Survival analysis

In survival analysis, the outcome corresponds to the time elapsed until an event of interest occurs. Some patients will observe the event while other patients may not. When the follow-up of the study is over, remaining patients are censored. Some patients leave the study early (i.e., before the end of follow-up), they are lost to follow-up and considered censored at their last visit. The event is usually death, but other events of interest (e.g., progression of a disease) can be analyzed with survival models. The shape of the distribution of survival times justifies the requirement for specific models, because survival times are always positive, they often have skewed shapes of distribution and thus, statistical methods that rely on normality are not directly applicable and may produce invalid results with survival data. However, a suitable transformation of the event times, such as the logarithm or the square root can help with this issue. The main specificity in the analysis of survival times is censoring (i.e., the event is only observed for some patients).