B1373
Title: Infinite hidden Markov models for multiple multivariate time series with missing data
Authors: Ander Wilson - Colorado State University (United States) [presenting]
Abstract: Exposure to air pollution is associated with increased morbidity and mortality. Recent technological advancements permit the collection of time-resolved personal exposure data. Such data are often incomplete, with missing observations and exposures below the limit of detection, which limits their use in health effects studies. An infinite hidden Markov model is developed for multiple partially or non-overlapping multivariate time series with missing data. The model is designed to include covariates that can inform the allocation of time points to hidden states. Beam sampling is implemented, a combination of slice sampling and dynamic programming, to sample the hidden states and a Bayesian multiple imputation algorithm to accommodate missing data. In simulation studies, the model excels in estimating hidden states and state-specific parameters and imputing observations that are missing at random or below the limit of detection. The imputation approach is validated on data from the Fort Collins commuter study. The estimated hidden states improve imputations for data that are missing at random compared to existing approaches. In a data analysis of the Fort Collins commuter study, the inferential gains obtained from the model are described, including estimating state-specific parameters that characterize exposures better than manually assigned microenvironments and identifying hidden state trajectories that are shared among repeated sampling days for the same individual.
