ABSTRACT
Sometimes, the exact time that an event occurs is not known, and the only information available is that the event has occurred in a particular interval of time, such as the time between two successive examinations. This leads to interval-censored survival data, and methods of analysis are developed in this chapter. One complication is that the survivor function is not defined within certain time intervals, known as Turnbull intervals. After describing how the survivor function can be estimated, possible models for interval-censored data are introduced. These include a model where the baseline survivor function is a non-parametric maximum likelihood estimator based on Turnbull intervals, and an analogue of the Cox regression model that assumes that the hazard function is piecewise constant. Parametric models based on an assumed probability distribution for the underlying event times are also described. Current status data, a special case of interval censoring where the status of an individual is only observed at one time point, is also briefly discussed.
