Applications Of Multiple Imputation To Interval-Censored Data

The problem of analyzing interval-censored data has become increasingly common in many applied fields,such as medicine, biology, epidemiology, engi-neering, economics,and demography. When the exact time of interest cannot be observed,but a certain interval in which it falls is known,we call this type of data as interval-censored data. Especially, when time of interest has been known to be greater than (or less than) a given time,and its exact value is unknown, we call it right(left)-censored data.For right-censored data, statisticians have written a number of books that provide varied and effective statistical analysis approaches deeply and widely. Although right-censored failure time data can be treated as a special case of interval-censored data, and some of inference approaches for right-censored data can be directly, or with minor modification (generalization),applied to the anal-ysis of interval-censored data, most of them for right-censored data are not ap-propriate for interval-censored data due to the complexity and special structure of interval censoring. In recent decades,a voluminous literature on the statistical analysis of interval-censored failure time data has appeared,but most of effective methods are very complicated.In this paper,we apply multiple imputation(MI) approach to the problem of estimating regression coefficients in accelerated failure time model with interval-censored data based on Tanner and Wong(1987) data augmentation algorithm. This method consists of iterating two steps:First,exact failure times are im-puted from finite interval-censored(but not right-censored)data based on the current estimates of the regression coefficient and the error distribution.Second, a standard statistical procedure for right-censored data is applied to imputed data to update the estimates.A major advantage of the approach is its sim-plicity because the estimation procedure can be easily implemented by taking advantage of existing statistical software packages for right-censored data.