Dear allBack to the Top
Does anyone have experience in exclusion of outlier data pointsin
pk-studies?
We use 4 animals at every time point.
Example of a data set:
At t = 60 min
Concentrations (ng/ml): 822, 759, 4288, 800.
Howconfidently can Iexclude the data point 4288 ng/ml?
References to any relevant publications will be highly appreciated.
Best regards,
Eva Dam
Department of Bioanalysis
NeuroSearch
Denmark
Back to the Top
Dear Eva
Robust fitting using an iteratively reweighted least squares algorithm
is a good tool to solve the outliers problem. Many software as SYSTAT
and other programs can be used for this purpose.
Best regards.
Jos\0xC8 M. Lanao.
Dpt. Pharmacy and Pharmaceutical Technology
University of Salamanca
Spain
Back to the Top
Outliers can be detected statistically using Z-score. A Z-score value
more than 3 or less
than -3 is considered an outlier value. Note that some studies may need
logarithmic
transformation of data before analysis.
Back to the Top
Dear Eva,
I think that only doing an exploration of your results you can be
confidently to exclude this value for your calculations but if you
prefer to
be sure you can apply a Dixon's test (test specific to find outliers). I
took the liberty of doing the test with your data and I can say you
that its
an outlier point. You can find this test in the statistical literature.
I think that it's important to know if its the complete kinetic of this
animal that have an estrange behaviour or only this sample at 60'
because in
the first case you can exclude the animal for your calculations when you
find the problem (more dose administered, animal with the minor
weight,...).
I hope it helps.
Sincerely,
Daniel Mart\0xCCnez
RIA Laboratory
Metabolism & Pharmacokinetics Service
Research & Development Department
IPSEN PHARMA, S.A.
Ctra. Laure\0x2021 Mir\0xDB 395
Sant Feliu de Llobregat, Barcelona, Spain
Tel\0xC8f.: 936858100
daniel.martinez.at.beaufour-ipsen.com
Back to the Top
Dear Dr. Dam,
Exclusion of outlier data is a very tricky topic, that has been dealt
with in this discussion group some time ago.
In my honest opinion there is no generally applicable solution for this
problem. The use of statistics may be appropriate, but is not warranted
in all cases. Omitting values deviating by more than some multiple of
the standard deviation (e.g. 3) is attractive from a practical point of
view, but it is somewhat casual, since the standard deviation is
dependent on the particular data set.
In my experience the use of iteratively reweighted least squares
fitting does not solve the problem adequately. Just as in linear
regression, nonlinear regression is quite sensitive to outliers.
As a general rule, I recommend to analyse your data both including and
excluding the possible outlier(s) (if there is more than one outlier,
the problem increasing dramatically!), and judge both solutions
carefully with respect to plausibility.
Hans Proost
Johannes H. Proost
Dept. of Pharmacokinetics and Drug Delivery
University Centre for Pharmacy
Antonius Deusinglaan 1
9713 AV Groningen, The Netherlands
tel. 31-50 363 3292
fax 31-50 363 3247
Email: j.h.proost.at.farm.rug.nl
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)