Previous Page | Next Page

The PHREG Procedure

Piecewise Constant Baseline Hazard Model

Single Failure Time Variable

Let be the observed data. Let be a partition of the time axis.

Hazards in Original Scale

The hazard function for subject is

     

where

     

The baseline cumulative hazard function is

     

where

     

The log likelihood is given by

     
     

where .

Note that for , the full conditional for is log-concave only when , but the full conditionals for the ’s are always log-concave.

For a given , gives

     

Substituting these values into gives the profile log likelihood for

     

where . Since the constant does not depend on , it can be discarded from in the optimization.

The MLE of is obtained by maximizing

     

with respect to , and the MLE of is given by

     

Let

     
     

The partial derivatives of are

     
     

The asymptotic covariance matrix for is obtained as the inverse of the information matrix given by

     
     
     

See Example 6.5.1 in Lawless (2003) for details.

Hazards in Log Scale

By letting

     

you can build a prior correlation among the ’s by using a correlated prior , where .

The log likelihood is given by

     

Then the MLE of is given by

     

Note that the full conditionals for ’s and ’s are always log-concave.

The asymptotic covariance matrix for is obtained as the inverse of the information matrix formed by

     
     
     

Counting Process Style of Input

Let be the observed data. Let be a partition of the time axis, where for all .

Replacing with

     

the formulation for the single failure time variable applies.

Previous Page | Next Page | Top of Page