Modelos para dados de sobrevivência multivariados com censura informativa

Abstract

To use the most of the methods presented in the literature to analyze survival data, it is necessary to make the assumption that the mechanism generating the censoring is noninformative, that is, assume that the distribution of the censoring times does not depend on any parameter of the lifetime distribution. However, in many situations, this assumption can be inadequate and lead to misleading inferences. With the objective to accommodate the dependence between lifetime times and informative censoring, we consider the approach provided by the frailty models, specically the approach proposed by Huang & Wolfe (2002). First, we propose a fully parametric version of the Huang & Wolfe (2002)s model, using the Weibull distribution to adjust the lifetime times and informative censoring times. Next, we also propose a semiparametric version, considering the Piecewise Exponential distribution to model the lifetime and censoring times. For these two models, we show the estimation steps via maximum likelihood and Bayesian approaches. Later we propose a Bayesian version of the Huang & Wolfe (2002)s model. Informative censoring can also inuence the estimates obtained by the models to analyze multivariate survival data with cure rate. In order to do so, we propose a fully parametric model capable of capturing the dependence between promotion times and informative censoring times, in which we use the Weibull distribution to adjust promotion times and informative censoring times. In addition to this model, we also propose a semiparametric model for cure rate data with informative censoring, using the Piecewise Exponential distribution to model the promotion times and the censoring times. We present the steps for estimation through the maximum likelihood and Bayesian approaches. Monte Carlo simulation studies were performed considering three scenarios: data generated with a positive, negative and null correlation between lifetime and censoring times for data with and without cure rate. An application is carried out on mortality data in renal dialysis centers in the United States, provided by the study called Dialysis Outcomes and Practice Patterns Study (DOPPS), using the models proposed with informative censoring. Finally, an application is presented in data on melanoma cancer, provided by the Surveillance, Epidemiology, and End Results (SEER) program, using the proposed models with informative censoring and cure rate