A Hierarchical Bayesian Markov-Dependent Model for Lifetime Persistence and Recurrence of Major Depressive Episodes

• Activity Number: 628 - Complex Data Analysis with Mental Health Applications
• Type: Contributed
• Date/Time: Thursday, August 2, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Mental Health Statistics Section
• Author(s): Chenyang Gu* and Alan Zaslavsky and Ronald Kessler
• Companies: Harvard Medical School and Harvard University Medical School and Harvard Medical School
• Keywords: Hierarchical Bayesian modeling; Major Depressive Episode; Markov-dependent Model; Markov Chain Monte Carlo; World Mental Health surveys

Abstract

In the World Mental Health (WMH) surveys, respondents who had experienced major depressive episodes (MDE) reported the first and last years of these episodes and the number of intervening years in which they occurred. The research question of interest is to estimate the lifetime persistence and recurrence of MDE. We develop a hierarchical Bayes estimation methodology for the analysis of two-state Markov chains observed from these heterogeneous respondents. The developed model provides a natural framework for estimating the transition probabilities while accounting for such heterogeneity by assuming that the logit transformation of two transition probabilities corresponding to each respondent's Markov-dependent sequence arise from a common, bivariate normal distribution and incorporating the respondent characteristics. We use Markov Chain Monte Carlo methods to obtain posterior inferences. The proposed method is illustrated by using a WMH survey dataset of 6,685 respondents in 28 countries who reported on episodes of MDE.