Back to the Top
The following message was posted to: PharmPK
Can somebody suggest me if there is a way to generate mean plasma conc.

time plot when we input individual subject data in WinNonlin? I know
that
this is possible in Kinetica just wondering if we can get the same in
WinNonlin.
Thanks in advance.
Prasad Tata
Mallinckrodt, Inc.
Back to the Top
The following message was posted to: PharmPK
Conduct descriptive statistics on the concentration data and plot the
mean
values vs. time.
Sibel D. Ucpinar, Ph.D.
Pharamcokinetic Division
CEDRA Coporation
Back to the Top
Hi Tata,
It's very simple process in WinNonlin.
First, you go to Tools>Descriptive Statistics in WNL.
Then, define your Sort Variables=Time and Summary
Variables=Concentration.
Click Calculate button.
Once you have this result, then plot X=Time, Y=Mean, which is what you
want.
If you want to add +SD on the timemean conc plot, just add those in
WNL using Error Bar menu on Chart Wizard.
I hope this proves helpful.
Joseph Kim
GlaxoSmithKline
Back to the Top
The following message was posted to: PharmPK
I know one way of constructing mean conc plots in WinNonlin.
Use the descriptive stats module and calculate the mean concentrations
..
Once this is done, open the chart wizard and select XY scatter, make
sure it
uses the descriptive stats workbook as the source. After that , you can
select the variables you want to plot , add error bars, etc.
Hope this helps!
Clapton Dias, Ph.D.
Pharmacokineticist II,
PRA International
16400 College Boulevard,
Lenexa, Kansas.
Ph: 9132277257
Back to the Top
The following message was posted to: PharmPK
Dear Prasad,
Do you mean arithmetic or geometric mean? Seems to be
experts in PK land are not quite agreeing which one is
more appropriate and should be used!
Rostam
Back to the Top
The following message was posted to: PharmPK
> Do you mean arithmetic or geometric mean? Seems to be
> experts in PK land are not quite agreeing which one is
> more appropriate and should be used!
>
It makes no difference what you plot. These kind of graphs are just
pretty pictures for marketing and other presentation purposes. Naived
pooled data analysis methods like this have no place in statistical
inference.

Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
email:n.holford.aaa.auckland.ac.nz tel:+64(9)3737599x86730 fax:3737556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
Back to the Top
The following message was posted to: PharmPK
Thanks Nick, is it fair to extend your opinion to
actual PK parameters as well? In general, do we have
enough confidence/data to say most PK parameters are
log normally distributed?
Rostam
Back to the Top
The following message was posted to: PharmPK
Rostam Namdari wrote:
>
> Thanks Nick, is it fair to extend your opinion to
> actual PK parameters as well? In general, do we have
> enough confidence/data to say most PK parameters are
> log normally distributed?
The answer to your question depends on why you care about the
distribution. For descriptive purposes assuming a specific theoretical
distribution such as normal or lognormal is not necessary. The
assumption of a specific distribution is only relevant when you want to
use that assumption to make some inference about the parameters e.g.
comparing the means of two groups or need to make a prediction such as
the 95% confidence interval. There are ways of avoiding these
assumptions about distributions if you use more robust statistical
techniques such as the bootstrap.
http://wfn.sourceforge.net/bshistry.htm
I generally prefer to use the log normal distribution model for PK
parameters because this guarantees that any predictions I make from the
distribution will always have positive values. If I assume a normal
distribution I could predict impossible things like negative clearances.
Please remember that distributions such as the normal or lognormal are
just models for reality. In this sense they are just like PK models.
George Box's aphorism "All models are wrong but some are useful"
applies to statistical distribution models too.

Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
email:n.holford.a.auckland.ac.nz tel:+64(9)3737599x86730 fax:3737556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
Back to the Top
The following message was posted to: PharmPK
Nick thanks again, your point with regard to the
purpose of analysis was well made. Prasad also had an
interesting point that I thought I will share it with
the group.
When you transform anything into log you will use
geometric mean, any untransformed data you will use
arithmetic mean. Let us say in BE studies you log
transform the data for AUCinf, AUCt, Cmax and do not
transform the data for T1/2, Tmax, Kel. For the first
three parameters geometric mean is more accurate where
as for the rest arithmetic mean is more accurate since
most of the programs are cookie cutter programs we
routinely generate and present both am and gm data for
all the parameters.
Rostam
Back to the Top
The following message was posted to: PharmPK
Dear Rostam Namdari,
You wrote:
> Let us say in BE studies you log
> transform the data for AUCinf, AUCt, Cmax and do not
> transform the data for T1/2, Tmax, Kel. For the first
> three parameters geometric mean is more accurate where
> as for the rest arithmetic mean is more accurate since
> most of the programs are cookie cutter programs we
> routinely generate and present both am and gm data for
> all the parameters.
I don't understand the rationale behind your preference for arithmetic
mean for T1/2 and Kel. Why would this be more accurate that the
geometric
mean, and what have 'cookie cutter programs' to do with this? Of course
the
'best' choice would be to use the arithmetic mean if the data are
normally
distributed and the geometric mean if they are lognormally distributed.
But
in practice the problem is that this cannot be established
satisfactorily.
Also, using the arithmetic mean in one case and the geometric mean in
another case would be confusing.
There is at least one major advantage of using the geometric mean for
T1/2/
and Kel: the geometric mean of T1/2 is Ln(2) / Kel and vice versa. This
is
certainly not the case for the arithmetic mean (and the difference may
be
quite large if the CV is large).
Best regards,
Hans Proost
Johannes H. Proost
Dept. of Pharmacokinetics and Drug Delivery
University Centre for Pharmacy
Antonius Deusinglaan 1
9713 AV Groningen, The Netherlands
tel. 3150 363 3292
fax 3150 363 3247
Email: j.h.proost.aaa.farm.rug.nl
PharmPK Discussion List Archive Index page
Copyright 19952010 David W. A. Bourne (david@boomer.org)