Date: Thu, 27 Jul 2006 09:12:41 +0200
Reply-To: "Sitals, Sonia" <Sonia.Sitals@KGU.DE>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "Sitals, Sonia" <Sonia.Sitals@KGU.DE>
Subject: AW: Sample size problem.
Content-Type: text/plain; charset="utf-8"
thanks for your answer, but I have version 8.2!
By the way, I will be out of office for 2 weeks (vaccationss ;) ).
How can I go out of the list for this period and then join the list again with other e-mail-address?
Thanks for helping,
Von: Zack, Matthew M. (CDC/CCHP/NCCDPHP) [mailto:firstname.lastname@example.org]
Gesendet: Montag, 24. Juli 2006 14:04
An: Sitals, Sonia
Betreff: RE: Sample size problem.
If you have SAS version 9, you might run several different possibilities through
the SAS procedure, PROC POWER, using its TWOSAMPLEFREQ statement, specifying
various levels of risk with the new method vs. the "known" risk of the current method
and a specified statistical significance level, alpha (for example, alpha=0.05).
Whether alpha should be one-tailed or two-tailed is something for you to decide.
This gives the research group the options of either increasing the sample size
or decreasing the proposed risk of the new risk relative to the risk of the current
method. You might also consider splitting the total sample size into two unequally
sized groups, though this may not get you much of an advantage with respect to
PROC POWER can also plot some of its output.
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Sitals, Sonia
Sent: Monday, July 24, 2006 6:49 AM
Subject: Sample size problem.
Hi , it would be nice if you cold help me.
Patients undergoing open-heart procedures are in risk for neurologic injuries.
With the Magnet Ressonanz Technic, has been shown that almost 50% of all patients shows adverse neurologic outcomes (yes or no).
With a new OP-method it is intended to reduce this risk in 25% of the cases.
Randomized two groups – Control and Study
Control – expected that 50% of all Patients will show adverse neurologic outcomes.
Study - it is expected that adverse neurologic outcomes will decresed in 25%.
The research group proposed 40 patients in each group, but the ethic commission asked for a sample size evaluation.
I thought that Logistic regression should be used to analyse the data.
I do not believe that this kind of data is normally distributed – they show me one publication, but not with this Magnet ressonanz test, and ther was no word about distribution.
Chi-Square and Fischer’s exact tests were used.
I made an evaluation: Gruppe 1: 50% Gruppe 2: 37.5% (α): 5% Power: 80%
Each Group: 193
For other α und Power the sample size varies between 485 (α=0.1%) and 128 (Power=65%).
Unfortunatly the research group found this sample size too big and they would like to have fewer patients, let’s say, 50 in each group, but this must be justified, once the power for this would be less then 30%.
In Lenth, R. V. (2001), (Lenth, R. V. ( (2001), ``Some Practical Guidelines for Effective Sample Size Determination,'' The American Statistician, 55, 187-193 .)
Dr. Sonia de Arêa Leão-Sitals
Klinikum der Johann Wolfgang Goethe-Universität
Klinik für Thorax-, Herz- und thorakale Gefäßchirurgie
Haus 23, UG, Raum 21 A Theodor-Stern-Kai 7
60590 Frankfurt am Main
Tel.: 069-6301 6111
Fax: 069-6301 81056 / 83964