Supplementary Materials to “Combining parametric, semi-parametric, and non-parametric survival models with stacked survival models”

Section 1 demonstrates the connection between the Brier Score in the absence of censoring versus the inverse probability-of-censoring weighted Brier Score. Section 2 derives the meansquared error decomposition presented in Section 3 of the main paper, and presents illustrations and examples regarding the impact of candidate survival models on performance. In addition, a simple example illustrates the situation of stacking a “parametric” and “non-parametric” survival model. Section 3 proves the asymptotic properties of stacked survival models presented in Section 4 of the main paper. In addition, Section 3.3 discusses the potential of distributional results for the conditional survival function. Section 4 discusses time-dependent stacking and compares the performance to time-independent stacking [equation (2.4) in the main paper]. Section 5 considers several extensions to the simulation study in the main paper.