Modelos semiparamétricos para análise de eventos recorrentes

Abstract

In areas such as medicine, public health, business, industry, reliability, social sciences and insurance, many situations arise in which the interest is to study processes that generate events repeatedly over time. These types of situations are called recurrent event processes, and the data they provide is called recurrent event data. In this context, the models proposed in the present work are, fundamentally, survival models based on the Poisson process and the renewal process, with the hazard function (or intensity) being constructed from a semiparametric perspective via Bernstein polynomials. In addition, two general classes of semiparametric models are proposed that have the above processes as particular cases, the hazards functions (or intensities) of these classes being obtained through the Bernstein polynomials and the piecewise exponential. The proposed models are flexible in the sense that they do not impose a specific form for the hazard function (or intensity), have qualities similar to those of the parametric models with regard to the estimation of the survivor, hazard (or intensity) and cumulative hazard (or intensity) functions. Some of these models do not assume proportional hazards (or intensities) and have computational characteristics that are attractive from the point of view of classical and Bayesian inference, which motivated to make inference for the models proposed under both paradigms. The analysis developed here presents the results of a simulation study aimed at investigating the behavior of the proposed models in different scenarios and also explores real data from classic studies in the literature for the analysis of recurring events.