Design Suggestion

A competing risk, or competing event, is an event that alters the probability of the event of interest, in the context of a study in which the primary outcome is the time to the event of interest (survival time). For example, if the event of interest is death due to a cardiac cause, death due to a non-cardiac cause is a competing event. A flawed study design might regard death due to a non-cardiac cause as independent censoring rather than a competing event, implicitly assuming that an individual dying from a non-cardiac cause had, just prior to death, the same risk of death due to a cardiac cause as an individual followed for the same duration who had not yet died. Of course, this is not a reasonable assumption since non-cardiac causes of death include, for example, stroke due to atherosclerotic carotid artery disease, a condition that is clearly associated with the risk of cardiac death.

Investigators frequently ignore the effects of competing risks when conducting survival analysis to compare time-to-event outcomes between treatment groups and use methods that assume independent censoring, such as the Kaplan-Meier estimator, log-rank test, and Cox proportional hazards regression. Each of these methods assumes that any factor that interrupts observation of the time from study entry to the event of interest is unrelated to an individual’s risk for the event of interest. If this assumption does not hold, these methods are biased.

A better approach is to acknowledge the potential for competing risks when they exist and to plan to conduct a competing risks analysis, using a statistical method that properly accounts for the competing risk or risks. A general approach for competing risks analysis is based on estimation of cumulative incidence functions for the event of interest and each of the competing events. Cumulative incidence functions, if properly estimated, are unbiased in the presence of competing risks and can be compared between treatment groups using specialized non-parametric tests and regression techniques.