|
Abstract:
|
Competing risks are intrinsic to failure time data e.g. cancer registry data in which individuals are exposed to multiple potential failures such as death from the primary cancer, distant metastases or non-cancer-related causes. The simplest practice has been cause-specific approach treating individuals who fail of extraneous causes as censored observations. Alternatively, we might consider subdistribution hazard to investigate cumulative incidence, reflecting "actually observed" survival or mortality pattern. Yet it requires additional caution to make inferences in the presence of competing risks when aforementioned approaches provide different results. This discrepancy may result not only from the model formulation but also from the data structure such as distributions of censoring or failure times. In this study, we aim to evaluate data-specific optimal choice for competing risks approach by presenting simulation studies under various realistic scenarios. Ultimately, we would like to provide insights on the impact of competing risks and when we should consider competing risks into statistical analysis utilizing public health data.
|
