Aplicação do pacote survstan em R para modelagem de dados de sobrevivência: uma comparação com o pacote survival

Abstract

Reliability analyses are greatly relevant in engineering, being a way of evaluating and predicting the occurrence of unwanted events and then, support processes and products improvement, aiming for greater efficiency in results. To model the behaviour of a system’s failures, it is needed to select a probability distribution for failure times, in addition to the best regression model that will explain the influence of covariates. R software is one of the tools that can support the analysis process, which would otherwise involve complex calculations. In view of this, this work seeks to demonstrate the application of survstan, a R package of functions that aims to provide a toolkit for fitting survival models. For this purpose, a dataset extracted from literature was fed as input to survstan functions, which were used to select the appropriate probability distribution, as well as fitting the Accelerated Failure Times regression model and obtaining models’ parameters of interest. The functionality of the package to perform these functions was compared with the survival package, one of the most used in R for survival analysis tasks. In comparison, survstan presented consistent results with those obtained through survival, with less effort applied in writing code and interpreting results.