NONMEM Users Network Archive

Hosted by Cognigen

RE: General question on modeling

From: Mark Sale - Next Level Solutions <mark>
Date: Tue, 20 Mar 2007 11:54:23 -0700

OK, one last,
  Yaning, I agree that we are unlikely to miss REALLY important
covariates. But, recently in this forum someone was confused about why
they got different answers with different sequences of steps (I think it
involved ETAs, but I'm not sure), then there is the paper by Janet Wade,
Stuart and Nancy Sambol (Interaction between structural, statistical,
and covariate models in population pharmacokinetic analysis. J PK
Biopharm). Finally, a recent personal experience. Had data on a drug,
found that weight and age were predictors of clearance (not a shock,
pretty standard stuff). It was known that the drug is almost entirely
renally cleared, so certainly CRCL (by Cockcroft-Gault) would predict
clearance. But, it didn't. The combination of age and weight
explained most of the variability in the estimate of CRCL, so CRCL had
no effect on clearance, when age and weight were in the model. When
age and weight were taken out, now CRCL is important. So, you could
have 2 of the 3 effects, age, weight or CRCL. Which will end up in the
model? Depends which sequence they are added - probably random. So, it
is possible that the dosing algorithm will depend on which sequence the
modeler chose to test the effects. I find this a little concerning. I
suspect that our algorithm usually works well, but sometimes (we likely
will never know how often) it fails us.


Next Level Solutions, LLC

> -------- Original Message --------
> Subject: RE: [NMusers] General question on modeling
> From: "Wang, Yaning" <
> Date: Tue, March 20, 2007 1:33 pm
> To: "Mark Sale - Next Level Solutions" <mark
> nmusers
> In Marie Davidian's book, Nonlinear Models for Repeated Measurement
> Data, she briefly discussed this as "Graphical approaches to model
> selection". The first paper about this approach was traced back to the
> 3-step approach in Maitre,et al (1991) (the credit probably should have
> been given to the Fig. 11.1 of the NONMEM Users Guide Part V). She also
> applied the simialr method in semi-parametric modeling and mentioned the
> Bayesian counterpart of this approach. Without rigorous evaluation of
> this approach, it seems that the statistical field also follows this
> intuitively reasonable method. With or without looping back, it is hard
> to miss a REALLY important covariate.
> Yaning Wang, Ph.D.
> Team Leader, Pharmacometrics
> Office of Clinical Pharmacology
> Office of Translational Science
> Center for Drug Evaluation and Research
> U.S. Food and Drug Administration
> Phone: 301-796-1624
> Email:
> "The contents of this message are mine personally and do not necessarily
> reflect any position of the Government or the Food and Drug
> Administration."
> On Mon, 19 Mar 2007 11:32:54 -0700, "Mark Sale - Next Level Solutions"
> <mark
> > Dear Colleagues, I've lately been reviewing the literature on model
> > building/selection algorithms. I have been unable to find any even
> > remotely rigorous discussion of the way we all build NONMEM models.
> > The structural first, then variances/forward addition/backward
> > elimination is generally mentioned in a number of places (Ene Ettes in
> > Ann Pharmacother, 2004, Jaap Mandemas series on POP PK series J PK
> > Biopharm in 1992, Jose Pinheiros paper from the Joint Stats meeting in
> > 1994, Peter Bonates AAPS journal article in 2005, Mats Karlsons AAPS
> > PharmSci, 2002, the FDA guidance on Pop PK). It is most explicitly
> > stated in the NONMEM manuals (Vol 5, figure 11.1) - without any
> > reference. From the NONMEM manuals it is reproduced in many courses,
> > and has become axiomatic. I've looked at the stats literature on
> > forward addition/backwards elimination in both linear and logistic
> > regression, where it is at least formally discussed (with some
> > disagreement about whether it is "correct"). But, I am unable to find
> > any justification for the structural first, then covariates (drive by
> > post-hoc plots), then variance effects approach we use (I'm sure many
> > people will point out that it is not nearly that linear a process,
> > although in figure 11.1, Vol 5 of the NONMEM manuals, it is depicted
> > as a step-by-step algorithm, without any looping back). Can anyone
> > point me to any rigorous discussion of this model building strategy?
> >
> > Mark Sale MD Next Level Solutions, LLC
> >
> >
> >
> --
> Alison Boeckmann
> alisonboeckmann
Received on Tue Mar 20 2007 - 14:54:23 EDT

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to:

Once subscribed, you may contribute to the discussion by emailing: