|
Abstract:
|
We consider estimation of the cumulative incidence functions for competing risks survival data subject to current status censoring. We show that the large-sample behavior of the nonparametric MLE depends fundamentally on smoothness assumptions that one makes about the observation time distribution. As a result, also the large-sample confidence intervals (CIs) depend on these smoothness assumptions. In practice, however, it is not always clear which smoothness assumptions are most appropriate. We therefore propose a subsampling-based method to construct CIs which automatically adapt to the smoothness, and yield the desired finite sample coverage under various smoothness conditions. We evaluate our methods in simulation studies and on two real data sets regarding i) age and type menopause of women in the US, and (ii) subtype-specific HIV infection in injecting drug users in Thailand.
|
