1995 年 25 巻 1 号 p. 81-96
Cox's regression model is widely used for analyzing survival data under competing risks as well as for single risk. We consider situations in which failure times are observed but failure types cannot be observed for some individuals. For this kind of incomplete data we propose a two-stage estimator of regression coefficients of covariates based on a pseudo-partial likelihood. We show that this estimator is consistent as the sample size tends to infinity. Simulation studies demonstrate that the proposed estimator for the regression coefficients gives smaller mean square errors than a conventional procedure which regards data with missing failure types as censored when sample sizes are small.