Regression Analysis for the Cause of Failure Type under Competing Risks

Abstract

In this talk, we are interested in a particular type of failure for competing risks data. Under a mixture formulation, the model contains two components. One component is the incidence rate of the event and the other is related to the latency distribution which describes the time to the event of interest given that it would occur. The focus here is on statistical inference of the incidence rate which measures the probability of the particular type of failure. A generalized linear model, such as logistic regression, is imposed and the major objective is to estimate the regression parameters when the true cause may be missing due to censoring. Despite of less interest the latency distribution plays an essential role if likelihood-based inference is pursued. However we avoid making parametric assumptions on the latency distribution by directly modifying the score equation. Several estimators are proposed which use techniques of inverse probability weighting and imputation to adjust for the sampling bias. Keywords: Imputation; Inverse Probability Weighting; Missing Data; Random cure time model.