The PHSELECT Procedure

In the getStarted data set that is presented in the section Getting Started: PHSELECT Procedure, suppose the variable C3 represents the gender of the subjects. It is conceivable that males and females belong to different subpopulations and might have different baseline hazard functions. A stratified analysis is conducted by specifying C3 as the stratifying variable in the STRATA statement. The CODE statement requests that a text file named ScoreCode.txt be created. This file contains SAS DATA step code to predict survival probabilities at time points that are specified by the TIMEPOINT= option. By default, PROC PHSELECT uses the estimated quartiles of the KM curve (no covariates) of the training data. The SHOWTIME option produces variables that contain these estimated time points.

proc phselect data=mycas.getStarted;
   class C1 C2;
   model Time*Status(0) = C1 C2 X1-X4;
   strata C3;
   code file='ScoreCode.txt' showtime;
run;

The "Parameter Estimates" table in Output 16.2.1 shows the resulting regression parameter estimates of the stratified analysis.

You can use the DATA step code from the CODE statement to compute direct adjusted survival probabilities by taking the average of individual predicted survival probabilities. The method of direct adjustment controls for possible confounders that result from an imbalance of subject characteristics between groups. This adjustment is especially useful in nonrandomized studies. In the following statements, the DATA step code is applied to the getStarted data set to obtain the predicted survival probabilities for individuals in that data set and the predicted probabilities at each time point are averaged by gender:

data Scores;
   set mycas.getStarted;
   %inc 'ScoreCode.txt';
run;

proc print data=Scores (obs=1);
   var Time_:;
run;

proc means data=Scores mean;
   class C3;
   var S_Time_:;
run;

Output 16.2.2 displays the time points at which the survival probabilities were predicted. These time points, Time_1=7, Time_2=19, and Time_3=54, are the quartiles of the failure times based on the Kaplan-Meier curve. Output 16.2.3 displays the direct adjusted survival probabilities at these quartiles. The predicted survival probability at Time_1 for females is slightly lower than that for males, but at Time_3 the direction is reversed. Also revealed is that the getStarted data set has 76 males but only 24 females.