15.14: Censoring Survival Data

Survival analysis is a statistical method used to analyze time-to-event data, often employed in fields such as medicine, engineering, and social sciences. One of the key challenges in survival analysis is dealing with incomplete data, a phenomenon known as "censoring." Censoring occurs when the event of interest (such as death, relapse, or system failure) has not occurred for some individuals by the end of the study period or is otherwise unobservable, and it might have many different reasons and patterns. This text describes the types of censoring, their implications, and the methods used to address them.

Types of Censoring

1. Right Censoring: This is the most common form of censoring, occurring when the event of interest has not happened by the study's end or when a participant is lost to follow-up. For example, if a patient in a cancer study is still alive at the study's conclusion, their exact time of death is unknown but beyond the study period. Right censoring requires special statistical techniques to accurately estimate survival times, as the actual event time is unknown but can only be inferred to be beyond the observed period.
2. Left Censoring: Left censoring occurs when the beginning of the event period is unknown. This can happen when patients enter a study after having already experienced the event. For instance, if a patient is known to have cancer but the exact onset date is unknown, the data are left censored.
3. Interval Censoring: This type occurs when the exact time of the event is unknown but falls within a certain interval. For example, if patients are checked for a health condition at intervals (e.g., every six months), the exact time of disease onset may be somewhere between two check-ups.
4. Type-I Censoring: This occurs when a study is set to conclude at a predetermined time, established by the researcher. At the end of this fixed period, any subjects who have not yet experienced the event of interest are censored. The number of events that have occurred by this time is considered a random variable.
5. Type-II Censoring: In this case, the study continues until a specific proportion of the sample has experienced the event. The observation ends after the event occurs, so only the first events are observed. The value is determined before the data collection begins, and the study continues until this number of events is observed.
6. Random Censoring: This type of censoring happens when the total observation period is fixed, but participants enter the study at different times. Some participants may experience the event, while others may not, and some might be lost to follow-up. The censoring time varies among individuals, as it is not uniformly applied.

Several statistical techniques have been developed to handle censored data in survival analysis such as Kaplan-Meier Estimator, Cox Proportional Hazards Model, and Multiple Imputation.

In survival data, censoring leads to incomplete data, and it typically occurs when subjects experience an event before or after the study ends.

Right-censoring is the most typical form, and it occurs when the subject drops out of the study before the event occurs or when the study ends before the event occurs.

For example, a clinical study on the occurrence of heart attacks is carried out for five years. If the subjects do not have a heart attack, the data is right-censored.

Left-censoring is relatively rare but can occur when the beginning of an event is unknown or when the event happens before the subjects participate in the study.

For instance, in a study of cancer recurrence following treatment, if subjects are examined five months post-treatment for recurrence, those who have a recurrence are left-censored.

Interval censoring occurs when a specific subject is studied for a period, gets lost to follow-up for a while, and returns to continue being studied.