, and people have thought that this

means that the population parameter estimates are more precise.

However, the likelihoods are usually not reported. When we have

examined this relationship, the pop parameter CV's were indeed smaller,

but the likelihoods were less that with the NP methods, showing that

the NP methods were better able to perceive the true diversity of the

parameter distributions in the population, as they were not inhibited

by any assumptions of shape such as normal or lognormal. The likelihood

is the important thing, not the usual indices of goodness of fit, as

the parameter distributions are often genetically polymorphic. This is

not nearly as easily seen with parametric methods that do not show you

the entire population parameter distributions, but only something that

has some assumed shape, normal or lognormal, etc.

In addition, NP methods lend themselves to the design of better

(more precise) dosage regimens, using "multiple model" dosage design,

which specifically develops dosage regimens to hit target goals with

maximum precision (minimum weighted squared error). Our new clinical

software uses this approach. More information is on our web site

www.lapk.org.

Very best regards,

Roger Jelliffe

Roger W. Jelliffe, M.D. Professor of Medicine,

Division of Geriatric Medicine,

Laboratory of Applied Pharmacokinetics,

USC Keck School of Medicine

2250 Alcazar St, Los Angeles CA 90033, USA

Phone (323)442-1300, fax (323)442-1302, email= jelliffe.-a-.usc.edu

Our web site= http://www.lapk.org - On 3 Jul 2003 at 08:43:06, =?ISO-8859-1?Q?Fr=E9d=E9ric_BOIS?= (Frederic.Bois.at.ineris.fr) sent the message

Back to the Top

The following message was posted to: PharmPK

Dear All,

We use MCSim (freely available at http://toxi.ineris.fr on the MCSim

page) to do

Monte Carlo simulations, including Markov Chain Monte Carlo.

Frederic Bois

Frederic Y. Bois,

Unite de Toxicologie Experimentale, responsable

INERIS

Parc ALATA, BP 2

60550 Verneuil en Halatte

FRANCE

tel: + 33 (0)3 44 55 65 96

fax: + 33 (0)3 44 55 66 05

email: frederic.bois.aaa.ineris.fr

web: http://www.ineris.fr, http://toxi.ineris.fr - On 8 Jul 2003 at 13:10:52, "Linda Hughes" (LHughes.at.Pharsight.com) sent the message

Back to the Top

Nick is correct that the current version of WinNonMix does not have a

built-in mechanism for automated replication of simulation data sets.

Only one simulation can be done at a time, and a constant seed is used

in the random number generator so that consistent simulation results

will be produced with the same parameter specification. A mechanism

for automated replication of simulation data sets is being considered

for future versions.

Regards,

Linda Hughes, M.S., Software Development Group

Pharsight Corporation, 5520 Dillard Dr., Suite 210, Cary, NC 27511

Want to post a follow-up message on this topic? If this link does not work with your browser send a follow-up message to PharmPK@boomer.org with "Monte Carlo Simulations" as the subject

PharmPK Discussion List Archive Index page

Copyright 1995-2010 David W. A. Bourne (david@boomer.org)