topic Re: Out of sample prediction with PROC PHREG in Statistical Procedures

Out of sample prediction with PROC PHREG

Vari — Wed, 19 Jul 2017 08:22:46 GMT

Hi,

I would like to make an out of sample prediction of the survival probability based on my PROC PHREG estimates. Is this possible in SAS?

Re: Out of sample prediction with PROC PHREG

ballardw — Wed, 19 Jul 2017 14:56:21 GMT

Please describe what you mean by 'out of sample'?

Do you mean to apply the model from Proc Phreg to a different data set? This is scoring. You can use the STORE option to save the Phreg information and then Proc PLM to score the data pointing to that saved information.

An example using Proc Mixed from the online help, replace the Proc Mixed code with your proc (though I suggest saving the store information into a different permanent library than SASUSER ) and then point to your store and other data set in the proc Proc Plm Score statement. The Show and LSMEANS are just from the example an you likely don't need them.

proc mixed data=MyBigDataSet;
   class Env A B sub;
   model y = A B x / ddfm=KenwardRoger;
   random int A*B / sub=Env;
   repeated / subject=Env*A*B type=AR(1);
   store sasuser.mixed;
run;

proc plm restore=sasuser.mixed;
   show cov Parms;
   lsmeans A B / diff;
   score data=NewData out=ScoreResults;
run;

Re: Out of sample prediction with PROC PHREG

Vari — Thu, 20 Jul 2017 11:34:39 GMT

Thanks you! This seems to have worked. I get predicted values for the new
observations.

I am not sure, however, how to interpret these predicted values. For about
99% of the cases the predicted values are negative (and they range between:
-2.96 to 4.27). How do these relate to failure times or hazard rates?

##- Please type your reply above this line. Simple formatting, no
attachments. -##