Um estudo de modelos de riscos competitivos em análise de sobrevivência

Abstract

Statistical analyzes that involve recording times and the occurrence of one or more events of interest are known as survival or reliability analysis. When analyzing data sets of this nature, one of the objectives is to estimate the survival or reliability function of the sampling units (people, systems, components, etc.). Examples of times and events that generate such times are: times until washing machines stop working, times until death/recovery of patients presenting with a certain disease, time until children learn to write, etc. Among some situations that can occur in this type of study, we highlight two: (a) not all sampling units suffer the event of interest (censorship) and, (b) the event can occur for several reasons (competitive risks). In both cases, this information can be incorporated into statistical analysis.

From records of complete times, censoring, types of cause and covariates, it is possible to adjust a model to the data, and also statistics or functions of interest. In this work the objective is to estimate the survival function of times in which a certain event occurs, considering competitive risks, under non-parametric approaches, by the Kaplan-Meier estimate and its modifications and, parametric, by the Cox model and the model proposed by Fine and Gray. Furthermore, show in which situations they are best used, as well as their advantages and disadvantages.