Keywords: disease mapping, CAR model, mental health, Bayesian hierarchical modeling
Nearly 1 in 5 American adults suffer from any mental illness in a given year. Existing literature has identified evidence of spatial clustering of specific mental health conditions, but none have examined the clustering of mental health related hospitalizations in the US. The present analysis uses Bayesian hierarchical models with an intrinsic CAR model prior to predict the rates of inpatient hospitalizations attributed to mental disorders in zip codes in the state of Missouri. Four separated models were run, incorporating the following variables: (1) no covariates, (2) percent of population with no health insurance, (3) percent of families receiving food stamps, and (4) percent of vacant housing. All models yielded similar estimates for the rate of mental health-related hospitalizations throughout the region (around 11 per 1000 population), and the percent of families receiving food stamps and percent of vacant housing were found to be significantly associated with hospitalization rates. There was also evidence of significant spatial clustering of these rates throughout the state, with high rates found in the St. Louis, Kansas City, and “bootheel” regions, and low rates found in the northeast portion of the state. Health professionals can use these results to prioritize regions throughout the state that have the greatest need for mental health service providers and interventions.