B1289
Title: Simultaneous modeling of counts with excess zeros and left-truncated survival data with time-varying effects
Authors: Xavier Piulachs - Universite Catholique de Louvain - University of Barcelona (Belgium) [presenting]
Montserrat Guillen - University of Barcelona (Spain)
Abstract: In the context of health care studies, repeated measurements of medical claims are collected on each subject as an indirect measure of health status. At the same time, the aging process taking place in developed countries leads to an obvious interest in assessing the relationship between emergency care demand and survival rate, paying particular attention to elderly policyholders. We propose a joint model to analyze the degree of dependence between the longitudinal and time-to-death outcomes for a large cohort of Spanish policyholders aged 65 and over. The longitudinal response is restricted to a small range of non-negative integer values, and is affected by some degree of overdispersion caused by unobserved heterogeneity and an excess of zeros. Additionally, only policyholders who have reached the age of 65 come under study, so death times may be subject to both left truncation and right censoring. Finally, we argue the existence of a time-varying relationship between longitudinal and survival responses. The estimation of parameters is performed under a Bayesian perspective by Markov chain Monte Carlo simulation methods.
