A kernel method for modelling interval censored competing risks: theory and methods
Résumé
Self-consistent algorithms for modelling the cause-specific cumulative incidence functions in an interval censored, competing risks context were proposed by Hudgens et al. (2001) and Adamic (2006). One of the drawbacks of these methods was that convergence was only achieved over a small class of points, called cause-specific innermost intervals. Braun et al. (2005) proposed a self-consistent kernel modification to Turnbull's univariate algorithm as a way to fill in the gaps between the resulting innermost intervals. We generalize this approach to the plateau of interval censored and masked competing risks. The new approach is illustrated by applying the method to a human trypanosomiasis data set.