Test Of Independence For Survival Data Based On The Censored Mean-Variance Index

In survival analysis,survival time is living time of patient which cannot be fully observed due to various reasons,so survival time may be censored which is represented as observed time and censored status.The censored survival time is mostly rightcensored which is interval data with a lower bound but no upper bound,so we mainly focused on the survival data with right-censored.And we focused on independence test which uses survival data with right-censored to test the relationship between survival time and other factors.The test can be used to judge the effectiveness of treatments and view the characteristics related to the disease,which has great practical significance.There have been many literatures which are roughly divided into two categories.The first is using survival models or survival curves to construct a parameter or nonparametric test.The second is converting the right-censored data into uncensored.But at present,few methods based on indexes without model assumptions and suitable for different variable types have been found.Therefore,we proposed a censored meanvariance independence test based on the censored mean-variance index proposed by Zhong et al.[1]to test the independence between survival time and other variables using right-censored survival data.The censored mean-variance index can measure the relationship between survival time and other variables using right-censored survival data,so the proposed test statistic based on this index can inherit its good characteristics.The test uses permutation test to realize the test process,and calculates the p-value to judge the significance of the null hypothesis.For univariate,the test can be applied to various variable types and model assumptions.For multivariate,the combination of principal component analysis makes the test suitable for multivariate condition.This thesis simulated the type-I error and power of the test under different model assumptions,and compared it with other tests.Then we applied the test to the actual dataset to test the significant variables and explain the actual meaning.Finally,it is concluded that the test can use right-censored survival data to effectively test the relationship between survival time and other variables under different model assumptions,and it is robust.