Back to the Top
Hello to the List,
Is there any other software besides ADAPT II that will help in the
selection of sampling points?
Susan E. Shoaf
[Have a look at
http://www.boomer.org/course/pk_bio/Ch9906c/sld004.htm and the next
few slides. Optimal sampling calculations look like sensitivity
analysis?? Also interesting is what happens when you add weights to
the data points - db]
Back to the Top
There is a program which does optimal sampling produced by Alfredo Ruggeri
at the University of Padova in Italy. Part of this work was supported by
our NIH Resource Grant that produced SAAM II. I suggest you contact Dr.
Ruggeri for more information.
Back to the Top
[Two replies - db]
X-Sender: jmlanao.aaa.gugu.usal.es (Unverified)
Date: Mon, 07 Feb 2000 11:35:35 +0100
To: PharmPK.at.boomer.org
From: jml
Subject: Re: PharmPK Re: Optimal sampling points
There is another program called DRUGTEST released by MEDISOFT for optimal
design of experiments in pharmacokinetics using Fisher Matrix information
and probability density functions of sampling times.
J.M.Lanao
Dpt. Pharmacy.
Univ. Salamanca. Spain
---
X-Originating-IP: [146.186.229.23]
From: "Vinay Desai"
To: PharmPK.at.boomer.org
Subject: Re: PharmPK Re: Optimal sampling points
Date: Mon, 07 Feb 2000 14:50:22 EST
WinNonlin allows you to plot various partial derivatives as part of
its modeling features. You can use these partial derivative plots
for optimizing time points.
Vinay Desai
Back to the Top
From: exfamadu.aaa.savba.savba.sk
To: PharmPK.-a-.boomer.org
Date: Tue, 8 Feb 2000 10:28:11 +0100
Subject: Re: PharmPK Re: Optimal sampling points
X-Confirm-Reading-To: exfamadu.-at-.savba.savba.sk
X-pmrqc: 1
Priority: normal
> few slides. Optimal sampling calculations look like sensitivity
> analysis?? Also interesting is what happens when you add weights to
> the data points - db]---
No, optimal sampling calculations do not look like sensitivity analysis.
With best regards,
Maria Durisova
Dipl. Engineer Maria Durisova D.Sc.
Senior Research Worker
Scientific Secretary
Institute of Experimental Pharmacology
Slovak Academy of Sciences
http://nic.savba.sk/sav/inst/exfa/advanced.htm
SK-842 16 Bratislava
Slovak Republic
---
Date: Tue, 08 Feb 2000 13:38:02 -0500
From: Mark Lovern
Organization: Pharsight, Inc.
X-Accept-Language: en
To: PharmPK.aaa.boomer.org
Subject: [Fwd: [Fwd: Re: PharmPK Re: Optimal sampling points]]
> Dear Susan:
>
> The optimization of a sampling regimen can be a very complicated issue.
> What software package will suit your needs really depends upon how much
> information you wish to incorporate into your analysis. For instance,
> you may choose to base your sampling design based on simulations using a
> fixed set of parameter values that are assumed to be typical for your
> study population. This is an approach that can be implemented quite
> easily in WinNonlin. Other approaches based on simulation are discussed
> below.
>
> WinNonlin's simulation output includes both variance inflation factors
> (VIFs) as well as partial derivatives of the concentration function with
> respect to model parameters. Variance inflation factors provide a
> relative measure of how precisely parameters may be estimated using a
> particular sampling regimen. A parameter's VIF value may be compared
> across competing sampling designs. The design which minimizes the
> parameter's VIF will maximize the precision with which the parameter is
> estimated.
>
> The sensitivity of the dependent variable (ie plasma concentration) to
> changes in parameter values at particular times may be assessed using
> WinNonlin's partial derivative output. The sensitivity of the dependent
> variable to a particular parameter value increases as that parameter's
> partial derivative deviates from zero. Sampling in regions of greater
> sensitivity will result in more precise estimates for the parameter, and
> plots of partial derivatives over time can be used to suggest what
> sampling times will be most strategic for estimating model parameters.
>
> For a discussion of how model simulations may be employed to optimize
> study designs, you may want to read pp. 310-318 of Gabrielsson and
> Weiner's Pharmacokinetic and Pharmacodynamic Data Analysis 2nd Ed.
> (Swedish Pharmaceutical Press, 1997). If you do not have this reference
> available, I would be happy to FAX you a copy of these pages. I have
> found this example is particularly helpful, because the authors discuss
> how to select a sampling regime that will provide good information even
> when one is unsure of what model underlies the data. (Many sample time
> selection algorithms require a priori selection of a model.)
>
> While the "average value" approach to study design may convey useful
> information, it is rather simplistic and fails to consider any number of
> factors that may be important in determining what sampling design is
> "truly" optimal. Such factors include inter-subject variability in PK
> parameters, covariate relationships, and the error structure of the
> bioanalytical assay. Simulation approaches to study design that
> incorporate such information are collectively referred to as
> computer-aided trial design (CATD). Pharsight offers trial simulation
> software that is specifically designed to aid in the implementation of
> the CATD approach.
>
> The advantage of CATD is that it allows protocol designers to answer
> questions that cannot be answered by the "average value" approach to
> study design. When one uses "average values", one is able to determine
> what sampling times will minimize the uncertainty in parameter
> estimates, provided the true parameter values are somewhere in the
> neighborhood of the values used to perform the simulation. The question
> that is not answered by such an approach is how many of your study
> subjects actually have parameter values that fall within this
> neighborhood. With the CATD approach, one has the ability to answer
> questions such as "What (if any) sampling regimen will allow Clearance
> to be estimated with less than a 15% CV for all subjects?" It is such
> questions that are truly of interest when designing a study.
>
> If you have any questions regarding our software and how it may be
> applied to your problem, please feel free to contact me directly (email:
> mlovern.-at-.pharsight.com; phone: (919) 859-6868 ext. 4007).
>
> Best Regards,
Mark Lovern
Consulting Scientist
Pharsight, Inc.
Phone: (919) 859-6868 ext. 4007
FAX: (919) 859-6871
Back to the Top
There is no such thing as optimal design of experiments in PK/PD. PK/PD
analysis is a methodology that is insensitive, when used properly, to the
data set. The tools you refer to can only assist in identifying the
boundary values within which the analytical tools converge on a single
solution. Mathematical transformations do not necessarily simply
information---sometimes they add complexities that render the original data
even more difficult to understand.
Daro Gross
Back to the Top
[Quite a few replies - hot topic! - db]
From: "Stephen Duffull"
To:
Subject: RE: PharmPK Re: Optimal sampling points
Date: Thu, 24 Feb 2000 15:57:07 +1000
X-Priority: 3 (Normal)
Importance: Normal
Daro
> There is no such thing as optimal design of
> experiments in PK/PD.
I am either missing the point - or I disagree with your
assertion. As I'm sure you're aware there are a number of
papers discussing optimal design of PK experiments both for
individuals, and more recently for populations (eg D'Argenio
JPB 1981;9:739-56 for individuals and Mentre et al
Biometrika 1997;84:429-442 for a population). Both papers
give an idea of how designs can be optimised - whether the
optimisation procedure will always locate the true optimum
is, however, up for debate.
> PK/PD
> analysis is a methodology that is insensitive,
> when used properly, to the
> data set.
Again - I can't agree. The standard errors of the parameter
estimates (computed from the square root of the diagonal
elements of the inverse of the Fisher information matrix)
are directly influenced by the design for non-linear mixed
effect models. Indeed it can be shown that as the number of
subjects in a population design increases the determinant of
the population Fisher information matrix increases.
On a more general note I would be worried if a PKPD analysis
was insensitive to the data - indeed this implies that the
model was selected independently of the data!
Regards
Steve
=================
Stephen Duffull
School of Pharmacy
University of Queensland
Brisbane, QLD 4072
Australia
Ph +61 7 3365 8808
Fax +61 7 3365 1688
---
From: "Della Pasqua, Oscar"
To: "'PharmPK.at.boomer.org'"
Subject: Re: Optimal sampling points
Date: Thu, 24 Feb 2000 08:22:16 -0000
Daro,
I wished you were right. I suggest you review the literature
(e.g.Park et al. J Pharmacokinet Biopharm 1998 Aug;26(4):471-92;
Gieschke et al. Int J Clin Pharmacol Ther 1997 Oct;35(10):469-74;
Palmer et al. Stat Med 1998 Jul 30;17(14):1613-22).
Oscar E. Della Pasqua
====================
Dr. O. E. Della Pasqua
Full Development Group
Worldwide Clinical Pharmacology
GlaxoWellcome Research & Developmet
* ++ 44 208 966 2404
fax ++ 44 208 966 2123
* odp72514.at.glaxowellcome.co.uk
====================
---
From: GLDrusano.-a-.aol.com
Date: Thu, 24 Feb 2000 10:23:08 EST
Subject: Re: PharmPK Re: Optimal sampling points
To: PharmPK.-a-.boomer.org
Such an arrogant resonse for an area with a huge literature that says exactly
the opposite and for which there is clinical validation!
George Drusano
---
X-Sender: n7950211.at.popin.ncl.ac.uk
Date: Thu, 24 Feb 2000 15:25:48 +0000
To: PharmPK.at.boomer.org
From: James
Subject: Re: Optimal sampling points
Dear Dr Gross,
When you say that there is no optimal design, do you mean that there are
many designs which give similar practical results? This may be true, but
there are still issues in how to design such experiments which may be
termed "optimal design". Personally, I dislike the word optimal because it
implies perfection and exists only in a perfectly defined sytem (however
that excludes essentially all real experiments). One could perhaps argue
that in any nonlinear system it is impossible to design an optimal
experiment without knowing what you want to find out in advance.
Much as I wish it was true PKPD modelling is not insensitive to the design
of the experiment. It is trivial to design a bad experiment.
A sensibly applied transformation (by which I take it you mean a mapping of
data onto summary parameters, using assumptions) should illuminate the key
features of the data - if it does not it is pointless.
James Wright
---
X-Sender: jelliffe.-a-.hsc.usc.edu
Date: Thu, 24 Feb 2000 09:58:03 -0800
To: PharmPK.at.boomer.org
From: Roger Jelliffe
Subject: Re: PharmPK Re: Optimal sampling points
Dear Daro:
Thanks for your note. It would be very interesting for me,
for example, to
understand better why you say what you say. For example, what do you mean
that PK/PD analysis is a method WHEN USED PROPERLY, that is insensitive to
the data set? It would be very interesting if you can expand on this
thought and support it, and your other thoughts, further. Further, how does
this relate to D- optimality in experimental design, for example?
Very best regards,
Roger Jelliffe
---
Date: Thu, 24 Feb 2000 13:59:07 -0500
From: "James Cherry"
To:
Cc:
Subject: PharmPK Re: Optimal sampling points
Thank you. I don't know who Daro Gross is, but he has restated things
that are well known and he has a firm grasp of the obvious. A cursory
reading of Belsley, Kuh and Welsch "Regression Diagnostics-
Identifying Influential Data and Sources of Collinearity" will show
the bias of his statement of the value of data transformations in
many cases. I have this book if you wish to read it. His statement
that transformations may make the info more complex is a one-sided
viewpoint which is already well known, that is why multiple
transformations are reviewed, and the ones that improve data
interpretation and prediction are kept and ones that make it more
meaningless are rejected. Duh. He is only partially right that our
tools only assist in identifying the boundary values( boundary values
are not easily determined from some non-lin models), but remember,
this is precisely what we are trying to do!
Remember, even the simplest aminoglycoside programs use log
transformation of the serum values to change the non-linear,
difficult to interpret picture of drug elimination to a simple,
straight linear graph. This is one obvious example of the use of
transformations to improve data interpretation. The cockcroft-gault
equation is nothing but a transformation of creatinine data to make
it simpler to interpret and decrease variance in the model. Can a
clinician as experienced as you deny the value of transformations in
those cases?
---
Back to the Top
[Two replies - db]
X-Sender: n7950211.aaa.popin.ncl.ac.uk
Date: Fri, 25 Feb 2000 10:15:21 +0000
To: PharmPK.-a-.boomer.org
From: James
Subject: Re: PharmPK Re: Optimal sampling points
Dear Dr Cherry,
It is interesting to see how people with different backgrounds have
different perspectives on these things. Personally, I would recommend the
text by Carroll & Rupert, Transformation and Weighting and Regression,
however none of these texts seem to emphasize the role of subject-specific
knowledge in deciding transformations. Both this and the Belsly, Kuh and
Welsch were written in the early eighties and may well be out-of-date
(although Carrol & Ruppert is a classic in my opinion). A lot of this
difference may be to do with terminology (what are boundary values exactly?
Why do they matter?) However, and without need to resort to my
undergraduate degree, I can confidently say that the only analysis that is
insensitive to the data (and how it is collected eg design) is one that
doesn't use it.
Incidentally, the Cockcroft & Gault formula is a misleading "transformation".
James Wright
---
X-Sender: mentor.at.hardlink.com
Date: Fri, 25 Feb 2000 06:41:12 -0500
To: PharmPK.aaa.boomer.org
From: Daro Gross
Subject: Re: Optimal sampling points
I feel like I have stirred up a hornet's nest by simply stating what is
already well understood---that PK/PD analytical techniques are mathematical
transformations that have been proven to be useful but for which there
cannot be an "optimal" data set, only an optimal choice of analytical tools.
The difficulty in choosing the correct PK/PD tools is a measure of a
person's experience in working with a particular data set (i.e., depth of
knowledge in a particular area of medicine) and is not related to the data
set itself.
Some persons have greater amounts of knowledge about the characteristics of
data sets related to a particular area of research. Other persons focus on
being able to adapt PK/PD tools to the parameters of data sets as presented
by the expert researcher.
Medicine is far too complicated to permit a person to acquire both in depth
expertise in an area of research as well as in the use of PK/PD analytical
tools required to analyze the data generated. No person can ever hope to
have the time or the energy in one life-time to absorb enough information
to become sufficiently expert in both specialty-specific data collection
and PK/PD analytical techniques.
An "optimal" data set implies that the researcher gathering the data is not
expert enough to gather the data rather than the more reasonable conclusion
that the selection of the PK/PD analytical techniques used to interpret the
data need conform to the parameters of the data gathered. Failure to chose
the correct data analysis tools does not render the data meaningless---but
may transform useful data into meaningless conclusions.
Daro Gross
P.S. I have run into too much good research that has been discarded because
it did not yield meaningful results when subjected to inappropriate
mathematical measures---PK/PD is intended to address this problem in
medicine, not preserve a status quo that discards valid research when it is
not parameterized in a manner that fits off-the-shelf computer software. (I
have always been under the impression that physicians were trained to
conduct medical research, not computer programmers.)
Back to the Top
A couple of clarifications about optimal designs for PK/PD studies.
With a given
optimality criterion and with stated modeling and statistical assumptions, an
optimal design can always be calculated. However, the design will
depend on the
assumptions. For instance, different variance model parameters can
yield widely
differing "optimal" designs.
Therefore, the robustness of optimal or even favorable experimental designs is
especially important. Fortunately, such designs are quite robust in
populations. The reason is that the optimal designs of the various
subjects are
different and tend to center around the best design that is calculated with the
population average parameters.
In the presence of various uncertainties, it is advantageous to pursue a
sequential strategy. Start with a classical sampling scheme which spreads the
observations around. Narrow the design toward a single-sample optimal plan as
information is gained about the parameters and their properties.
Laszlo Endrenyi
University of Toronto
Back to the Top
Dear All:
Dr. Endrenyi's remarks are very much to the point. Especially
appropriate
is the strategy of starting with a fairly conventional design, doing a few
subjects, making a population model, getting the parameter distributions,
and then defining an optimal design based on those results. Then repeat
this every so often, every 5 - 10 patients, for example, until the model
gets stable. In this way one never puts all one's eggs in one basket of an
experimental design of protocol, but refines it as the results come in.
This optimizes the information per subject and per level and per dollar,
for all of that. Well said, Laszlo!
Roger Jelliffe
Roger W. Jelliffe, M.D. Professor of Medicine, USC
USC Laboratory of Applied Pharmacokinetics
2250 Alcazar St, Los Angeles CA 90033, USA
Phone (323)442-1300, fax (323)442-1302, email= jelliffe.-a-.hsc.usc.edu
Our web site= http://www.usc.edu/hsc/lab_apk
Back to the Top
Ref: Optimal sampling points
Dear All
In one of my earlier papers: Wu G. An explanation for failure to predict
cyclosporine area under the curve using a limited sampling strategy: a
beginner=EDs second note. Pharmacol Res 1997; 35: 547-52, I used the Fourie=
r
seires to decompose the blood drug concentration into linear and nonlinear
components, which is also an optimal sampling points.
Guang Wu, MD, PhD
Laboratoire de Toxicocinetique et Pharmacocinetique
=46aculte de Pharmacie
Universite de la Mediterranee Axi-Marseille II
27 Boulevard Jean Moulin
=46-13385 Marseile Cedex 05
=46rance
Tel: +33 4 91 83 56 45
=46ax: +33 4 91 80 26 12
Back to the Top
I agree with the approach of Dr. Endrenyi endorsed by Dr. Jelliffe, however
there may be a slight problem in doing human research with this approach -
IRB's. At our institution, if one makes any change in a protocol it must be
approved, and adding or deleting samples or changing sample times requires an
amendment. A pharmacokinetic protocol with the approximate number of samples
and approximate sample times, from my experience, would not be approved, even
if the volume of blood did not exceed the our set limit. Although, minor
protocol changes do often represent major inconveniences, I totally agree with
keeping IRB's informed in every aspect of the research under their auspicious.
Just a reminder that conducting pharmacokinetic research (or really any human
research for that matter) can sometimes become a real nightmare for the
investigator and the institution when all the i's aren't dotted and the t's
crossed.
Art Straughn, Member
University of Tennessee IRB
Back to the Top
[A few replies - db]
From: "Bruce CHARLES"
Organization: School of Pharmacy
To: PharmPK.at.boomer.org
Date: Tue, 7 Mar 2000 11:03:55 +1000
Subject: Re: PharmPK Re: Optimal sampling points
X-Confirm-Reading-To: "Bruce CHARLES"
X-pmrqc: 1
Priority: normal
Taking this a step further to the ops level, most (if not all) clinical
activities in Australian teaching hospitals are run on a cost-centre
basis where a unit is responsible for the costing of all their
activities. This could be a stumbling block for the optimal sampling
practice (which I do support).
Further, blood sampling for path, TDM etc is tied quite strictly to the
scheduled rounds of the phlebotomists which, for most drugs,
corresponds to the morning trough level at around 8 am. 'Non-
standard' sampling times can be a logistic headache especially
where there are multiple sampling regimens among patients and/or
which cross shifts of duty of the med and nursing staff.
I guess like most things in health the bottom line at the end of the
day is $ and if we can show favourable cost-effectiveness for such
a practice then optimal sampling will thrive.
Cheers,
BC
Bruce CHARLES, PhD
Associate Professor
Director, The Australian Centre for Paediatric Pharmacokinetics
University of Queensland, School of Pharmacy, QLD 4072
Australia
+61 7 336 53194 (TEL)
+61 7 336 51688 (FAX)
0403 022 252 (MOBILE)
bruce.-a-.pharmacy.uq.edu.au
---
Date: Tue, 07 Mar 2000 14:16:42 +1300
From: Nick Holford
X-Accept-Language: en
To: PharmPK.at.boomer.org
Subject: Re: PharmPK Re: Optimal sampling points
If the design of a study is not optimal then I would argue that it is
unethical. So changing the protocol to make the design better should
always be a priority for an ethical researcher. IMHO it is unethical to
use the burden of paperwork as an excuse for not improving the study
design.
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, Private Bag 92019, Auckland, New Zealand
email:n.holford.-at-.auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556
http://www.phm.auckland.ac.nz/Staff/NHolford/nholford.htm
---
X-Sender: jelliffe.at.hsc.usc.edu
Date: Mon, 06 Mar 2000 17:21:19 -0800
To: PharmPK.-a-.boomer.org
From: Roger Jelliffe
Subject: Re: PharmPK Re: Optimal sampling points
Dear Art:
It is good to hear from you. As to the changes in the
protocol, describe
all this in the original protocol. State the number of samples, state the
original protocol, and state how the changes will be made. Dot all the i's
and cross all the t's. IRB's are able to do this stuff. Many have already
done so, as evidenced by a number of investigators who have done this already.
Very best regards,
Roger Jelliffe
Roger W. Jelliffe, M.D. Professor of Medicine, USC
USC Laboratory of Applied Pharmacokinetics
2250 Alcazar St, Los Angeles CA 90033, USA
Phone (323)442-1300, fax (323)442-1302, email= jelliffe.aaa.hsc.usc.edu
Our web site= http://www.usc.edu/hsc/lab_apk
Back to the Top
[Two replies - db]
From: Laszlo Endrenyi
X-Accept-Language: en
To: PharmPK.aaa.boomer.org
Subject: Re: Optimal sampling points
Date: Mon, 6 Mar 2000 23:41:57 -0500
Nick,
The word "optimal" can be interpreted in different ways. If it means
"as good as
possible under the given constraints", I agree with you: we should
pursue studies
as much free of bias and as efficient as possible. If, however, it
means "optimal
in the statistical sense" then this is perhaps an ideal goal which is, however,
subject to several assumptions. I believe that it is more useful to consider
designs that are favorable under many assumptions and conditions, i.e., robust
designs.
Laszlo Endrenyi
University of Toronto
---
X-Sender: jelliffe.at.hsc.usc.edu
Date: Mon, 06 Mar 2000 21:51:04 -0800
To: PharmPK.-a-.boomer.org
From: Roger Jelliffe
Subject: Re: PharmPK Re: Optimal sampling points
Dear Bruce:
Thanks for your comments about optimal sampling times.
Another thing you
can do is to adjust the times of the doses so that the relation between the
dose and the level can be close to D-Optimal. For example, under many
circumstances, a pretty much D-Optimal strategy is to combine a peak level
with the first dose with another level, often after a different dose, which
is obtained at about 3 hours before the trough, for many patients with a
creatinine clearance above 40. Consider that if morning blood draw time is
at 8 am, for example, one can center the aminoglycoside dosing around an 11
am dose, so that the 8am blood sample is more informative than the trough,
and the pair, as outlined above, is then close to D-Optimal. It is just
another way to juggle things around to get a practical approach to
realizing optimal designs.
Veery best regards,
Roger Jelliffe
Roger W. Jelliffe, M.D. Professor of Medicine, USC
USC Laboratory of Applied Pharmacokinetics
2250 Alcazar St, Los Angeles CA 90033, USA
Phone (323)442-1300, fax (323)442-1302, email= jelliffe.aaa.hsc.usc.edu
Our web site= http://www.usc.edu/hsc/lab_apk
***************
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)