2010 Volume 22 Issue 1 Pages 3-21
In survival analysis, the useful tool for exploration of the factors is tree structured method. Eto et al. (2007) have propose the multi-split tree structured method based on k-samples generalized rank test statistics (MUSTGRAS). However, these nonparametric approaches do not have consistency in data analysis process. Then, we proposed the methodology of data-adaptive multi-split tree structured method (DAMUST), assuming the power-normal distribution as the survival distribution of each terminal node, where power-normal distribution is defined as the distribution specified before the power-normal transformation. We evaluated the performance of the DAMUST by some practical examples with survival data and small scale simulation. As a result, DAMUST has better performance than MUSTGRAS. On the other hand, we can evaluate the survival distribution for each terminal nodes based on the power normal distribution using DAMUST by way of simulation, clinical study, etc. Consequently, DAMUST is better useful method than MUSTGRAS.