Empirical Likelihood Inference For The Accelerated Failure Time Model With Interval-Censored Data

Accelerated failure time model is a useful model to analyze the relation-ship between survival time and the covariate. In the field of survival analysis, Compared with the proportional hazards model accelerated failure time model is more simple, intuitive and easy to understand is the preferred method of sur-vival analysis. However, when obtaining survival data in experiment, due to the limitations of observation and the individual differences of the observer entering or exiting the experiment, the data we obtained always appear interval-censored. This article focuses on empirical likelihood inference for the accelerated failure time model with interval-censored data. The main structure is as follows:In chapter1, we analysized the research situation of accelerated failure time model with interval data and gave an outline of accelerated failure time model, handling method of interval-censored data, estimation method of re-gression parameter, empirical likelihood(EL) method and jackknife empirical likelihood(JEL) method.In chapter2, firstly, we handled interval-censored data by quoting a nov-el method; Then, estimated the unknown parameter in accelerated failure time model with interval-censored data by a method similar to generalized estimating equations; Finally, derived the expression of the confidence interval of the re-gression parameters in the model Separately by normal approximation method, EL and jackknife EL method.In chapter3, Using the randomly generated data to simulate for the three methods in the preceding chapter, we can get the Confidence intervals and coverage probability respectively and compared with each otherIn this paper, We quoted relatively simple and novel methods to hand interval-censored data and estimated the unkown parameters of the model. Es-pecially, we made jackknife EL inference for the accelerated failure time model with interval-censored data then compared with EL and normal approximation method. The simulation results show that Confidence intervals of EL and JEL are wider and coverage probability are Closer to the confidence level than nor-mal approximation method. JEL method is better than EL method in some cases and more effective in practical. Because it can be used for calculating complex statistics which empirical likelihood method can’t do it.