생존 분석에서의 절단

Left truncation occurs when individuals who experienced an event before a certain time are excluded from the study. This commonly happens through delayed entry, where only participants who survive until a specific entry point are observed. For example, in occupational studies, workers who retired or died before the study began are not included, creating bias that only represents those still at risk after entry.

Right truncation excludes individuals whose event time exceeds a specific value, such as the study period. While left truncation involves delayed entry into observation, right truncation occurs when only individuals who experienced the event by a certain time are included. For instance, a mortality study recording only deaths within a specific timeframe excludes those who lived beyond the observation period.

Truncation completely excludes subjects from analysis because they do not meet entry criteria, providing no data on them. Censoring, by contrast, provides partial information on subjects whose exact event time is unknown. For example, censoring indicates an event has not occurred up to a certain point, whereas truncation means those individuals are entirely absent from the dataset.

Time zero establishes the reference point from which all participants are observed until they experience the event or are censored. Ideally, time zero represents the true start of risk exposure, such as employment start date in occupational studies. However, when workers who left before the study began are excluded, the chosen time zero may not reflect actual exposure history, introducing left truncation bias.

Left truncation occurs in occupational exposure studies where workers hired before the study began are excluded if they left the factory previously. Disease studies also exhibit left truncation when only individuals diagnosed after a specific date are included, excluding those diagnosed earlier. These exclusions create datasets containing only survivors or those meeting delayed entry criteria.

Truncation introduces bias by systematically excluding certain individuals, which can distort group comparisons in survival analysis. When comparing the survival analysis of two or more groups, truncation may differentially affect groups if exclusion criteria apply unequally. This selective exclusion compromises the representativeness of each group and can lead to misleading conclusions about survival differences.

Truncation results in complete data loss for excluded subjects; no information exists about them in the dataset. Censoring preserves some information, such as knowing an individual survived until a specific date. This fundamental difference means truncated data cannot be recovered or analyzed, while censored data can be incorporated into survival models using specialized statistical methods.