Applications of competing risks analysis in public health

Abstract

In medical and public health research, survival statistics are of particular interest as they can reflect patient prognoses and improvements in health care systems. However, measures of survival differ in their use and interpretation depending on how they deal with competing causes of death. Cause-specific survival estimates survival function based on the event of interest while treating other events as censored, thereby representing the "net" impact of cancer on survival. On the other hand, cumulative incidence function and subdistribution hazard consider the number of subjects experiencing competing risks in their formulations, thereby providing measures for investigating the patients' actual prognoses. In this paper, we review competing risks survival models used in public health study. We introduce the concept of competing risks methods and compare these with traditional net approaches (e.g. relative and cause-specific). We demonstrate how competing risks analysis can be used in population-based cancer survival analysis utilizing the Surveillance, Epidemiology, and End Result (SEER) cancer registry data. As the scope of public health study has extended beyond prognosis and risk prediction, competing risks analysis has been applied in such studies as well. We further discuss the uptake of the competing risks approach in personalized and precision medicine. Various methods and applications used in risk predicting prognostic models with competing risks are reviewed, aiming to provide effective analytical tools for researchers who plan to implement competing risks models on public health studies.