From:"Joern Loetsch"
Subject:[NMusers] error models log domain
Date:Mon, January 28, 2002 12:57 pm
Dear nonmem users,
what error model could be used for fitting in the log domain when the simple proportional
Y=LOG(F)+EPS(1) appears to result in a tendency of underestimation of higher plasma concentrations.
Thank you in advance
J. lotsch
____________________________________________
Jorn Lotsch, MD
pharmazentrum frankfurt, Dept. of Clinical Pharmacology
Johann Wolfgang Goethe-University Hospital
Theodor-Stern-Kai 7
D-60590 Frankfurt am Main
GERMANY
Tel.:+49-69-6301-4589
Fax.:+49-69-6301-7636
error models log domain
From:LSheiner
Subject:Re: [NMusers] error models log domain
Date:Mon, January 28, 2002 1:22 pm
What makes you think that a bias at high concentrations is due to the
error model?
--
_/ _/ _/_/ _/_/_/ _/_/_/ Profesor Lewis B Sheiner, MD
_/ _/ _/ _/_ _/_/ letter: Box 0626, UCSF, SF, CA, 94143-0626
_/ _/ _/ _/ _/ package: Rm C255, UCSF, SF, CA, 94122-2722
_/_/ _/_/ _/_/_/ _/ 415-476-1965 (v), 415-476-2796 (fax)
From:Steve_Charnick@vpharm.com
Subject:Re: [NMusers] error models log domain
Date:Mon, 28 Jan 2002 13:39:48 -0500
Joern:
Could you perhaps clarify the data situation? I agree with Lew
that the bias at the high concentrations is not necessarily
due to the error model. I'm more curious about the dataset
and the structural model you used as possible sources of
the problem you implied you're having.
Steven
From:"Piotrovskij, Vladimir [PRDBE]"
Subject: RE: [NMusers] error models log domain
Date:Wed, 30 Jan 2002 16:39:05 +0100
Jorn,
As the matter of fact, you can use any reasonable error model.
Particularly, you can assume the additive (not proportional!)
error variance to be concentration-dependent,
e.g., a stepwise function.
However, I am not sure that the bias can be
eliminated by making the residual error model
more complex. Modifying your structural
model might be more logical.
Best regards,
Vladimir