Chuanpu,
Thanks for bringing my attention to your excellent simulation study of
the problems associated with predicting doses from model based analyses.
In your study the highest effect for which doses was predicted was 50%
of Emax so I would not expect any real difference between choosing a
linear or Emax model to make that kind of prediction. Also the study
designs were essentially setup for interpolation. I would also expect
that an empirical model should be able to interpolate equally well as a
more mechanistic model (provided the empirical model is not too bizarre).
I wrote earlier:
The usual purpose of the model is to predict the effect over a range
of concentrations. If you choose a linear model because your
subjective impression is that the model is "overparameterised" due to
large standard errors then you can be certain that any extrapolation
will overpredict the size of the effect. If you choose an Emax model
you may still have a biased prediction but it will be a better
prediction than one from a linear model. In the interpolation range of
predictions the Emax model will still do better. I cannot see how it
can do worse than the linear model (assuming the model passes other
tests of plausibility and the VPC looks OK).
Referring to this assertion you said in another email:
Our previously mentioned simulations showed exactly the opposite in
certain situations - i.e., when the power is low. The Emax model
predicted worse because of instability, even though it was the "true"
model.
I had some difficulty identifying where this is described in your paper.
Would you please guide me to the page(s) where this can be found? It is
however not surprising that if the design is very poor then any model is
going to be poor at making predictions.
So until I understand your paper better I think I will stick with my
original assertion that an Emax model is better than a linear model when
one is intends to use it for extrapolation and will be equivalent to
empirical models for interpolation.
Chuanpu H, Yingwen D. Estimating the predictive quality of dose-response
after model selection. Stat Med. 2007;26(16):3114-39.
Hu, Chuanpu [CNTUS] wrote:
We have conducted simulations to show that an over-parameterized model,
even if "true" and "significant," could give worse predictions (ref
below). The simulations were conducted perhaps more like in the context
of interpolations. What happens in extrapolation will be very much
depend on the specific situation. However this suggests that the
empirical model may deserve to be given more consideration.
Reference: Hu C, Dong Y., Estimating the predictive quality of
dose-response after model selection. Statistics in Medicine 2007;
26:3114-3139.
Chuanpu
~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*
Chuanpu Hu, Ph.D.
Director, Pharmacometrics
Pharmacokinetics
C-3-3
Biotechnology, Immunology & Oncology (B.I.O.)
Johnson and Johnson
200 Great Valley Parkway
Malvern, PA 19355
Tel: 610-651-7423
Fax: (610) 993-7801
E-mail: [email protected]
~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*
--
Nick Holford, Professor Clinical Pharmacology
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
mobile: +64 21 46 23 53
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
