Análise de sobrevivência na presença de censura informativa

Abstract

In survival analysis data modeling, the most of procedures lead to the non-informative censoring, i.e, the event and censoring time are independent (the censoring time distribution parameters does not carry any information about the failure time distribution parameters).However, in some cases the non-informative censoring assumption is violated and thus it is necessary to include a structure which takes into account the dependence between T and C. In this work, the modeling consider the assumption the T and C are conditionallyindependents given a frailty Z, which follows a gamma distribution with location and scale parameters equal to . To this case (and the default case), we assume parametric models such as Exponential e Weibull for the times variables. By incorporating the frailty Z in the modeling, the maximum likelihood estimators are found by maximizing the marginallikelihood function. We perform a Monte Carlo simulation study to evaluate the informative and usual models under scenarios in which the censoring mechanism is informative and noninformative. The analysis of a data set regarding the patients admitted to HC/UFMG, aftera bone marrow transplantation, from may 2010 to june 2011.