Survival - Time to Failure

Analysis of survival data is more complex than than other methods we've seen so far; We can't just take the mean survival time a a confidence interval to predict when the last patient will die. Also, survival times are unlikely to follow a Normal distribution, so simple regression techniques are not applicable.

Censoring

Censoring of data can happen in a number of ways:

1. Some subjects still have not experienced the event of interest by the end of the study. Most studies have a recruitment period followed by a pure follow-up period. Patients who are enrolled earlier have a higher chance of experiencing the event.
2. Some participants drop out early, either due to dropping out or another event unrelated to the study.

In either case, you know that the subject has participated in a study to a certain time without witnessing the event, but have no information thereafter. Such incomplete information is said to be right censored. Note that we censor the data, not the subject. Throughout this lecture we will assume independent censoring - the participants remaining in the study are representative of the target population.