A1147
Title: Hierarchical Bayesian accelerated failure time models for clustered survival data
Authors: Dongu Han - Korea University (Korea, South) [presenting]
Daeyoung Lim - Food and Drug Administration (United States)
Jichan Park - Korea University (Korea, South)
Taeryon Choi - Korea University (Korea, South)
Abstract: The proposal is to apply a hierarchical Bayesian spectral analysis regression to accelerate the failure time model for the clustered survival data. The proposed models employ hierarchical Gaussian process priors to lift the restrictive parametric assumption and smooth the estimated functions within and between clusters. This hierarchical approach allows the provision of more accurate estimates than fitting separate models to each cluster when data within clusters are sparse. A Dirichlet process mixture model is investigated with a Weibull base measure for clustered survival data. An efficient Markov chain Monte Carlo algorithm is presented, and it describes how to approximate functional estimands such as hazard and residual life functions that are of interest to researchers as well as practitioners. A theoretical justification is provided for the proposed method through weak posterior consistency. An extensive empirical study based on artificial and real datasets has been conducted to assess the performances of the proposed models.
