Date: Mon, 26 Aug 2002 20:37:41 +0100 Kevin "SAS(r) Discussion" Kevin blueyonder (post doesn't reflect views of blueyonder) Re: Stratified Non-parametric analysis / confidence intervals

Hi,

I would recommend going ahead with the van Elteren test which can be carried out using proc freq (see www.sas.com for more info). Basically, the textbook is saying that the test is not valid if there is a treatment by subgroup interaction. This is true for just about every analysis that would be carried out on a clinical trial.

The Hodghes-Lehmann estimator is an appropriate estimator to use with the van-Elteren test. To generate a confidence interval that relates to the van Elteren test, you could create a look using a macro %do statement and add small increments to the patients' values for one of the treatment groups. Carry out the van Elteren test within this loop and the range of added values that give a significant result for a treatment effect is your confidence interval.

I have to note that in my experience most people use the van Elteren test and give the Hodghes Lehmann estimator and confidence intervals.

I have a SAS macro that does this as part of the PHASTAR application but unfortunately it is not free!

Hope this helps,

Kevin Kane kevink@phastar.co.uk http://www.phastar.co.uk

"Stefan Hantel" <Stefan.Hantel@t-online.de> wrote in message news:000501c24c39\$4c2c8c70\$0100a8c0@HANTEL002... > Dear all, > > I want to analyze data from a multi-center trial. Since the data are > obviously not normally distributed, if have to perform a non-parametric > test. Without considering the center, I would use a Hodges-Lehman > estimator and confidence-intervals. Does somebody know a similar method, > which considers also centre? > > The van Elteren test could be an alternative, but according to a text > book, it is only applicable, if the results per center are not > contradicting each other (i.e., in one center treatment A is better than > B and in an other center B is better than A). > > Unfortunately, I have to present confidence intervals for the difference > of the medians / means according to the study protocol. Therefore, I > cannot use confidence intervals for the effect measures P(X<Y) + 0.5 > P(X=Y). > > Does somebody can provide me with a reference / hint how to solve the > problem? Is somebody aware of a SAS-solution? > > Thank you very much for your help. Any hint is highly appreciated. > > > > Stefan

Back to: Top of message | Previous page | Main SAS-L page