From: musor000@optonline.net
Subject: [NMusers] PD modeling:  What is the most efficient approach to transfer data?
Date: Mon, 19 Dec 2005 21:14:22 -0500

Hello Nonmem Users,

I am working on a PD model.  Usually, it is not recommended to combine PK and PD models. 
First, it is necessary to develop a PK model.  Then predicted values can be used in a PD model.

What is the most efficient way to transfer data from PK to PD model?  I simply took
output PK dataset, modified it, and used it in a PD model.  It was time consuming.  Is
there another way to transfer data (supermodel or something else)?

Frequently, it is necessary to have more predicted timepoints than observed timepoints. 
One way to get predicted values is to create missing observations in an input dataset. 
This approach is time consuming and prone to errors.  Is there a more simple way to get
the predicted values?

Kind regards,

Pavel
_______________________________________________________

From: Nick Holford n.holford@auckland.ac.nz
Subject: Re: [NMusers] PD modeling:  What is the most efficient approach to transfer data?
Date: Tue, 20 Dec 2005 17:03:47 +1300

Pavel,

Where did you get this idea from?
" Usually, it is not recommended to combine PK and PD models."

If you read Zhang et al. I think you will find the opposite conclusion. i.e. best
estimates are obtained by simultaneous analyis of PK and PD observations.

Zhang L, Beal SL, Sheiner LB. Simultaneous vs. sequential analysis for population PK/PD
data I: best-case performance. J Pharmacokinet Pharmacodyn. 2003 Dec;30(6):387-404.

Because of time contraints you may wish to do sequential PK then PD analyses. Zhang et
al describe 3 flavours -- PPP, IPP and PPPD. Overall they recommend the PPPD approach. This
is quite easy to set up once you have combined the PK and PD observations into a single
dataset. You just write the full model as if for a simultaneous analysis but FIX all the
PK parameters. The data set includes the PK observations (the 'D' in PPPD) and the model
uses the fixed population PK parameters (the 'PPP' in PPPD).

Nick
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
email:n.holford@auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
_______________________________________________________

From: "Bachman, William (MYD)" bachmanw@iconus.com
Subject: RE: [NMusers] PD modeling: What is the most efficient approach to transfer data?
Date: Tue, 20 Dec 2005 08:06:04 -0500

The argument made "against" simultaneous PK-PD modeling is that potentially
the PD data could drive the PD model (the cart before the horse).  Make your
own conclusion, just stating the usual argument.

The alternative to "transferring data from PK to PD model" is to generate
the individualized PK parameters (POSTHOC or FOCE estimates) and include
them in the PD data file.  Then insert them into the PK model that generates
predictions for the PD model.

Adding more time points with EVID=2 is the only way I can think of to
generate more time points.  The only errors you can make are getting the
times wrong or not asking for the prediction in the correct CMT.
_______________________________________________________

From: "Bachman, William (MYD)" bachmanw@iconus.com
Subject: RE: [NMusers] PD modeling: What is the most efficient approach to transfer data?
Date: Tue, 20 Dec 2005 11:45:20 -0500

 that should have read "the PK data could drive the PD model"
_______________________________________________________

From: "Bachman, William (MYD)" bachmanw@iconus.com
Subject: RE: [NMusers] PD modeling: What is the most efficient approach to transfer data? 
Date:  Tue, 20 Dec 2005 11:48:26 -0500

I'll get it right yet!  "the PD data could drive the PK model "
_______________________________________________________

From: "Pereira, Luis" Luis.Pereira@bos.mcphs.edu
Subject: RE: [NMusers] PD modeling: What is the most efficient approach to transfer data?
Date: Tue, 20 Dec 2005 12:10:46 -0500

I just cannot agree with any argument against simultaneous PKPD modeling. What
'drives' the model is Information, whether PK or PD, and one should not forget
the initial objective setup for the modeling exercise. Time points should
ultimately be chosen based on an optimality criterion.

---------------------------------------------------------------
Luis M. Pereira, Ph.D.
Assistant Professor, Biopharmaceutics and Pharmacokinetics
Massachusetts College of Pharmacy and Health Sciences
179 Longwood Ave, Boston, MA  02115
Phone: (617) 732-2905
Fax: (617) 732-2228
Luis.Pereira@bos.mcphs.edu
 
_______________________________________________________

From: Nick Holford 
Subject: Re: [NMusers] PD modeling: What is the most efficient approach to transfer data?
Date:  Wed, 21 Dec 2005 07:02:37 +1300

As Bill has pointed out there is a caution one should be especially aware of when doing
simultaneous PKPD. If the PD model (including any link between PK and PD such as effect
compartment model or turnover model) is badly wrong then it can indeed cause the PK estimates
to become biased. This is discussed in the follow on paper by Zheng et al.

Zhang L, Beal SL, Sheiner LB. Simultaneous vs. sequential analysis for population PK/PD
data II: robustness of methods. J Pharmacokinet Pharmacodyn 2003;30(6):405-16.

--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
email:n.holford@auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
_______________________________________________________