Back to the Top
Hi All,
I have a general question for number of data points and parameters to
be estimated for a model. Are there some rules for having minimum
number of data points to estimate the number of parameters?
Thank you all.
Vipul
Vipul Kumar
Post Doc fellow
Pharmaceutics
University of Florida
Gainesville-FL (USA)
Back to the Top
Dear Vipul,
"The more the merrier" (or better), is a good motto. Our PK Solutions
pharmacokinetics data analysis software [ see www.SummitPK.com ]
requires 5 sample points as a minimum, but this would put a strain on
any attempt to calculate parameters following an extravascular dose.
If the analyses are accurate and the individual subject variation is
minimal, 6+ samples can get you 75 PK parameters using PK Solutions.
But that should be considered only a pilot experiment to work out a
better and more complete sampling protocol.
David S. Farrier, Ph.D.
Summit Research Services
/\ /\
SummitPK.com /\ / \ /\ / \
/ / / /\ / \
==
David S. Farrier, Ph.D. Phone: 970-249-1389
Summit Research Services Fax: 970-249-1360
68911 Open Field Dr. Email: DFarrier.-a-.SummitPK.com
Montrose, CO 81401 Web: http://www.SummitPK.com
Back to the Top
The following message was posted to: PharmPK
Vipul
> I have a general question for number of data points and parameters to
> be estimated for a model. Are there some rules for having minimum
> number of data points to estimate the number of parameters?
Yes and no :-)
Can you make your question more specific? If so then I may be able to
help more...
If this a single subject or a population design? If it is single
subject
then the minimum number of samples per subject = the number of fixed
effects parameters. If you are estimating residual variance then you
will
need another sample if the variance is additive.
If this is a population design - then it is more complex and I
suggest you
have a look at: Duffull et al. JPKPD 2005; 32:441-457.
Cheers
Steve
--
Professor Stephen Duffull
School of Pharmacy
University of Otago
PO Box 913
Dunedin
P 03 479 5044
F 03 479 7034
Back to the Top
The following message was posted to: PharmPK
Vipul
> I have a general question for number of data points and parameters to
> be estimated for a model. Are there some rules for having minimum
> number of data points to estimate the number of parameters?
Yes and no :-)
Can you make your question more specific? If so then I may be able to
help more...
If this a single subject or a population design? If it is single
subject
then the minimum number of samples per subject = the number of fixed
effects parameters. If you are estimating residual variance then you
will
need another sample if the variance is additive.
If this is a population design - then it is more complex and I
suggest you
have a look at: Duffull et al. JPKPD 2005; 32:441-457.
Cheers
Steve
--
Professor Stephen Duffull
School of Pharmacy
University of Otago
PO Box 913
Dunedin
P 03 479 5044
F 03 479 7034
Back to the Top
Dear Prof Duffull and David S. Farrier,
Thank you for your response. To be more specific, my question
concerns estimating PK parameters for small number of subject (n=13)
and I am interested to calculate parameters for each individual
separately. This is because of the different PK profiles (single peak
vs Double peak) observed for different individuals.
I have around 9-11 data poionts per individual, however when I tried
to estimate model parameters using WinNonlin user define model for
two absorption sites, sometimes I got very high %CV for parameters
and at occassions WinNonlin crashed and I had to reinstall it again
to use it. So I was wondering was it because I do not have enough
data points given the number of parameters (6 in case of a two
absorption site model) or am I still missing something here!!!!
Thank you
Vipul
Vipul Kumar
Post Doc Associate
Dept of Pharmaceutics
University of Florida
Gainesville, FL-32601
Back to the Top
ear Vipul Kumar,
As my understanding, the more data points maybe could not solve
your problem. The more data points give the more exact description
about the model. But if the truth is that there is high CV% for the
parameters of the hypothetic model on this data set, the more data
points could do nothing.
The technique to estimate parameters is to minimize least square
(or some other function) as a target function. The limination of data
points comes not only from statistical part but also numerical part.
Yours
Ma Guangli
Back to the Top
Dear groupmembers,
Is there any guideline or published paper which states the minimum
number of
concentration per subject required to obtain PK parameter like Cmax,
AUCt,
AUCi, t1/2,Kel,Tmax etc efficiently?
Please help me out in this....
thanks in advance.
Regards,
Dipesh Jayswal
Biostatistician,
BioArc Research Solution,
Vadodara
Back to the Top
The following message was posted to: PharmPK
Dear David,
you wrote:
>If the analyses are accurate and the individual subject variation is
>minimal, 6+ samples can get you 75 PK parameters using PK Solutions.
>
Maybe I'm just old-fashioned, but I always thought that:
degrees of freedom = number of data points - number of parameters
Maybe you can force your software--I cannot do that with mine--
to calculate 75 parameters from 6 samples, but most likely only
with a confidence interval of -infinity to +infinity ;-)
best regards,
Helmut
--
Helmut Schuetz
BEBAC
Consultancy Services for Bioequivalence and Bioavailability Studies
Neubaugasse 36/11
1070 Vienna/Austria
tel/fax +43 1 2311746
Web http://BEBAC.at
BE/BA Forum http://forum.bebac.at
Back to the Top
The following message was posted to: PharmPK
Hi Vipul
I am not aware of any particular designs that are so bad that would
require
you to have to reinstall your software ...
But without knowing more, it seems odd that you would need to have
more than
9-11 data points per individual in order to estimate the parameters
for a 6
parameter model, unless the sampling times were taken at very inoptimal
times.
Since you only have 13 individuals it might be difficult to get a
population
approach to run and get estimates of the variance of between subject
effects
for all parameters. However, it is probable that patients will share
some
common components of the PK profile, e.g. perhaps they have the same
structural aspects to their disposition model, and hence a population
approach might be helpful.
In the meantime, I suggest you quantify the informativeness of your
design
by either computing the asymptotic standard errors for your design
using say
ADAPT II or performing simulation-estimations to get empirical
estimates of
the standard errors. If you wanted to look at the informativeness of
your
design for a population setting then you could use POPT.
Regards
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 913 Dunedin
New Zealand
E: stephen.duffull.-at-.otago.ac.nz
P: +64 3 479 5044
F: +64 3 479 7034
Back to the Top
The following message was posted to: PharmPK
Dear David,
you wrote:
>If the analyses are accurate and the individual subject variation is
>minimal, 6+ samples can get you 75 PK parameters using PK Solutions.
>
Maybe I'm just old-fashioned, but I always thought that:
degrees of freedom = number of data points - number of parameters
Maybe you can force your software--I cannot do that with mine--
to calculate 75 parameters from 6 samples, but most likely only
with a confidence interval of -infinity to +infinity ;-)
best regards,
Helmut
--
Helmut Schuetz
BEBAC
Consultancy Services for Bioequivalence and Bioavailability Studies
Neubaugasse 36/11
1070 Vienna/Austria
tel/fax +43 1 2311746
Web http://BEBAC.at
BE/BA Forum http://forum.bebac.at
Back to the Top
Dear Dipesh Jayswal,
I believe that there is no a minimum number for all of questions.
Compartment and noncompartment analysis are very different.
As my understanding, that 9 data points as a minimum value are
chosen in noncompartment analysis is because that there should be 3
data points in absorption, distribution, and elimation phase,
respectively to estimate the parameters,
To compartment analysis, the minimum number of data points is
according to your model.
Population PK analysis give a minimum number of data points anyway.
Yours
Ma Guangli
Back to the Top
The following message was posted to: PharmPK
Helmut
Theoretically parameters to be estimated cannot exceed data points.
It's right. But the condition is that each parameters should be
orthogonal
(independent to each other).
For PK parameters they are usually recalculated by others like ke =
CL/Vd as
you know.
Most of 75 PK parameters that PK sulutions put out are such ones, I
think.
Kyun-Seop Bae MD PhD
Assistant Professor
Department of Clinical Pharmacology & Therapeutics
Asan Medical Center, University of Ulsan
Tel: +82-2-3010-4611
Fax: +82-2-3010-4623
Mobile: +82-17-253-5048
Email: ksbae.at.amc.seoul.kr
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)