Back to the Top
Dear All,
I have a question regarding Interpretation of Results for a
Bioequivalence Study.
For one of our dataset,when Statistical Analysis (ANOVA, Calculation
of 90% CI) is performed on untransformed data, it Fails to show
Bioequivalence.
However for after Log Transformation on the same dataset, study shows
Bioequivalence.
Can anyone help me understanding the reason for this behaviour.
Thanks in Advance.
Regards,
Vikesh S
Back to the Top
Dear Vikesh,
My thought is you probably have high variability on the raw
untransformed data. Log transformation trends to decrease the data
variability and this may be the reason of the results you have
obtained when applied ANOVA.
Hope this can clarify something,
Best regards,
Prof. Sergio F. Sanchez Bruni
Research Scientist of CONICET
Laboratory of Pharmacology
Faculty of Veterinary Medicine
UNCPBA-Tandil (7000), Argentina
Back to the Top
The following message was posted to: PharmPK
Dear Vikesh and Sergio,
The reason for using a log transformation in bioequivalence testing
is that
it is likely that AUC values are (at least close to) log-normally
distributed, mainly because clearance is usually (close to) log-normally
distributed between subjects. A log-normal distribution implies that the
log-transformed values are normally distributed. And when AUC values are
log-normally distributed, one should apply ANOVA and any other
statistical
test on the log-transformed values.
Sergio wrote:
> Log transformation trends to decrease the data variability
This is indeed more or less the case, but when you write this, one could
think of some magic trick to decrease data variability. This is not
the case
here. As stated above, log-transformation is the right thing to do if
the
data are (close to) log-normally distributed.
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. 31-50 363 3292
fax 31-50 363 3247
Email: j.h.proost.-a-.rug.nl
Back to the Top
Dear Hans and Vikesh,
I do agree your thought, you are right and explained very well the
why of using log data transformation in bioquivalence studies.
But I think we did not fully reply the Vikesh's question which was:
"For one of our dataset,when Statistical Analysis (ANOVA,
Calculation of 90% CI) is performed on untransformed data, it Fails
to show Bioequivalence. However for after Log Transformation on the
same dataset, study shows Bioequivalence."
If it is true that my phrase induces to a magic trick for forcing
some results, I cannot deny that it is frequently used for that
reason. When data variability are very low there is a high
likelihood that ANOVA testing of the PK parameters is consistent
for transformed and untransformed data giving the same results of
bioequivalence.
Hans, I'm very sorry and don't get me wrong please, but how do you
know that there is not variability enough on the Pk parameters
obtained for the Vikesh's study.
Vikesh I'd like to hear a word from you. .
Best regards,
Prof. Sergio F. Sanchez Bruni
Research Scientist of CONICET
Laboratory of Pharmacology
Faculty of Veterinary Medicine
UNCPBA-Tandil (7000), Argentina
Telefax:+54- (0)2293-422357/426667
Back to the Top
The following message was posted to: PharmPK
Dear Nav!
you wrote:
>Is anyone in the group familiar with whether or not data for BE/BA
>type studies should be ln-transformed prior to outlier analysis.
>
Yes ;-)
>I am using Grubb's test which is based on the Z ratio and depending
>on whether the data is transformed or not, there is a slight
>variation in the results.
>
<>
What /exactly/ are you testing right now?
Since you are interested in doing the final analysis in the log-scale
you may test the individual ratios of untransformed responses (or
differences of logs).
Data mining does not make sence in a confirmatory trial (e.g.,
trying different statistical tests or playing around with
distributional assumptions).
What variant of Grubb's test are you using - the one detecting
1 outlier or the one detecting 2 outliers?
Anyway in BE analyses we are not interested in the individual
BE-ratios - only reminds me on FDA's 75/75-rule ;-)
If you are interested whether your statistical assumptions for
ANOVA are valid, you should primarily look on the intra-subjects
residuals:
Shapiro-Wilk for normality and for outliers maybe Hund's test, or
Tukey's sum-differnce plot, or the Mahalanobis distance, or ...)
best regards,
Helmut
--
Helmut Schuetz
BEBAC
Consultancy Services for Bioequivalence and Bioavailability Studies
Neubaugasse 36/11
1070 Vienna/Austria
tel/fax +43 1 2311746
http://BEBAC.at
Bioequivalence/Bioavailability Forum at http://forum.bebac.at
http://www.goldmark.org/netrants/no-word/attach.html
Back to the Top
The following message was posted to: PharmPK
Dear Sergio,
Thank you for your reply. This thread started with:
> "For one of our dataset,when Statistical Analysis
>(ANOVA, Calculation of 90% CI) is performed on
>untransformed data, it Fails to show Bioequivalence.
>However for after Log Transformation on the same
>dataset, study shows Bioequivalence."
This is not uncommon. A test on untransformed data is
different from a test on transformed data, and the result
can be different. If not, we would not worry about the
transformation! As stated earlier, the choice between
untransformed and transformed data should be based on the
statistical distribution of the values. As this
distribution cannot be obtained from the data in many
cases, a plausible a priori assumption is usually made. In
case of AUC and Cmax, the log-normal distribution is the
logical choice.
> If it is true that my phrase induces to a magic trick
>for forcing some results, I cannot deny that it is
>frequently used for that reason.
Yes, I suppose that this is the case. But we all know, or
should know, that one is not free to choose a statistical
test. Preferrably the test is chosen before the actual
experiment is performed. Deviation from the originally
chosen test requires a sound and solid justification.
'Forcing some results' is definitely not a sound and solid
justification!
> Hans, I'm very sorry and don't get me wrong please, but
>how do you know that there is not variability enough on
>the Pk parameters obtained for the Vikesh's study.
I didn't say that. On the contrary, I agree with your
statement that in case of low variability the test on
untransformed and transformed data will give essentially
the same results.
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. 31-50 363 3292
fax 31-50 363 3247
Email: j.h.proost.-a-.rug.nl
Back to the Top
Dear Hans,
Thanks very much for your contribution in the present discussion. You
were absolutely consistent and clear enough. Hope this discussion
helps to Vikesh to solve the concern on the study results
interpretation.
Best regards, and see you all next time.
Prof. Sergio F. Sanchez Bruni
Research Scientist of CONICET
Laboratory of Pharmacology
Faculty of Veterinary Medicine
UNCPBA-Tandil (7000), Argentina
Back to the Top
The following message was posted to: PharmPK
Dear Hans and Sergio,
Thanks for your comments.
As far as my data set is concern i observed a high Intra CV for
Untransformed analysis,
however after ln-transformation Intra CV reduced after ln-
transformation.
May be this was the reason for dataset failing under Untransfmed
analysis and passing Bioequivalence Criteria after
ln-transformation.
Also i think this type of case occurs with dataset having small
sample size.
I guess if a large sample size is used then may be the difference in
the results of Untransformed and ln-transformed analysis should not
be very large.(I mean failing BE under untransformed state and
Passing after ln-transformation).
Another possible reason that i could see may be any occurence of some
Outlying Values in the dataset.
However I also feel that A log-normal distribution implies that the
log-transformed values are normally distributed.
Hence ANOVA results are more appropriate and reliable after Log-
Transformation.
Best Regards,
Vikesh S
Back to the Top
The following message was posted to: PharmPK
One of the basic assumptions of ANOVA is that the variables be normally
distributed. Log transform may accomplish that. Ideally, it should be
determined that the tranform did normalize the data; if not a different
transform can be used to accomplish that.
James D. Prah, PhD
US EPA
Human Studies Division MD (58B)
Research Triangle Park, NC, 27711
919 966 6244
919 966 6367 FAX
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)