Predicting regional variations in mortality from motor vehicle crashes

Objective: To show that the previously-observed inverse relationship betwee n population density and per-capita mortality from motor vehicle crashes ca n be derived from a simple mathematical model that can be used for predicti on. Methods: The authors proposed models in which the number of fatal crash es in an area was directly proportional to the population and also to some power of the mean distance between hospitals. Alternatively, these can be p arameterized as Weibull survival models. Using county and state data from t he U.S. Census, the authors fitted linear regression equations on a logarit hmic scale to test the validity of these models. Results: The southern stat es conformed to a different model from the other states. If an indicator va riable was used to distinguish these groups, the resulting model accounted for 74% of the variation from state to state (Alaska excepted). After contr olling for mean interhospital distance, the southern states had a per-capit a mortality 1.37 times that of the other states. Conclusions: Simply knowin g the mean distance between hospitals in a region allows a fairly accurate estimate of its per-capita mortality from vehicle crashes. After controllin g for this factor, vehicle crash mortality per capita is higher in the sout hern states, for reasons yet to be explained.