Administrative and artificial censoring in censored regression models

Administrative censoring, in which potential censoring times are known even for subjects who fail, is common in clinical and epidemiologic studies. No netheless, most statistical methods for failure-time data do not use the in formation contained in these potential censoring times. Robins has proposed two approaches for using this information to estimate parameters in an acc elerated failure-time model; the methods generally require the analyst to t reat as censored some subjects whose failure time is observed. This paper p rovides a rationale for this 'artificial censoring', discusses some of its consequences, and illustrates some of these points with data from a randomi zed trial of breast cancer screening. Copyright (C) 2001 John Wiley & Sons, Ltd.