Censure des données de survie

Right-censoring occurs when the event of interest has not happened by the study's end or when a participant is lost to follow-up. This is the most common censoring type in survival analysis. For example, in a five-year clinical study on heart attacks, subjects without events are right-censored. Right-censored data requires special statistical techniques to accurately estimate survival times.

Left-censoring occurs when the beginning of an event period is unknown, typically when participants enter a study after already experiencing the event. Unlike right-censoring, where the event time is unknown but beyond the study period, left-censoring means the exact onset date is unknown. Left-censoring is relatively rare but can occur in studies like cancer recurrence monitoring where subjects are examined months post-treatment.

Interval censoring occurs when the exact time of an event is unknown but falls within a specific time range. This happens when subjects are studied, lost to follow-up temporarily, and then return to continue being studied. For example, if patients are checked for a health condition every six months, the exact disease onset may fall between two check-ups, creating an interval of uncertainty.

Type-I censoring occurs when a study concludes at a predetermined time set by the researcher, with any subjects not experiencing the event being censored at that fixed point. Type-II censoring continues until a specific proportion of the sample experiences the event, with the study ending after that predetermined number of events is observed. Both types differ in whether time or event count determines study conclusion.

Random censoring occurs when participants enter a study at different times while the total observation period remains fixed. Some participants may experience the event, others may not, and some might be lost to follow-up. The censoring time varies among individuals since it is not uniformly applied, making this type particularly common in real-world studies where enrollment is staggered.

Censoring creates incomplete data because the event of interest has not occurred for some individuals by study end or is otherwise unobservable. This incompleteness makes it difficult to determine exact survival times. Statistical techniques like the Kaplan-Meier Estimator and Cox Proportional Hazards Model have been developed to handle censored data and accurately estimate survival times despite these gaps.

Several statistical techniques handle censored data, including the Kaplan-Meier Estimator, Cox Proportional Hazards Model, and Multiple Imputation. These methods allow researchers to work with incomplete data and make valid inferences about survival times. When comparing groups, techniques like comparing the survival analysis of two or more groups help determine whether survival differences are statistically significant.