Abstract
Advances in the availability of geographically referenced health and environmental quality data of high spatial resolution have created new opportunities in environmental epidemiology. Novel statistical methods for linking health, exposure, and hazards are required to underpin the development of public health tracking. A test for the association between spatial contours of health risk and exposure is outlined. This test is examined using, as an example, the spatial contours of congenital malformation risk obtained from a routine dataset in the vicinity of a landfill site and an exposure model based on exponential reduction with distance from the site. Spatial contours of risk of congenital malformation were simulated using the exposure model stated and a given population pattern. These were compared with the corresponding expected risk derived from routine birth data to yield relative risk contours. For each simulation three test statistics were devised: the slope of the regression line of standardized relative risk on exposure level, the proportion of standardized relative risks above zero, and the mean standardized relative risk of individuals not subject to exposure. The distributions of these test statistics (under the null no exposure from site and alternative hypotheses) were determined from a simulation exercise. A comparison of receiver operator characteristic (ROC) curves between those relating to the proposed test and those relating to a widely used method proposed by CitationStone (1988) demonstrated our test to be more efficient. Formal statistical testing of the concordance between spatial contours of risk and environmental exposure enables optimal use of spatial data.
We acknowledge the funding of a PhD studentship to J. Read by the Medical Research Council, UK.