While the Bayesian hierarchical regression framework is
well established, the application of these methods for defining
CAs is novel. Existing methods for defining a CA are not
probability based, may require a priori selection of a distance
or patient flow threshold, or do not adjust easily for covariates.
In contrast with previously proposed and applied approaches
for CA analysis, the Bayesian regression models can estimate
the CA stochastically from the data using exceedance probabilities,
while adjusting for several covariates. Estimating
effects for patient demographics were beneficial for understanding
differences in the likelihood of being seen at MCC
according to patient characteristics. For example, we found
that non-white patients were significantly more likely to be
diagnosed at MCC than white patients. In addition, we could
handle different types of available data with different forms
of the regression model. Also, we could evaluate the benefit
of including a prior on random effects that assume spatial
correlation across counties by comparing a Bayesian measure
of goodness-of-fit. In our case, including a spatially correlated
prior for county effects was unnecessary because of the
strong spatial signal present in the patient data for being seen
at MCC.