Bayesian Mixed Hidden Markov Models: A Multi-Level Approach to Modeling Outcomes with Misclassification

• Activity Number: 422
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yue Zhang*+ and Kiros Berhane
• Companies: University of Southern California and University of Southern California
• Keywords: Mixed Hidden Markov Model (MHMM) ; Multi-Level Model ; Misclassification ; MCMC ; Bayesian ; Asthma

Abstract

Questionnaire-based health status responses are often prone to misclassification. When studying the effect of risk factors on such responses, ignoring the possible misclassifications may lead to biased effect estimates. Analytical challenges posed by these misclassified responses are further complicated when simultaneously exploring the factors for both misclassification and health process in a multi-level setting. We propose a fully Bayesian Mixed Hidden Markov Model (BMHMM) for handling differential misclassification in discrete responses in a multi-level setting. The BMHMM generalizes the Hidden Markov Model (HMM) by introducing random effects into three sets of HMM parameters for prevalence, transition and emission probabilities, allowing for cluster level heterogeneity under a multi-level model structure. An extensive simulation study is undertaken to illustrate the gains from properly accounting for the misclassification. We apply our method to the Southern California Children's Health Study, where questionnaire based information on asthma diagnosis in children may be observed with misclassification. Risk factors for both asthma transition and misclassification are examined.