An approach for constructing survival tree based on combination of covariates

We propose a new approach for constructing a survival tree which is used to analysis of time to event data. The construction of survival tree is performed by the classification and regression tree (CART) algorithm, which constructs a tree model by recursively dividing the data. In the traditional approach, internal-node data are divided by a splitting rule which consider only one covariate for each node. In the proposed approach, two or more combinations of covariates are considered for dividing the internal-node data in the algorithm. We show that the proposed method is more suitable than the traditional approach in some situations through the comparative research by simulations. We also present the result of an actual analysis based on proposed approach.