Hi Norman,
Priors is the summary of the data. You always loose something when you
summarize the data. Sometime you have to use summaries because data are
not available or when you want to support some parts of the model while
allowing more flexibility to the other parts (for example, when you use
adult priors + allometric scaling to support pediatric model). In other
cases, it is better to use raw data without intermediate steps.
One (artificial but still valid) example is the nonlinear drug. At each
dose level, you could be able to describe it by the linear model. But if
you have data from several dose levels, you can develop the nonlinear
model. Now assume that you have several (say 3) separate studies, each
at one dose level. You will have 3 different linear models, each with
different parameters, and with some uncertainty. I am not sure how to
get a meaningful model out of these data summaries. Similarly, if you
have different populations, or different distribution/representation of
covariates, or different sampling scheme - I think it would be more
beneficial to have all these diverse data in one set that to transfer
information through the parameters.
Still, each particular case should be approached individually, I would
not rule out any of the approaches (priors vs combining the data)
without knowing details of the specific case.
Thanks
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
On 10/14/2011 10:45 PM, Norman Z wrote:
Hi Leonid,
Would you please explain why combine the data may be a better option
than using priors? I've met similar situations.
Thanks,
Norman
On Fri, Oct 14, 2011 at 5:01 PM, Leonid Gibiansky
<[email protected] <mailto:[email protected]>> wrote:
Nele,
Nonmem 7.2 has UNCONDITIONAL option in cov step that would compute
SEs even for problems with error in minimization.
Also, it is perfectly OK to have variance-covariance computed from
bootstrap; some may even argue that it is better than use Nonmem cov
matrix.
And finally (just an opinion) I would rather combine the data than
use priors.
Thanks
Leonid
------------------------------__--------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com <http://www.quantpharm.com/>
e-mail: LGibiansky at quantpharm.com <http://quantpharm.com/>
tel: (301) 767 5566 <tel:%28301%29%20767%205566>
On 10/13/2011 1:31 AM, Kaessner, Nele wrote:
Dear all,
I am currently in a planning phase for a PopPK evaluation for an
upcoming study. I have already developed a PopPK model for this
compound, and the idea would now be to fit the available model
to the
new data. One idea was to do this via the PRIOR functionality in
NONMEM,
so that I do not have to mix the old data into the new data set.
So now
I wanted to test the performance of this approach via
simulations, and
have the following problem:
My previously developed final model did not converge successfully
(terminated with rounding errors), so I do not have a covariance
matrix
of estimates to describe the uncertainty in the THETAs. However I
obtained confidence intervals from a nonparametric bootstrap,
and I was
wondering if these results can somehow be used instead for the prior
information.
In some internal discussions we had the idea to just use all the
parameter estimates from the performed 1000 bootstrap runs, and just
calculate variance and covariance of THETAs from these.
I would highly appreciate your comments on the validity of such an
approach, or suggestions of other alternatives.
Thanks and best regards
Nele
__________________________________________________________________
Dr. Nele Kنكner
PK/PD Sciences
Modelling and Simulation (RDP/MS)
*Nycomed: A Takeda Company*
Nycomed GmbH
Byk-Gulden-Str. 2
D-78467 Konstanz, Germany
Fon: (+49) 7531 / 84 - 4759
<tel:%28%2B49%29%207531%20%2F%2084%20-%204759>
Fax: (+49) 7531 / 84 - 94759
<tel:%28%2B49%29%207531%20%2F%2084%20-%2094759>
mailto: [email protected]
<mailto:[email protected]>
<mailto:nele.kaessner@nycomed.__com
<mailto:[email protected]>>
http://www.nycomed.com <http://www.nycomed.com/>
<http://www.nycomed.com/>
County Court: Freiburg, Commercial Register HRB 701257
Chairman Supervisory Board: Charles Depasse (Speaker)
Management Board: Gilbert Rademacher (Speaker),
Konstantin von Alvensleben, Dr. Rainer Wiartalla
------------------------------__------------------------------__--------
The content of this email and of any files transmitted may contain
confidential, proprietary or legally privileged information and is
intended solely for the use of the person/s or entity/ies to
whom it is
addressed. If you have received this email in error you have no
permission whatsoever to use, copy, disclose or forward all or
any of
its contents. Please immediately notify the sender and
thereafter delete
this email and any attachments.
------------------------------__------------------------------__--------
