Modelos de sobrevivência para dados correlacionados na presença de riscos competitivos e censura intervalar

Abstract

In the classic survival analysis approach, a single cause is considered for the occurrence of the event of interest, however, there are situations in which several causes of failure are possible, but only the occurrence of the first of them can be observed. When an individual is at risk of failing due to different types of causes, these different types are called competitive or competing risks. Many clinical studies involving competing risks data are often subject to interval censoring. This means that the failure time is not observed precisely, but is only known between two observation times, such as clinical visits. In several studies correlated data are present and appropriate statistical analysis requires that these correlations be considered through frailty modeling or by adjusting the intra-cluster correlation in a marginal model. However, studies investigating correlated data in the presence of competing risks and interval censoring were not found in the literature. In this context, we propose a parametric regression model in which the accumulated incidence function is modeled using cause-specific failure rate functions. Our second contribution consists of a semiparametric cause-specific regression model using Taylor series expansion to approximate the basal failure rate function. In both models, to accommodate the presence of clusters, we use a Generalized Estimation Equation (GEE) model with an independent work matrix and thus, a sandwich variance estimator is used to adjust the correlation within the cluster. A Monte Carlo simulation study was conducted and indicated good performance in terms of inference in both models proposed in this work. Another analysis was conducted on a set of real data on dental trauma from the Dental Trauma Program of the UFMG School of Dentistry.