In the most extreme case, a participant could be lost to follow-up, and hence censored, because failure is about to occur. Censoring might therefore be thought to be indicative that the participant is more likely to subsequently fail more quickly because, for example, dropout could be associated with a deterioration in health and hence also associated with failure.
One reason for this is because dropout is a common reason for censoring. However, this assumption is untestable and will often be doubtful for individuals censored before the scheduled end of the study.
Standard software assumes independent censoring, conditional on the covariates in the analysis model. This type of censoring is called right censoring and will occur if individuals are still at risk of failure at the scheduled end of the study, but often a non-trivial proportion of participants will be right censored before this time. Typically, the analysis is complicated because the failure times are unobserved for a proportion of individuals instead, we record the last time that they were under observation, known as the censoring time. Models for survival analysis 1 – 4 are very commonly applied to time-to-event data in medical statistics.
0 kommentar(er)
