B0986
Title: Accelerated failure time modeling with time-dependent covariates via nonparametric Gaussian scale mixtures
Authors: Sangwook Kang - Yonsei University (Korea, South) [presenting]
Ju-young Park - Yonsei University (Korea, South)
Byungtae Seo - Sungkyunkwan University (Korea, South)
Jinkwon Kim - Yonsei College of Medicine (Korea, South)
Abstract: The accelerated failure time (AFT) model is a widely utilized regression model employed in survival analysis for the purpose of examining the association between failure time and a set of covariates. The model includes a logarithmic link function and a random error term. The model can be classified as either parametric or semiparametric depending on the degree of specification in the error distribution. In many biomedical research, it is customary for covariates to be regarded as fixed and independent of time. However, in numerous instances, time-dependent covariates are frequently encountered. The focus is on a semiparametric AFT model that accounts for time-dependent covariates. It is assumed that the baseline failure time follows an infinite scale mixture of Gaussian densities, making the model highly flexible compared to models assuming a one-component parametric density. To estimate the model parameters and mixing distributions, a maximum likelihood estimation approach is employed and a feasible algorithm is proposed that utilizes a constrained Newton method. To assess the finite sample properties of the proposed methods, simulation studies are conducted. Furthermore, the application of these methods is illustrated using a nationwide population-based health screening database from Korea.
