Spatio-Temporal Hierarchical Models for Analyzing Atlanta Pediatric Asthma ER Visit Rates

Abstract

Prevalence and mortality rates of asthma among children have been increasing over the past twenty y ears, particularly among African American children and children of lower socioeconomic status. In this paper we i n vestigate the link between ambient ozone and pediatric ER visits for asthma in the Atlanta metro area during the summers of 1993, 1994, and 1995. Our statistical model allows for several demographic and meteorologic covariates, spatial and spatio-temporal autocorrelation, and errors in the ozone estimates (which are obtained from a kriging procedure to smooth ozone monitoring station data). As with most recent B a yesian analyses we employ a MCMC computing strategy, highlighting convergence problems we encountered due to the high collinearity o f s e v eral predictors included in early versions of our model. After providing our choice of prior distributions, we present our results and consider the issues of model selection and ade-quacy. In particular, we ooer graphical displays which suggest the presense of unobserved spatially varying covariates outside the city o f A tlanta, and reveal the value of our errors-in-covarates approach, respectively. Finally, we summarize our discuss limitations of (and possible remedies for) both our data set and analytic approach, and compare frequentist and Bayesian approaches in this case study.