PHREG multilevel response

[lillymaginta1]

Hi Everyone,

How can I modify PHREG statement to run for a multilevel response variable (stm= 0, 1, 2). The following statement yield one HR, I need to get two HR for stm(1 vs. 0 and 2 vs. 0)

proc phreg data=s ;

   class drug (ref='1') stm (ref='0') / param=ref;

   model period1*stm(0)= drug  year/ risklimits;

run;

any help is appreciated 

[SASJedi]

Moved this post to Statistical Procedures, where it should get better exposure.

[SteveDenham]

Try adding a HAZARDRATIO statement, such as:

hazardratio stm/diff=ref;

SteveDenham

[lillymaginta1]

Thank you for the response! I tried the statement but I received the following error, the variable specified in the hazardratio statement must be in the model. 

[StatDave · Accepted solution]

If the response that you want to model is multinomial (either ordinal or nominal), then you should be using PROC LOGISTIC, not PROC PHREG. You can find examples of both in the Examples section of the LOGISTIC documentation. From your description, it sounds like you want to treat the response levels as nominal. The comparisons you mention will be obtained if you use the LINK=GLOGIT option in the MODEL statement and the DESCENDING response variable option:

model stm(descending) = drug year / link=glogit;