Dear NM-users,
Currently I am working on the PK data of a drug that is taken twice daily orally. I have full AUC data from 45 individuals with 10 samples per individual over a 12-hour period and abbreviated AUC data from 99 individuals with 5 samples per individual over a 2-hour period. Individual plots and literature demonstrate that a 2-compartment PK model with first order absorption and elimination and with a lagtime should adequately describe the data. To model the data I used ADVAN4 TRANS4 with FO and a combined error model.
Problems occur with fitting the absorption phase of the drug, which results in a structural underestimation of the peak concentration. Plots of predicted versus observed concentration showed deviations from the line of unity. Application of transit-compartments (Erlang) and more flexible absorption profiles (eg Weibull) did not solve this problem. However the fit could be improved by forcing the volume of the central compartment to higher values, which resulted in higher Ka values, higher peak concentrations, better individual plots in Xpose, but an increase of the objective function.
What could be the possible reasons for these observations? Is it appropriate to fix the distribution volume to higher values which produce better goodness-of-fit plots but higher minimum objective function values?
Thanks in advance for your time.
Brenda de Winter
--
/B.C.M. de Winter
Erasmus Medical Center
Department Hospital Pharmacy
Unit Clinical Pharmacology
Room L-056
's-Gravendijkwal 230
3015CE Rotterdam
the Netherlands
T +31 (0)10 46 33202
F +31 (0)10 43 66605
[EMAIL PROTECTED] /
Improved absorption profile by forcing volume of the central compartment
Hello Brenda,
Your sample looks small.
I have 38 subjects with rich data. FO method works, but plots for sume
subjects do not look very good and there are few local maxima. The method with
interaction does not work due to the small sample size. Although the
prelimenary data are useful, nothing can make the plots perfect. I have to
increase the sample size.
Try to run the model for each subject without population estimates. It will
give you an idea how the first order approcximation affects your plots.
Try to play with the model of error. If your doses are very different, you may
need to use CV+additive model.
Try to implement correlation between Cl and V2, but set correlation between Ka
and the other parameters to zero.
Good luck,
Pavel
Quoted reply history
----- Original Message -----
From: Jurgen Bulitta
Date: Tuesday, April 24, 2007 3:47 pm
Subject: Re: [NMusers] Improved absorption profile by forcing volume of the
central compartment
To: "B.C.M. Winter - De" , [email protected]
> Dear Brenda,
>
> Just a couple of comments and questions:
> 1) Is there a specific reason why you are using FO for a dataset
> with
> frequent sampling? How large is your between subject
> variability?
> I would recommend considering FOCE+I in your case (for a more
> detailed assessment, see e.g. Bauer R et al. AAPS J. 2007;9:E60-83.)
>
> 2) Have you tried a 3 compartment model? I would try this, as
> you
> potentially have frequent observations during the absorption
> phase.
> The number of compartments depends on the rate of absorption,
> sampling frequency, and (sometimes) method of analysis. So I
> would
> not worry too much, if you get something else than other reports
> in literature.
>
> 3) I would recommend running visual predictive checks for each
> group
> of patients to check, if you have adequate predictive
> performance in
> each group. The parameter variability model could be one reason,
> why you observe better individual fits but a worse objective
> function
> when you fix the volume. In addition, you could prepare some
> boxplots
> for the eta distributions in each group.
>
> 4) Did you try some of the models described by Dr. Nick Holford
> in his 1992
> cefetamet pivoxil paper? (J Pharmacokinet Biopharm. 1992;20:421-42.)
>
> 5) Sometimes, using a full variance covariance matrix for the
> absorption
> parameters improves the predictive performance during the
> absorption
> phase notably.
>
> Hope some of this will work (-:
> Best regards
> Juergen
>
>
> -----------------------------------------------
> Juergen Bulitta, PhD, Post-doctoral Fellow
> Pharmacometrics, University at Buffalo, NY, USA
> Phone: +1 716 645 2855 ext. 281, [EMAIL PROTECTED]
> -----------------------------------------------
>
>
>
>
"The method with interaction does not work due to the small sample size."
I do not think that this is correct. Sample size should not be an issue for FOCEI. If you have lag time with random effect, this could be a problem for FOCE, but not the sample size.
Leonid
[EMAIL PROTECTED] wrote:
> Hello Brenda,
>
> Your sample looks small. I have 38 subjects with rich data. FO method works, but plots for sume subjects do not look very good and there are few local maxima. The method with interaction does not work due to the small sample size. Although the prelimenary data are useful, nothing can make the plots perfect. I have to increase the sample size. Try to run the model for each subject without population estimates. It will give you an idea how the first order approcximation affects your plots. Try to play with the model of error. If your doses are very different, you may need to use CV+additive model. Try to implement correlation between Cl and V2, but set correlation between Ka and the other parameters to zero. Good luck,
>
> Pavel
>
Quoted reply history
> ----- Original Message -----
> From: Jurgen Bulitta
> Date: Tuesday, April 24, 2007 3:47 pm
>
> Subject: Re: [NMusers] Improved absorption profile by forcing volume of the central compartment
>
> To: "B.C.M. Winter - De" , [email protected]
>
> > Dear Brenda,
> >
> > Just a couple of comments and questions:
> > 1) Is there a specific reason why you are using FO for a dataset
> > with
> > frequent sampling? How large is your between subject
> > variability?
> > I would recommend considering FOCE+I in your case (for a more
> > detailed assessment, see e.g. Bauer R et al. AAPS J. 2007;9:E60-83.)
> >
> > 2) Have you tried a 3 compartment model? I would try this, as
> > you
> > potentially have frequent observations during the absorption
> > phase.
> > The number of compartments depends on the rate of absorption,
> > sampling frequency, and (sometimes) method of analysis. So I
> > would
> > not worry too much, if you get something else than other reports
> > in literature.
> >
> > 3) I would recommend running visual predictive checks for each
> > group
> > of patients to check, if you have adequate predictive
> > performance in
> > each group. The parameter variability model could be one reason,
> > why you observe better individual fits but a worse objective
> > function
> > when you fix the volume. In addition, you could prepare some
> > boxplots
> > for the eta distributions in each group.
> >
> > 4) Did you try some of the models described by Dr. Nick Holford
> > in his 1992
> > cefetamet pivoxil paper? (J Pharmacokinet Biopharm. 1992;20:421-42.)
> >
> > 5) Sometimes, using a full variance covariance matrix for the
> > absorption
> > parameters improves the predictive performance during the
> > absorption
> > phase notably.
> >
> > Hope some of this will work (-:
> > Best regards
> > Juergen
> >
> >
> > -----------------------------------------------
> > Juergen Bulitta, PhD, Post-doctoral Fellow
> > Pharmacometrics, University at Buffalo, NY, USA
> > Phone: +1 716 645 2855 ext. 281, [EMAIL PROTECTED]
> > -----------------------------------------------
> >
> >
> >
> >
Hello Leonid,
More complex NONMEM methods may need more observations. This is my expirience,
but I am glad to find out more about that.
Regards,
Pavel
Quoted reply history
----- Original Message -----
From: Leonid Gibiansky
Date: Monday, May 7, 2007 3:43 pm
Subject: Re: [NMusers] Improved absorption profile by forcing volume of the
central compartment
To: [EMAIL PROTECTED]
Cc: [email protected]
> "The method with interaction does not work due to the small
> sample size."
>
> I do not think that this is correct. Sample size should not be
> an issue for FOCEI. If you have lag
> time with random effect, this could be a problem for FOCE, but
> not the sample size.
> Leonid
>
> [EMAIL PROTECTED] wrote:
> > Hello Brenda,
> >
> > Your sample looks small.
> >
> > I have 38 subjects with rich data. FO method works, but plots
> for sume
> > subjects do not look very good and there are few local maxima.
> The
> > method with interaction does not work due to the small sample
> size.
> > Although the prelimenary data are useful, nothing can make the
> plots
> > perfect. I have to increase the sample size.
> >
> > Try to run the model for each subject without population
> estimates. It
> > will give you an idea how the first order approcximation
> affects your
> > plots.
> >
> > Try to play with the model of error. If your doses are very
> different,
> > you may need to use CV+additive model.
> >
> > Try to implement correlation between Cl and V2, but set
> correlation
> > between Ka and the other parameters to zero.
> >
> > Good luck,
> > Pavel
> >
> > ----- Original Message -----
> > From: Jurgen Bulitta
> > Date: Tuesday, April 24, 2007 3:47 pm
> > Subject: Re: [NMusers] Improved absorption profile by forcing
> volume of
> > the central compartment
> > To: "B.C.M. Winter - De" , [email protected]
> >
> > > Dear Brenda,
> > >
> > > Just a couple of comments and questions:
> > > 1) Is there a specific reason why you are using FO for a dataset
> > > with
> > > frequent sampling? How large is your between subject
> > > variability?
> > > I would recommend considering FOCE+I in your case (for a more
> > > detailed assessment, see e.g. Bauer R et al. AAPS J.
> 2007;9:E60-83.)
> > >
> > > 2) Have you tried a 3 compartment model? I would try this, as
> > > you
> > > potentially have frequent observations during the absorption
> > > phase.
> > > The number of compartments depends on the rate of absorption,
> > > sampling frequency, and (sometimes) method of analysis. So I
> > > would
> > > not worry too much, if you get something else than other reports
> > > in literature.
> > >
> > > 3) I would recommend running visual predictive checks for each
> > > group
> > > of patients to check, if you have adequate predictive
> > > performance in
> > > each group. The parameter variability model could be one reason,
> > > why you observe better individual fits but a worse objective
> > > function
> > > when you fix the volume. In addition, you could prepare some
> > > boxplots
> > > for the eta distributions in each group.
> > >
> > > 4) Did you try some of the models described by Dr. Nick Holford
> > > in his 1992
> > > cefetamet pivoxil paper? (J Pharmacokinet Biopharm.
> 1992;20:421-42.)
> > >
> > > 5) Sometimes, using a full variance covariance matrix for the
> > > absorption
> > > parameters improves the predictive performance during the
> > > absorption
> > > phase notably.
> > >
> > > Hope some of this will work (-:
> > > Best regards
> > > Juergen
> > >
> > >
> > > -----------------------------------------------
> > > Juergen Bulitta, PhD, Post-doctoral Fellow
> > > Pharmacometrics, University at Buffalo, NY, USA
> > > Phone: +1 716 645 2855 ext. 281, [EMAIL PROTECTED]
> > > -----------------------------------------------
> > >
> > >
> > >
> > >
>