|
Abstract:
|
Competing risks are mutually exclusive event types. While a marginal analysis quantifies the occurrence of one specific event type over time for the situation that the competing risks were absent, a competing risks analysis considers each event type as a possible end point. We explain when a competing risks analysis is the appropriate approach, what quantities can be defined and estimated, and how the results can be interpreted. We describe two approaches to competing risks analysis. The multi-state approach is based on the cause-specific hazard and has a direct extension to Markov multi-state models. The subdistribution approach is based on the subdistribution hazard, which is the hazard that uniquely defines the cause-specific cumulative incidence. In the regression setting, we consider models that assume proportionality on either of the hazards. Fitting such a model for the cause-specific hazard boils down to fitting a standard Cox model. The proportional subdistribution hazards model is known as the Fine and Gray model. We briefly explain how the creation of a stacked data set allows for large flexibility in modeling effects on all competing risks in a single analysis. Basic knowledge of classical survival analysis is assumed (Kaplan-Meier and Cox model).
|
