ABSTRACT
The methods described in this book for the analysis of censored survival data are only valid if the censoring is independent or non-informative. Essentially, this means that the censoring is not associated with the actual survival time, so that individuals with censored survival times are representative of all others who are at risk at that time, and who have the same values of measured explanatory variables. For example, censoring would be considered to be independent if a censored time occurs because a patient has withdrawn from a study on relocating to a dierent part of the country, or when survival data are analysed at a prespecied time. On the other hand, if patients are withdrawn because they experience life-threatening side eects from a particular treatment, or if patients that do not respond to an active treatment are given rescue medication, the censoring time is no longer independent of the survival time. The censoring is then said to be dependent or informative. This chapter shows how dependent censoring can be taken account of when modelling survival data.
