Premium
Cumulative incidence analysis and relative survival
Author(s) -
KIVELÄ T
Publication year - 2008
Publication title -
acta ophthalmologica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.534
H-Index - 87
eISSN - 1755-3768
pISSN - 1755-375X
DOI - 10.1111/j.1755-3768.2008.3243.x
Subject(s) - proportional hazards model , cumulative incidence , survival analysis , incidence (geometry) , regression analysis , demography , event (particle physics) , population , confounding , medicine , statistics , hazard ratio , relative risk , confidence interval , cohort , mathematics , environmental health , geometry , physics , quantum mechanics , sociology
Purpose To highlight concepts related to competing events in time‐to‐event data sets. Methods Introduction to cumulative incidence and relative survival analyses and competing risks proportional hazards regression with examples from recent literature. Results Kaplan‐Meier and Cox regression analysis were designed to study mortality. They return biased estimates in the presence of competing events that render subjects immune to the event of interest (e.g. one is no longer at risk of vision loss, bleb failure or graft rejection after dying). Kaplan‐Meier can then be supplemented with cumulative incidence analysis and Cox with competing risks regression. The data needed are time‐to‐event or last follow‐up, last status (e.g. experienced an event, under follow‐up, lost to follow‐up) and explanatory or confounding variables. Subjects who experienced a competing event are treated as such and subjects who did not experience any event are “censored” at last follow‐up. A set of stepped curves is produced which show the cumulative incidence of each event as a function of time by study group; groups can be compared using dedicated tests. Competing risks regression provides a hazard ratio, adjusted for the effect of other variables in the model. Relative survival is an alternative to cumulative incidence method when analyzing mortality. It does not require that the status at last follow‐up be known. Survival of the study group is compared with that of the underlying population. The difference is equivalent to the cumulative incidence of disease‐specific death, but cumulative incidences of competing events are not available. Conclusion After this talk, participants should be able to recognize competing events, assess whether Kaplan‐Meier and Cox regression were appropriate methods and know alternatives to them.