A Review of R Packages for Bayesian Model-based Clustering of High-dimensional Multivariate Environmental Exposures

PURPOSE OF REVIEW: The analysis of a collective set of environmental health exposures in large populations can be complex due to its dimensionality, correlation between exposures, and skewed data structures. Bayesian model-based clustering has gained traction for its computational advantages and model flexibility for the analysis of big data. This review aims to highlight R packages available to derive environmental exposure patterns in heterogeneous populations using a Bayesian mixture model framework. RECENT FINDINGS: Twenty-one packages have been developed since 2006 with the capability of analyzing multivariate exposures, with more than half being released in the last ten years. Packages are available to identify both cross-sectional and longitudinal exposure patterns from a joint set of continuous, binary, categorical, directional, and ranked data types. Bayesian nonparametric priors are available to determine the appropriate number of latent clusters to fit. Variable selection is also available in some packages to focus primarily on exposures that may drive underlying patterns present. Bayesian mixture models are an attractive solution to analyze joint environmental exposures and its subsequent impact on population health. While software packages have been developed to provide researchers with advanced tools and models suitable for data complexities seen in environmental exposure data, their applications in the literature is significantly low. Further awareness of software packages and their capabilities could increase utilization and improve analysis of environmental exposures and population health.