Analysis of failure time data under competing censoring mechanisms

We derive estimators of the survival curve of a failure time in the presence of competing right censoring mechanisms. Our approach allows for the possibility that some or all of the competing censoring mechanisms are associated with the end point, even after adjustment for recorded prognostic factors. It also allows the degree of residual association to be possibly different for distinct censoring processes. Our methods generalize from one to several competing censoring mechanisms the methods of Scharfstein and Robins. Copyright 2007 Royal Statistical Society.