Date:         Fri, 13 Nov 2009 12:13:33 -0800
Reply-To:     Oliver Kuss <Oliver.Kuss@MEDIZIN.UNI-HALLE.DE>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Oliver Kuss <Oliver.Kuss@MEDIZIN.UNI-HALLE.DE>
Organization: http://groups.google.com
Subject:      Re: Incremental Contribution to Regression of Correlated
              Dichotomous Variables
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset=ISO-8859-1

On 12 Nov., 22:38, Haris <karoval...@gmail.com> wrote:
> I have a dataset where Gold Standard Diagnosis (Y/N) is known for each
> patient.  Then there also are three tests (T1-T3, also Y/N) that are
> trying to match the Gold Standard.  This is a a case of four
> correlated non-independent dichotomous variables.  Taken two at a
> time, I can use McNewmar test but I need to consider them all
> simultaneously.
>
> 1. How do I test if T2 contributes anything to the prediction of Gold
> Standard when CONTROLLING for what is already explained by T1?  Proc
> CATMOD came up in my search but I am not familiar with this procedure
> and it's not immediately intuitive.  I also found an old reference to
> GEE but seems to me that NLMixed and CatMod are the newer more
> advanced methods.  Anyone?
>
> 2. Can I compute a kappa inter-test reliability coefficient with SAS
> when there are more than two correlated variables involved?
>
> I am stuck for now.  Please help.

Dear Haris,
if one of your dichotomous variables is the gold standard, then you
should not use kappa statistics for analysis but the methods from
diagnostic accuracy testing, that is, sensitivity, specificity, and
predictive values. If one has several tests to compare I have good
experience using the ideas of Leisenring et al. (Leisenring W, Pepe
MS, Longton G. A marginal regression modelling framework for
evaluating medical diagnostic tests. Stat Med. 1997 Jun 15;16(11):
1263-81.). They show how this situation can be interpreted as a
marginal regression model which can be fitted by using the GEE methods
in PROC GENMOD.

Hope that helps,
Oliver