Decision trees have been recently introduced for survival data, including data with competing risks. The prognostic ability of these trees is not commonly examined. In this dissertation, we introduce a novel approach to building survival decision trees. We test the prognostic ability of decision trees constructed using inverse probability of censoring weighted prediction error curves on data from the National Burn Registry. Lastly we develop a non-parametric inference framework for comparing the prognostic ability of different prediction procedures. |