January 10th, 2018
Censoring is a common practice in survival data analyses. Survival data concerns the recording of the time it takes for a particular event of interest to occur (Antonisamy, Christopher & Samuel 2010, p. 214). In most survival-data studies, survival times (i.e. time before occurrence of the event of interest) of some of the study subjects may never be known with certainty. Such for instance could result where the study ends before some of the subjects depict the outcome of interest or a subject drops out of the study out of reasons unrelated to the occurrence of the event under interest (Antonisamy, Christopher & Samuel 2010, p. 215). Accordingly, the data collected for such subjects, referred as censored data, is that which was recorded up to the end of the study or up to the subject dropping out of the study, rather than up to the time of developing the event of interest. This paper discusses the types of censoring and the challenges they present for data analysis.
Right censoring. Right censoring occurs where the researcher knows the time of start of monitoring but does not know the exact time of occurrence of the event of interest in some of the study subjects; therefore the recorded observed survival time lower than the actual survival time (Antonisamy, Christopher & Samuel 2010, p. 214). Right censoring can be of three types, type I, Type II and type III (random censoring) (Lee & Wang 2003, pp. 2-4).
Type I censoring. Type I censuring occurs where the event of interest is observed only when it occurs before a predetermined time (Klein & Moeschberger, 2003, p. 64). Accordingly, where there are no subjects lost during the period of the study, in type I censoring, all censored observations will be equal to the study period’s length. However, subjects could have different predetermined censoring (fixed-sacrifice) times, in which case the resultant censoring would be progressive type I censoring (Klein & Moeschberger, 2003, p. 65).
Type II censoring. Type II censoring occurs where the fixed-sacrifice times are determined according to the number of subjects who achieve the event of interest. When such a number is reached, the remaining subjects, who have not achieved the event, are sacrificed with their recorded survival times being equal to the largest uncensored observation (Lee & Wang 2003, p. 2). A progressive type II censoring could also occur where after reaching the number of subjects required for the first censoring, a specific proportion of the subjects who have not achieved the event are censored (Klein & Moeschberger, 2003, p. 69). The remaining subjects then are observed until a second predetermined number of subjects achieves the event, necessitating a second round of sacrificing a proportion of the subjects who still have not achieved the event (Klein & Moeschberger, 2003, p. 69). Such a series of events continues until a predetermined number of cycles is completed.
Type III (Random) Censoring. Random censoring occurs where a competing event, other than the event under interest, leads to the removal of the subject from the study (Klein & Moeschberger, 2003, p. 69). Such could for instance arise where subjects are lost to follow-up. Accordingly, the researcher sacrifices such subjects out of unplanned events, which arise randomly. Accordingly, the observed survival values for such subjects will be up to the time of in which the competing event results into removal from the study.
Go to part two here.