[Detecting spatial autocorrelation of cancer risk when population density is heterogeneous]

Several studies have shown that cancer incidence and cancer mortality are spatially autocorrelated. Implicit to the demonstration of this characteristic is the assumption of similar variability of risk estimates in all geographic units. As this assumption is often incorrect in the context of most geographical studies of cancer incidence and mortality, we propose a simulation method which takes into account the heterogeneity of population density to study the distributions of the Moran I and Smans D statistics under the correct hypothesis. The results are compared with the classical approach using incidence data from the "Département" of Isère.