Back to the Top
The following message was posted to: PharmPK
Hi, I'd like to know what is the most appropriate
nonparametric method to do treatment ratio comparison
in average BE study? I have seen
Wilcoxon-Mann-Whitney, Hodges-Lehmann, and Boostrap
Confidence Interval approaches on a book and articles.
Is there any preference and what's the most
user-friendly approach? In addition, if there is any
existing SAS or S-PLUS source codes available?
Any feedback will be greatly appreciated.
Thanks.
Back to the Top
The following message was posted to: PharmPK
Hi Chang,
You should only have to analyze Tmax nonparametrically (except for
submissions to Canada I believe they want to see parametric still
[sorry, my
reference material is at the office]). If you have a single fixed
sequence
then I use Proc Paired in either PC StatXact or third party software
bundled
right in Unix SAS. If you have a crossover or parallel design then you
should use Proc Twosampl to analyze the distribution shift. Both
utilize
Hodges-Lehmann in the analyses.
Word of caution there are two bugs in Proc Paired. The first is that it
doesn't bring ties into the analysis which could give you erroneous
point
estimates and CI. To fix this bug simply add an extremely small amount
(say
0.000001) unbiasly to all the "post" measurements to break all the
ties. If
you format your output for only one or two decimal places then you don't
have to worry about subtracting it back out of your results. The other
bug
is a memory bug (doesn't reset the variable -99999.00) which is caused
when
you have too many ties in your analysis. The only way to reset the
memory
is to exit out of SAS and get back in.
I have reported both bugs to Cytel and they will be addressing these
issues
in future releases.
Regards.
Rob
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)