Date: Thu, 18 Nov 2004 00:05:36 GMT
Reply-To: Kevin Kane <kk014d0889@BLUEYONDER.CO.UK>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Kevin Kane <kk014d0889@BLUEYONDER.CO.UK>
Subject: Non-parametric analysis in clinical trials - Nightmare!
PHASTAR SOFTWARE... free experimental version of PHASTAR available at
Have you ever been in the unfortunate situation where you have to carry out
a non-parametric analysis for a clinical trial? If you've had to do this,
and you didn't need to include confidence intervals, then you might not
think it's too bad. But add in the horror of confidence intervals and it
becomes a SAS programmer's nightmare! Although the latest versions of SAS
produce some non-parametric confidence intervals, the standard way of
calculating these confidence intervals in the pharmaceutical world involves
using a lookup table to choose variables that are at some location in the
dataset, after it's been ordered. These can be approximated when the number
of patients in the trial becomes large, but that really only makes the
problem of writing a generic SAS program even more difficult.
Fear no more! PHASTAR can produce these reports - and, of course, macro
free SAS programs to accompany them - directly for your report.
Increasingly these days, adjusting for other factors is becoming more common
in non-parametric analyses. The method that is normally used is called the
van-Elteren extension to the Wilcoxon Rank Sum Test - this allows the
analysis of the treatment effect of your drug to be adjusted for some
categorical variables. The developer of PHASTAR, Kevin Kane, has not only
implemented the ability to produce P-values from this test, but PHASTAR is
the only statistical software that readily produces confidence intervals
that relate directly to the van-Elteren extension. Until now, most people
have been using unadjusted confidence intervals with the adjusted P-values,
but this is not necessary with PHASTAR.
As with the other report generating wizards, you can select various overall
statistics (overll P-value, degrees of freedom etc) and statistics relating
to the treatment effect (Hodghes-Lehmann estimator, confidence interval at a
specified level of alpha etc). You can select specific treatment
comparisons if you are interested in certain contrasts between treatments.
And the output can be sent to a text, HTML, RTF or PDF file.
Thanks to PHASTAR, the headache of non-parametric analyses should be over
A FREE trial version of PHASTAR can be downloaded for free from