Dear all,
First of all, I wanna say HAPPY NEW YEAR to
you guys. Best Wishes in this very beginning day of the "Horse
Year"(Chinese calendar lol )
Well,as titled,
here is the question:
I'm now doing external validation of the
PPK models pubished before by other groups with my own data. Ｉfind extreme
differences in MPE(mean prediction error), RMSE(root of mean squared prediction
error) and so on while using different estimation methods like FO, FOCE, or
FOCE-INTER, especially the one with interaction from the former 2.
I do understand the diversity in these 3
methods. But how much is the influence as to POSTHOC in external validation.
For example:
$ESTIMATION
MAXEVAL=0 METHOD=1 NOABORT
POSTHOC
Could anybody
explain to me and clarify the problem?
Great thanks!
Yours
Zhao Chenyan
Department of
Clinical Pharmacy, Huashan Hospital, Fudan University, Shanghai, P.R.China
Question: FO VS. FOCE VS. FOCE-INTER in POSTHOC 2014-1-3
Zhao,
For MAXEVAL=0, FOCE and FO should provide identical results (POSTHOC is
needed for FO; for FOCE it is not necessary but you can use it as well).
Concerning INTER, it should be used in all cases except when the error
is purely additive (and you can use it even for additive error, this
should not affect your solution). So if you need POSTHOC run, the
recommendation is to use
$ESTIMATION MAXEVAL=0 METHOD=1 INTER NOABORT POSTHOC
in all cases. For external validation, VPC-style diagnostics (as well as
regular model diagnostic plots) would be preferable to the overall
measures like MPE or RMSE that are more or less useless in identifying
the direction of the bias (if you have any differences with the
literature models).
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
Quoted reply history
On 1/4/2014 9:31 PM, 赵赵 wrote:
> Dear all,
>
> First of all, I wanna say HAPPY NEW YEAR to you guys. Best Wishes in
> this very beginning day of the "Horse Year"(Chinese calendar lol )
>
> Well,as titled, here is the question:
>
> I'm now doing external validation of the PPK models pubished before by
> other groups with my own data. Ｉfind extreme differences in MPE(mean
> prediction error), RMSE(root of mean squared prediction error) and so on
> while using different estimation methods like FO, FOCE, or FOCE-INTER,
> especially the one with interaction from the former 2.
>
> I do understand the diversity in these 3 methods. But how much is the
> influence as to POSTHOC in external validation. For example:
>
> $ESTIMATION MAXEVAL=0 METHOD=1 NOABORT POSTHOC
>
> Could anybody explain to me and clarify the problem?
>
> Great thanks!
>
> Yours
>
> Zhao Chenyan
>
> Department of Clinical Pharmacy, Huashan Hospital, Fudan University,
> Shanghai, P.R.China
>
Dear Dr Leonid，
Thanks so much! Your answer gave me a very clear direction of what to do next.
Still, I just wonder the reason,the principle behind it. Would you please give
me a further explanation?
Thanks in advance
Regards,Zhao
> Date: Sun, 5 Jan 2014 10:56:17 -0500
> From: [email protected]
> To: [email protected]; [email protected]
> Subject: Re: [NMusers] Question: FO VS. FOCE VS. FOCE-INTER in POSTHOC
> 2014-1-3
>
> Zhao,
>
> For MAXEVAL=0, FOCE and FO should provide identical results (POSTHOC is
> needed for FO; for FOCE it is not necessary but you can use it as well).
> Concerning INTER, it should be used in all cases except when the error
> is purely additive (and you can use it even for additive error, this
> should not affect your solution). So if you need POSTHOC run, the
> recommendation is to use
>
> $ESTIMATION MAXEVAL=0 METHOD=1 INTER NOABORT POSTHOC
>
> in all cases. For external validation, VPC-style diagnostics (as well as
> regular model diagnostic plots) would be preferable to the overall
> measures like MPE or RMSE that are more or less useless in identifying
> the direction of the bias (if you have any differences with the
> literature models).
>
> Leonid
>
>
> --------------------------------------
> Leonid Gibiansky, Ph.D.
> President, QuantPharm LLC
> web: www.quantpharm.com
> e-mail: LGibiansky at quantpharm.com
> tel: (301) 767 5566
>
>
>
Quoted reply history
> On 1/4/2014 9:31 PM, 赵赵 wrote:
> > Dear all,
> >
> > First of all, I wanna say HAPPY NEW YEAR to you guys. Best Wishes in
> > this very beginning day of the "Horse Year"(Chinese calendar lol )
> >
> > Well,as titled, here is the question:
> >
> > I'm now doing external validation of the PPK models pubished before by
> > other groups with my own data. Ｉfind extreme differences in MPE(mean
> > prediction error), RMSE(root of mean squared prediction error) and so on
> > while using different estimation methods like FO, FOCE, or FOCE-INTER,
> > especially the one with interaction from the former 2.
> >
> > I do understand the diversity in these 3 methods. But how much is the
> > influence as to POSTHOC in external validation. For example:
> >
> > $ESTIMATION MAXEVAL=0 METHOD=1 NOABORT POSTHOC
> >
> > Could anybody explain to me and clarify the problem?
> >
> > Great thanks!
> >
> > Yours
> >
> > Zhao Chenyan
> >
> > Department of Clinical Pharmacy, Huashan Hospital, Fudan University,
> > Shanghai, P.R.China
> >
Zhao,
For more details you can look at Nonmem help and manuals. POSTHOC step
of FO and MAXEVAL=0 of FOCE will do exactly the same computations (given
the population parameters, will compute individual parameters and
predictions). INTERACTION options allows a more precise interpretation
of the error model (from help: with INTER option, "The dependence on
etas of the model for intra-individual random error is preserved in the
computation of the objective function" while without it, the program
"set etas to 0 during the computation of the model for
intraindividual random error")
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
Quoted reply history
On 1/6/2014 6:53 AM, 赵赵 wrote:
> Dear Dr Leonid，
>
> Thanks so much! Your answer gave me a very clear direction of what to do
> next. Still, I just wonder the reason,the principle behind it. Would you
> please give me a further explanation?
>
> Thanks in advance
>
> Regards,
> Zhao
>
> > Date: Sun, 5 Jan 2014 10:56:17 -0500
> > From: [email protected]
> > To: [email protected]; [email protected]
> > Subject: Re: [NMusers] Question: FO VS. FOCE VS. FOCE-INTER in
> POSTHOC 2014-1-3
> >
> > Zhao,
> >
> > For MAXEVAL=0, FOCE and FO should provide identical results (POSTHOC is
> > needed for FO; for FOCE it is not necessary but you can use it as well).
> > Concerning INTER, it should be used in all cases except when the error
> > is purely additive (and you can use it even for additive error, this
> > should not affect your solution). So if you need POSTHOC run, the
> > recommendation is to use
> >
> > $ESTIMATION MAXEVAL=0 METHOD=1 INTER NOABORT POSTHOC
> >
> > in all cases. For external validation, VPC-style diagnostics (as well as
> > regular model diagnostic plots) would be preferable to the overall
> > measures like MPE or RMSE that are more or less useless in identifying
> > the direction of the bias (if you have any differences with the
> > literature models).
> >
> > Leonid
> >
> >
> > --------------------------------------
> > Leonid Gibiansky, Ph.D.
> > President, QuantPharm LLC
> > web: www.quantpharm.com
> > e-mail: LGibiansky at quantpharm.com
> > tel: (301) 767 5566
> >
> >
> >
> > On 1/4/2014 9:31 PM, 赵赵 wrote:
> > > Dear all,
> > >
> > > First of all, I wanna say HAPPY NEW YEAR to you guys. Best Wishes in
> > > this very beginning day of the "Horse Year"(Chinese calendar lol )
> > >
> > > Well,as titled, here is the question:
> > >
> > > I'm now doing external validation of the PPK models pubished before by
> > > other groups with my own data. Ｉfind extreme differences in MPE(mean
> > > prediction error), RMSE(root of mean squared prediction error) and
> so on
> > > while using different estimation methods like FO, FOCE, or FOCE-INTER,
> > > especially the one with interaction from the former 2.
> > >
> > > I do understand the diversity in these 3 methods. But how much is the
> > > influence as to POSTHOC in external validation. For example:
> > >
> > > $ESTIMATION MAXEVAL=0 METHOD=1 NOABORT POSTHOC
> > >
> > > Could anybody explain to me and clarify the problem?
> > >
> > > Great thanks!
> > >
> > > Yours
> > >
> > > Zhao Chenyan
> > >
> > > Department of Clinical Pharmacy, Huashan Hospital, Fudan University,
> > > Shanghai, P.R.China
> > >