Dear All,
With NM7.2/SAEM, the root.phi output file contains the conditional means
(and variances) of the individual phi's (mu(i)+eta(i)) - which is great!
I found that the these values can be different from what you get from a
$TABLE in more classical way:
$PRED
MU_1=THETA(1)
K=MU_1+ETA(1)
...
$TABLE K ...
For a some subjects, these are very different, and the ones coming from
the $TABLE are clearly less optimal, yielding a bad individual fit, as
judged from IPRE calculated in $PRED.
I guess one part of the solution is using CPRED and CPREDI - but
unfortunately I cannot use these as I (have to) use LIKELIHOOD in $EST
(M3 method for BLQ-data).
Kind regards,
Filip De Ridder
Janssen R&D, Beerse, Belgium.
NM7.2 SAEM with LIKE
Filip:
The $TABLE results are obtained from a "post-hoc" assessment at the best fit
eta values (modal, EBE), regardless of the method used. To evaluate $TABLE
parameters at the conditional mean positions, you may select FNLETA=0.
Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics, R&D
ICON Development Solutions
7740 Milestone Parkway
Suite 150
Hanover, MD 21076
Tel: (215) 616-6428
Mob: (925) 286-0769
Email: [email protected]
Web: http://www.iconplc.com/
Quoted reply history
________________________________
From: [email protected] [mailto:[email protected]] On
Behalf Of De Ridder, Filip [JRDBE]
Sent: Tuesday, August 07, 2012 10:18 AM
To: [email protected]
Subject: [NMusers] NM7.2 SAEM with LIKE
Dear All,
With NM7.2/SAEM, the root.phi output file contains the conditional means (and
variances) of the individual phi's (mu(i)+eta(i)) - which is great! I found
that the these values can be different from what you get from a $TABLE in more
classical way:
$PRED
MU_1=THETA(1)
K=MU_1+ETA(1)
...
$TABLE K ...
For a some subjects, these are very different, and the ones coming from the
$TABLE are clearly less optimal, yielding a bad individual fit, as judged from
IPRE calculated in $PRED.
I guess one part of the solution is using CPRED and CPREDI - but unfortunately
I cannot use these as I (have to) use LIKELIHOOD in $EST (M3 method for
BLQ-data).
Kind regards,
Filip De Ridder
Janssen R&D, Beerse, Belgium.
Using these estimates from the phi file may not always be possible/easy.
One example is with between-occasion variability (BOV) on PK
parameter(s). The .phi file will provide one single phi value while the
individual PK parameter will be different between occasions.
Regards
Navin Goyal
Quoted reply history
On Tue, Aug 7, 2012 at 6:30 PM, Bauer, Robert <[email protected]>wrote:
> **
> Filip:
> In your case it may be that conditional means are more appropriate than
> conditional modes, but I would not necessarily conclude that in general.
> I do not know what Monolix reports regarding individual parameters.
>
>
> Robert J. Bauer, Ph.D. ** ****
>
> Vice President, Pharmacometrics, R&D****
>
> ICON Development Solutions
>
> 7740 Milestone Parkway
>
> Suite 150
>
> Hanover, MD 21076****
>
> Tel: (215) 616-6428
>
> Mob: (925) 286-0769****
>
> Email: [email protected]****
>
> Web: www.iconplc.com
>
>
> ------------------------------
> *From:* De Ridder, Filip [JRDBE] [mailto:[email protected]]
> *Sent:* Tuesday, August 07, 2012 3:08 PM
> *To:* Bauer, Robert; [email protected]
> *Subject:* RE: NM7.2 SAEM with LIKE
>
>
>
> Hi Bob,
>
> Thanks for the clarification. Would you agree that the conditional means
> are more trustworthy, in case there is a relevant difference? In my
> specific dataset, there are 5 subjects for whom the conditional means yield
> a much better individual fit, than the "post-hoc" eta's.
> I believe that in Monolix, the individual parameter estimates are the
> conditional means, no? In my example
>
> Kind regards,
>
> Filip
> -----Oorspronkelijk bericht-----
> Van: Bauer, Robert [mailto:[email protected]<[email protected]>
> ]
> Verzonden: di 7/08/2012 18:28
> Aan: De Ridder, Filip [JRDBE]; [email protected]
> Onderwerp: RE: NM7.2 SAEM with LIKE
>
> Filip:
> The $TABLE results are obtained from a "post-hoc" assessment at the best
> fit eta values (modal, EBE), regardless of the method used. To evaluate
> $TABLE parameters at the conditional mean positions, you may select
> FNLETA=0.
>
> Robert J. Bauer, Ph.D.
> Vice President, Pharmacometrics, R&D
> ICON Development Solutions
> 7740 Milestone Parkway
> Suite 150
> Hanover, MD 21076
> Tel: (215) 616-6428
> Mob: (925) 286-0769
> Email: [email protected]
> Web: http://www.iconplc.com/
>
>
> ________________________________
> From: [email protected]
> [mailto:[email protected]<[email protected]>]
> On Behalf Of De Ridder, Filip [JRDBE]
> Sent: Tuesday, August 07, 2012 10:18 AM
> To: [email protected]
> Subject: [NMusers] NM7.2 SAEM with LIKE
>
> Dear All,
>
> With NM7.2/SAEM, the root.phi output file contains the conditional means
> (and variances) of the individual phi's (mu(i)+eta(i)) - which is great!
> I found that the these values can be different from what you get from a
> $TABLE in more classical way:
>
> $PRED
> MU_1=THETA(1)
> K=MU_1+ETA(1)
> ...
> $TABLE K ...
>
> For a some subjects, these are very different, and the ones coming from
> the $TABLE are clearly less optimal, yielding a bad individual fit, as
> judged from IPRE calculated in $PRED.
>
> I guess one part of the solution is using CPRED and CPREDI - but
> unfortunately I cannot use these as I (have to) use LIKELIHOOD in $EST (M3
> method for BLQ-data).
>
>
> Kind regards,
>
> Filip De Ridder
> Janssen R&D, Beerse, Belgium.
>
>
>
>
Hi Bob,
Apparently, Monolix offers two options side by side: mean or mode.
However, is there not another difference here: root.phi gives the conditional
mean of the sampling-based posterior density, whereas using FNLETA gives you
the "classical" EBE, i.e. not sampling based?
It seems a bit contradictory that you would use the sampled-based densities to
integrate out the ETA's in SAEM to obtain the pop pars, and then not use the
sampled-based densities to get the ETA's?
Any thoughts are welcome.
Kind regards,
Filip
-----Oorspronkelijk bericht-----
Quoted reply history
Van: Bauer, Robert [mailto:[email protected]]
Verzonden: wo 8/08/2012 0:30
Aan: De Ridder, Filip [JRDBE]; [email protected]
Onderwerp: RE: NM7.2 SAEM with LIKE
Filip:
In your case it may be that conditional means are more appropriate than
conditional modes, but I would not necessarily conclude that in general.
I do not know what Monolix reports regarding individual parameters.
Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics, R&D
ICON Development Solutions
7740 Milestone Parkway
Suite 150
Hanover, MD 21076
Tel: (215) 616-6428
Mob: (925) 286-0769
Email: [email protected]
Web: http://www.iconplc.com/
________________________________
From: De Ridder, Filip [JRDBE] [mailto:[email protected]]
Sent: Tuesday, August 07, 2012 3:08 PM
To: Bauer, Robert; [email protected]
Subject: RE: NM7.2 SAEM with LIKE
Hi Bob,
Thanks for the clarification. Would you agree that the conditional means are
more trustworthy, in case there is a relevant difference? In my specific
dataset, there are 5 subjects for whom the conditional means yield a much
better individual fit, than the "post-hoc" eta's.
I believe that in Monolix, the individual parameter estimates are the
conditional means, no? In my example
Kind regards,
Filip
-----Oorspronkelijk bericht-----
Van: Bauer, Robert [mailto:[email protected]]
Verzonden: di 7/08/2012 18:28
Aan: De Ridder, Filip [JRDBE]; [email protected]
Onderwerp: RE: NM7.2 SAEM with LIKE
Filip:
The $TABLE results are obtained from a "post-hoc" assessment at the best fit
eta values (modal, EBE), regardless of the method used. To evaluate $TABLE
parameters at the conditional mean positions, you may select FNLETA=0.
Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics, R&D
ICON Development Solutions
7740 Milestone Parkway
Suite 150
Hanover, MD 21076
Tel: (215) 616-6428
Mob: (925) 286-0769
Email: [email protected]
Web: http://www.iconplc.com/
________________________________
From: [email protected] [mailto:[email protected]] On
Behalf Of De Ridder, Filip [JRDBE]
Sent: Tuesday, August 07, 2012 10:18 AM
To: [email protected]
Subject: [NMusers] NM7.2 SAEM with LIKE
Dear All,
With NM7.2/SAEM, the root.phi output file contains the conditional means (and
variances) of the individual phi's (mu(i)+eta(i)) - which is great! I found
that the these values can be different from what you get from a $TABLE in more
classical way:
$PRED
MU_1=THETA(1)
K=MU_1+ETA(1)
...
$TABLE K ...
For a some subjects, these are very different, and the ones coming from the
$TABLE are clearly less optimal, yielding a bad individual fit, as judged from
IPRE calculated in $PRED.
I guess one part of the solution is using CPRED and CPREDI - but unfortunately
I cannot use these as I (have to) use LIKELIHOOD in $EST (M3 method for
BLQ-data).
Kind regards,
Filip De Ridder
Janssen R&D, Beerse, Belgium.
Hi All,
I dont think it is contradictory to use the sample-based densities for
integration and then use the classical EBE for reporting individual values.
When integrating you want to see the entire individual density so you can give
correct weight to large areas of low probability. But when you are reporting
individual values you dont really care about large areas of low probability,
you only want the “most likely” parameters for an individual, this is classical
EBEs or the (approximated) mode of the distribution.
The mean of this density is numerically useful but doesnt have a easy to
understand interpretation (for me at least).
warm regards,
Douglas Eleveld
Quoted reply history
________________________________________
From: [email protected] [[email protected]] on behalf of
De Ridder, Filip [JRDBE] [[email protected]]
Sent: Wednesday, August 08, 2012 10:06 PM
To: Bauer, Robert; [email protected]
Subject: [NMusers] RE: NM7.2 SAEM with LIKE
Hi Bob,
Apparently, Monolix offers two options side by side: mean or mode.
However, is there not another difference here: root.phi gives the conditional
mean of the sampling-based posterior density, whereas using FNLETA gives you
the "classical" EBE, i.e. not sampling based?
It seems a bit contradictory that you would use the sampled-based densities to
integrate out the ETA's in SAEM to obtain the pop pars, and then not use the
sampled-based densities to get the ETA's?
Any thoughts are welcome.
Kind regards,
Filip
-----Oorspronkelijk bericht-----
Van: Bauer, Robert [mailto:[email protected]]
Verzonden: wo 8/08/2012 0:30
Aan: De Ridder, Filip [JRDBE]; [email protected]
Onderwerp: RE: NM7.2 SAEM with LIKE
Filip:
In your case it may be that conditional means are more appropriate than
conditional modes, but I would not necessarily conclude that in general.
I do not know what Monolix reports regarding individual parameters.
Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics, R&D
ICON Development Solutions
7740 Milestone Parkway
Suite 150
Hanover, MD 21076
Tel: (215) 616-6428
Mob: (925) 286-0769
Email: [email protected]
Web: http://www.iconplc.com/
________________________________
From: De Ridder, Filip [JRDBE] [mailto:[email protected]]
Sent: Tuesday, August 07, 2012 3:08 PM
To: Bauer, Robert; [email protected]
Subject: RE: NM7.2 SAEM with LIKE
Hi Bob,
Thanks for the clarification. Would you agree that the conditional means are
more trustworthy, in case there is a relevant difference? In my specific
dataset, there are 5 subjects for whom the conditional means yield a much
better individual fit, than the "post-hoc" eta's.
I believe that in Monolix, the individual parameter estimates are the
conditional means, no? In my example
Kind regards,
Filip
-----Oorspronkelijk bericht-----
Van: Bauer, Robert [mailto:[email protected]]
Verzonden: di 7/08/2012 18:28
Aan: De Ridder, Filip [JRDBE]; [email protected]
Onderwerp: RE: NM7.2 SAEM with LIKE
Filip:
The $TABLE results are obtained from a "post-hoc" assessment at the best fit
eta values (modal, EBE), regardless of the method used. To evaluate $TABLE
parameters at the conditional mean positions, you may select FNLETA=0.
Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics, R&D
ICON Development Solutions
7740 Milestone Parkway
Suite 150
Hanover, MD 21076
Tel: (215) 616-6428
Mob: (925) 286-0769
Email: [email protected]
Web: http://www.iconplc.com/
________________________________
From: [email protected] [mailto:[email protected]] On
Behalf Of De Ridder, Filip [JRDBE]
Sent: Tuesday, August 07, 2012 10:18 AM
To: [email protected]
Subject: [NMusers] NM7.2 SAEM with LIKE
Dear All,
With NM7.2/SAEM, the root.phi output file contains the conditional means (and
variances) of the individual phi's (mu(i)+eta(i)) - which is great! I found
that the these values can be different from what you get from a $TABLE in more
classical way:
$PRED
MU_1=THETA(1)
K=MU_1+ETA(1)
...
$TABLE K ...
For a some subjects, these are very different, and the ones coming from the
$TABLE are clearly less optimal, yielding a bad individual fit, as judged from
IPRE calculated in $PRED.
I guess one part of the solution is using CPRED and CPREDI - but unfortunately
I cannot use these as I (have to) use LIKELIHOOD in $EST (M3 method for
BLQ-data).
Kind regards,
Filip De Ridder
Janssen R&D, Beerse, Belgium.