Goodness-of-fit methods for right-censored data: theory, implementation in R, and application in accelerated failure time models.
Goodness-of-fit (GOF) techniques are an important tool to test the validity of parametric models. GOF presents a very difficult framework that has been widely studied. However, for censored data, fewer papers can be found in the literature. We present a set of GOF methods for right-censored data and their implementation in the R package GofCens, which includes GOF tests and graphical tools.
Moreover, the p-value distribution under the null hypothesis has been studied for those GOF tests for right-censored data developed and implemented in GofCens. Interesting results have been obtained showing that, even when assuming non-informative censoring, the distribution of censoring times plays an important role in the p-value distribution.
An application of the methods to the accelerated failure time model (AFTM), a commonly used model in survival analysis, will be presented. We propose a simple approach that uses standardized residuals and the methods outlined to create GOF tools for AFTMs.
Palabras clave: Goodness of fit survival analysis censored data p-values accelerated failure time models