Date:         Wed, 14 Jan 2004 10:38:39 -0600
Reply-To:     Herbert Morley A <Morley.Herbert@LONESTARHEALTH.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Herbert Morley A <Morley.Herbert@LONESTARHEALTH.COM>
Subject:      Cox Regression
Content-Type: text/plain; charset="iso-8859-1"

I am comparing survival curves of patients using Cox propotional hazards.
The occurrence of an 'event' is the end-point and the covariates are
including gender (male =1; female =0) and whether they are diabetic and a
how they are treated variable- diabcont (non-diabetic =0; diabetic with diet
control =1; diabetic with oral meds =2; diabetic with insulin =3) and the
group (control =0; treatment =1).

PROC PHREG DATA = combo;
MODEL timevent*event(0) =type diabcont gender ;
RUN;

The analysis of max likelihood gives me a hazard ratio of 1.157 for
diabcont. I am not sure what we are comparing- the book says a change of 1
unit. However from  the Kaplan-Meier curves I would expect the hazard to be
much large for the diabcont=2 group compared to any of  the others. The
diabcont=0 and diabcont=3 curves are very close together so I expect a ratio
of 1 or so for that. However I would expect the ratio to be 2-3 or so for
the diabcont=2 group.

I would like to get ratios referred to diabcont =0 or any other reference
level, like in Logistic where the  (ref=) after the class variable gives the
baseline. I can't figure out how to do this. I tried TEST but am missing
something.

Appreciate any guidance.
Thanks in advance

Morley Herbert