Examining the Relationship between Survival Time and Time-dependent Covariables Using the Phsplm Procedure

The PHSPLM procedure describes survival data using a. parametric proportional hazards model where the baseline hazard is a eubic spline function with tails that are linearly restricted. Herndon (1988) has propooed this flexible hazard modd 8$ an alternative to the Cox model (1972). Properties of this model include: efficiency of estimation of cavariable coefficients and survival probabilities, smooth survival curves, and ~nfidence limits for survival and hazard estimates, even when cavanables are time-dependent. In addition, it has been sh{)wn that the use of the restricted cubic spline model when examining the erred of time-dependent covariables can :reduce computation time by a factor of 213 over that of the Cox mooel. The pnSPLM procedure can fit the homogeneous restricted cubic spline model, ignoring covariables; and also, the model adjusted for fixed ~(lvariables and/or time-dependent covariables. The implementation of the PHSPLM procedure is similar to the PHGLM procedure {Harrell, 1986). Features i.nclude those mentioned above, in addition to those included in the PHGLM prowdure. The use of the PHSPLM procedure will be demonstrated and contrasted against analytical results obtained fiom BMDP (Di:lOOn, 1981).