Bias by censoring for competing events in survival analysis

Abstract

In survival analysis, competing events preclude the occurrence of the event of interest. The censoring of competing events is common in medical studies but leads to biased cumulative incidence estimators. Competing risks methods, such as the non-parametric Aalen-Johansen method or the semi-parametric Fine and Gray model, alleviate this bias and should be preferred above the Kaplan-Meier method and the Cox model, respectively. As an illustrative example, in a large European cohort, we report on the differences in the cumulative incidence estimates of graft failure after kidney transplantation, caused by censoring for recipient death.