Handling of Tied Failures in Competing Risks Analysis

Abstract

In a competing risks framework where a subject is exposed to more than one cause of failure, multiple or tied `first¿ failures may be detected simultaneously at a particular follow-up because of periodical investigation. Standard competing risks analysis methods such as cause-specific hazard regression may not be applied satisfactorily when considerable information is missing due to the unknown cause of failure. This study develops a shared frailty model to identify the `true¿ cause of failure in the event of ties by adopting a cause-specific hazard regression model for each failure type with assumptions of exponential failure time distribution and gamma frailty. The MLE method is used to estimate parameters of interest, and standard cause-specific hazard regression may be implemented to evaluate covariate effects after ties are broken. Simulation studies demonstrated accuracy, efficiency and robustness of the proposed method in parameter estimation and improved accuracy in identifying unknown cause of failure compared to existing methods. The model was applied to a randomised clinical trial to compare a double-drug regimen versus a multiple-drug regimen in treatment of paediatrics with osteosarcoma, and found no significant difference between these two treatments.