Regression Analysis Of Clustered Failure Time Data Under The Additive Hazards Model

Due to the objective conditions,in many practical situations,the data we observed is often missing.The duration of the study is greater than or equal to a certain time or within a certain period of time.This incomplete data is called censored data in survival analysis.In recent years,censored data theory has been rapidly developed and widely applicated,such as in epidemiology,biostatistics,insurance actuarial and other aspects.Research methods also involve parametric,semi-parametric,nonparametric estimation methods.In the research of censorship data,left censored data and right censored data can be seen as special interval censorship.This paper is based on the II type interval censorship data.In the study of interval censorship data,there are two analytical methods commonly be used,namely Cox proportional model(1972)and additive risk model.Cox regression model can not only analyze the impact of various independent variables on survival time,but also make no demand to distribution of baseline risk function.However,The Cox proportional hazards model requires that the ratio of the risk functions of two individuals with different regression vectors does not change with time.Therefore,in practical research,the data is difficult to meet this requirement.To this end,in this paper,we use another significant model,which is applicated in a wider range,namely additive hazards model.In the actual study,clustered data often appears.For example,the failure time can be the patient's onset time from the same family or the same clinic,which is called clustered data.The paper will study the regression problem of clustered data interval.Since the failure time of the clustering data may be related to the size of the class,it is unreasonable to simply think of all the data as a class.A simple and common approach is to simplify interval-censored data due to the lack of proper inference procedures for direct analysis.For this reason,we proposed the method of within-cluster resampling.The WCR solves the problems may be related.Within-cluster resampling method is simple but computationally intensive.A major advantage of the proposed method is that the estimator can be easily implemented when the cluster size is informative.Asymptotic properties and some simulation results are provided and indicate that the proposed method works well.