A particular type of failure for competing risks data is modeled by a binary regression model. Several inference procedures are proposed when the true cause of failure type may be missing due to censoring. Likelihood analysis shows that the latency distribution of the failure time given the failure type plays an essential role to adjust for the censoring bias although it is useless if complete data were available. The proposed methods avoid making parametric assumptions on the latency distributions but with the price of mak-ing additions. One crucial assumption is that the follow-up is sufficient. Nevertheless methods for bias correction are proposed if this assumption is violated. Simulations are performed to evaluate finite-sample properties of the proposed estimators. The methods are applied to analyze the Stanford heart transplant data for illustrative purposes.
|Original language||American English|
|State||Published - Sep 2004|