Testing treatment effects in the presence of competing risks

Competing risks are often encountered in clinical research. In the presence of multiple failure types, the time to the first failure of any type is typically used as an overall measure of the clinical impact for the patients. On the other hand, use of endpoints based on the type of failure directly related to the treatment mechanism of action allows one to focus on the aspect of the disease targeted by treatment. We review the methodology commonly used for testing failure specific treatment effects. Simulation results demonstrate that the cause-specific log-rank test is robust (in the sense of preserving the nominal level of the test) and has good power properties for testing for differences in the marginal latent failure-time distributions, whereas the use of a popular cumulative incidence based approach may be problematic for this aim.