Use of spatial statistics to identify and test significance in geographic disease patterns

Abstract

Spatial analysis provides a powerful method to test epidemiologic hypotheses about patterns of disease occurrence. Available techniques can be classified based on the type of data which they utilize, generally areas or points, and the primary question of interest. Three issues relevant to epidemiology are: whether a disease is clustered; whether two diseases or a disease and potential risk factor have the same distribution; and if there are specific definable relationships between the values of the same variable at different locations. Join count statistics, which relate actual and expected number of joins between areas with dissimilar values, and second order analysis, which compares the actual and expected distances between all points weighted by their values, give estimates of the magnitude and statistical significance of clustering in patterns. To test for codistribution between areal patterns, the kappa statistic evaluates the degree of pattern overlap corrected for chance. Tjosteim's statistic measures the correlation, corrected for point locations, of the ranked values. Spatial autocorrelation analysis can be used to test for specific network and distance associations between values, for the scale of a pattern, for defined complex spatial relationships or to remove spatial effects from more traditional regression analysis of other risk factors. Epidemiologists have uncovered spatial associations between diseases and risk factors using traditional methods. However, more widespread application of spatial analyses to objectively quantify the magnitude and significance of hypothesized geographical associations could offer new insights into disease questions.