Modelo de mistura padrão com tempo de falha exponencial e censura informativa

Abstract

In this work we consider the long-term survival model introduced by Berkson & Gage (1952), for modeling survival data of nonhomogeneous populations, where a subpopulation does not present the event of interest, despite a long follow-up period. The cure rate models presented in the literature usually are developed under the assumption that censorship is noninformative. In the usual survival models Lawless (1982) considers that the variable of censoring is informative if its density function and its distribution function involve some parameter of interest. We propose a new definition of informative censoring in a similar way. This de_nition is extended for the unified long-term survival models (Rodrigues et al., 2009). Moreover, we verify, with simulated data, the impact caused by informative censoring in the coverage probabilities and in the lengths of asymptotic confidence intervals of the parameters of interest. A Bayesian approach with Jeffreys prior is also proposed. An example with real data is analysed.