SAS Code for the First-Stage Model
The Cox proportional hazard model formulated in (12.18) can be fitted using SAS procedure PHREG.
proc phreg data=firststage covs(aggregate) covout outest=mat;
class center endpoint/param=glm;
model outcome*status(0)= endpoint*treat*center;
strata center endpoint;
It is assumed that there are two records per subject in the input dataset, the first one corresponding to the surrogate endpoint and the second one to the true endpoint. The option covs(aggregate) applies the robust sandwich estimator of Lin and Wei (1989) for the covariance matrix that will be used in the second stage to correct for the uncertainty in the estimated parameters. The outest option creates an output SAS dataset containing estimates of the regression coefficients and the option covout adds the estimated covariance matrix of the parameter estimates to the outest dataset. In the MODEL statement, status is the censoring indicator variable (0=censoring, 1=event). In the strata statement we specify the variables that determine the stratification.
The parameter estimates obtained from model (12.18) are shown in Figure 12.24.