Using the Time Dependent ROC Curve to Build Better Survival Model

The ROC curve is mainly used to evaluate the discrimination power of a continuous variable for a binary outcome. Recently, time dependent ROC curves have been used to assess the predictive power of diagnostic markers for time dependent disease outcomes, thus to analyze censored survival data. Among the various methods of estimating the time dependent ROC curves, the Kaplan-Meier method is based on Bayes’ theorem and Kaplan-Meier survival function. It is easy to understand, implement and use. In this paper we implement the Kaplan-Meier estimate of time dependent ROC curves in SAS. Using the data of a clinical study, we demonstrate how time dependent ROC curves and area under the curve can be used to select predictive covariates and build better survival models.