Exploratory spatial analysis of birth defect rates in an urban population.

Are birth defect rates unusually high in particular urban localities? The answer requires that the birth defect rate for which 'significance' is claimed be adjusted for the variable population sizes of each area for which the rate is computed and for the spatial dependence of rates based on shared observations between neighboring areas. By address-matching birth and birth defect records to a digital road map, we are able to compute local birth defect rates at regular grid locations by dividing the number of birth defects that occurred in the geographical vicinity of a grid location by the total number of births in the same vicinity. We test for significance, at regular spatial intervals, against the null hypothesis that the observed rate at any locality could reasonably have arisen by chance alone, given the underlying geographical variation in births. Significance is determined by using Monte Carlo simulations, where each birth location has an identical probability of being a defect. From 1000 simulations, a statistical distribution of the birth defect rate for each grid location is determined. The proportion of the simulated birth defect rates that are less than the observed rate at any grid location is the probability that the observed rate is significant. These probabilities, portrayed as isarithmic maps, show areas that have significantly high birth defect rates. Our results show birth defect rates for the period 1983 to 1990 in Des Moines, Iowa, U.S.A.