The PHREG Procedure

Example 66.14 Bayesian Analysis of Piecewise Exponential Model

This example illustrates using a piecewise exponential model in a Bayesian analysis. Consider the Rats data set in the section Getting Started: PHREG Procedure. In the following statements, PROC PHREG is used to carry out a Bayesian analysis for the piecewise exponential model. In the BAYES statement, the option PIECEWISE stipulates a piecewise exponential model, and PIECEWISE=HAZARD requests that the constant hazards be modeled in the original scale. By default, eight intervals of constant hazards are used, and the intervals are chosen such that each has roughly the same number of events.

   data Rats;
      label Days  ='Days from Exposure to Death';
      input Days Status Group @@;
      datalines;
   143 1 0   164 1 0   188 1 0   188 1 0
   190 1 0   192 1 0   206 1 0   209 1 0
   213 1 0   216 1 0   220 1 0   227 1 0
   230 1 0   234 1 0   246 1 0   265 1 0
   304 1 0   216 0 0   244 0 0   142 1 1
   156 1 1   163 1 1   198 1 1   205 1 1
   232 1 1   232 1 1   233 1 1   233 1 1
   233 1 1   233 1 1   239 1 1   240 1 1
   261 1 1   280 1 1   280 1 1   296 1 1
   296 1 1   323 1 1   204 0 1   344 0 1
   ;

proc phreg data=Rats;
   model Days*Status(0)=Group;
   bayes seed=1 piecewise=hazard;
run;

The "Model Information" table in Output 66.14.1 shows that the piecewise exponential model is being used.

By default the time axis is partitioned into eight intervals of constant hazard. Output 66.14.2 details the number of events and observations in each interval. Note that the constant hazard parameters are named Lambda1,..., Lambda8. You can supply your own partition by using the INTERVALS= suboption within the PIECEWISE=HAZARD option.

The model parameters consist of the eight hazard parameters Lambda1, ..., Lambda8, and the regression coefficient Group. The maximum likelihood estimates are displayed in Output 66.14.3. Again, these estimates are used as the starting values for simulation of the posterior distribution.

Without using the PRIOR= suboption within the PIECEWISE=HAZARD option to specify the prior of the hazard parameters, the default is to use the noninformative and improper prior displayed in Output 66.14.4.

The noninformative uniform prior is used for the regression coefficient Group (Output 66.14.5), as in the section Bayesian Analysis.

Summary statistics for all model parameters are shown in Output 66.14.6 and Output 66.14.7.

The default diagnostics—namely, lag1, lag5, lag10, lag50 autocorrelations (Output 66.14.8), the Geweke diagnostics (Output 66.14.9), and the effective sample size diagnostics (Output 66.14.10)—show a good mixing of the Markov chain.