Karina,
Luann Phillips is right.
Additionally,
which initial estimates do you use for the respective rate constants? Which
ADVAN subroutine are you currently using?
Maybe there is some misspecification that your rate constants "try" to become
zero...
Sebastian Frechen
Clinical Pharmacology
University Hospital Cologne
-----Ursprüngliche Nachricht-----
Von: [email protected] [mailto:[email protected]] Im
Auftrag von Luann Phillips
Gesendet: Freitag, 11. Januar 2013 20:08
An: VegaVilla, Karina R (HSC)
Cc: [email protected]
Betreff: Re: [NMusers] low parameter estimates and TOL value
Karina,
TOL=4 is on the low side for most models. A TOL=4 implies a SIGL<=4 and
SIGDIG=4/3 or 1.33.
This would mean that all of the final parameter estimates could have less than
2 significant digits.
(See Intro to NONMEM 7 guide section I.14 for further details about controlling
the precision of the gradient evaluation).
I typically use SIGDIG=3 with SIGL=9 and TOL=9 in NONMEM 7. For NONMEM 6, SIGL
can not be specified and I have not had much success with TOL values > 6.
Luann Phillips
Director, PK/PD
Cognigen Corporation
Quoted reply history
On 1/11/2013 12:56 PM, VegaVilla, Karina R (HSC) wrote:
> Dear NMusers,
>
> I have developed a model that has some rate constant estimates with
> very low values, (i.e. 10E-05) and I'm running NONMEM with TOL=4. Is that the
> appropriate approach? If not, what would you suggest.
>
> Thanks.
>
> Karina R. Vega-Villa, Ph.D.
>
> Postdoctoral Research Associate
>