The persistence of tuberculosis in the United States: Spatial analysis and predictive modeling in the move toward elimination of tuberculosis (TB)

Though tuberculosis (TB) prevalence has decreased dramatically in the United States, its continual presence remains a threat to those whose needs are often overlooked. Those already impacted by poverty are the most vulnerable to TB, and stand to bear the worst health impacts, should they contract this disease. Mathematical modeling and spatial analysis have become invaluable tools in TB surveillance monitoring and elimination efforts. In this contribution, we demonstrate the capability of employing a time series, interpolative, vulnerability model to forecasted, state-level TB prevalence in the United States by determining areas influenced by poverty, as well as existing TB data acquired from the Center of Disease Control (CDC). The random effects term in this orthogonal eigenvector spatial filter model was comprised of spatially structured and stochastic effects (that is, spatially unstructured) terms, which were substituted for diagnostic, remote, and clinical covariates in our model. It was assumed that random effects terms in the TB risk model had followed a Gaussian frequency distribution with a mean of zero. The estimate equations were as follows:   and  .  The resulting estimated number of cases for a given state and year was  .  The Moran coefficient (MC) was 0.66, and its Geary Ratio (GR) was 0.35. The spatially unstructured random effects terms have only trace levels of spatial autocorrelation, with MC = 0.02, and Gr = 0.89.  Thus, the assumption of non-zero spatial autocorrelation was violated.  The forecast revealed possible hyperendemic transmission of TB in non-coastal, Northwestern states, as well as in some Northeastern states.  As such, more intervention efforts should be directed towards these areas.   Key words: Tuberculosis (TB), poverty, center of disease control (CDC).