Composite Endpoints in the Semi-Competing Risks Setting
*Ina Jazic, Department of Biostatistics, Harvard T.H. Chan School of Public Health
Keywords: semi-competing risks, composite endpoints, illness-death model, palliative care
Considerations about palliative care for patients with terminal illness have become increasingly prominent in health policy research. When evaluating the impact of interventions on quality of life in these populations, it is crucial to correctly handle the competing risk of death. This exemplifies the semi-competing risks setting, where primary interest lies in a non-terminal event, such as hospital readmission, whose occurrence is subject to a terminal event. In this context, investigators often choose to use composite endpoints, which combine the non-terminal event and death with the goal of (1) boosting event rates and, thereby, the power to detect a treatment effect, and (2) accounting for death as a competing risk. Here, we present simulations showing that composite endpoints may, in fact, dilute power if component outcomes have different rates and treatment effects. Either way, except in trivial situations, simply including death in a composite endpoint changes the quantity being estimated. To avoid this, we illustrate the use of the illness-death model proposed by Xu et al. (2010), which appropriately accounts for death without altering the scientific question.