Exploring and regionalizing geographic distributions of rare causes of mortality

Purpose: This presentation shows the results from experimentation of different smoothing techniques used in the detection of mapped clusters of mortality from rare causes. It also describes spatial autocorrelation analyses that can be used to evaluate clusters.

Data and Methods: The data are derived from the Compressed Mortality Files and the 2000 US Census for populations 15 years and greater. Geographic units are the county and Health Service Area (HSA) and the study period is for the years 1994 to 2003. The mortality variables studied include laryngeal cancer, silicosis, alcohol related liver disease, and cancer of the trachea, bronchus and lung. Occupational variables are also included. Two publicly available software programs, Head-Bang and GeoDA, produce the smoothed and borrowed strength values that are later mapped. Using global and local spatial autocorrelation analytical techniques from both GeoDA and ESRI's ArcGIS, spatial clusters are evaluated for potential as regions of statistically significant excess mortality.

Results: The Head-Bang method (non-parametric and can use population weights) produces smoothed maps from spatially heterogeneous and unstable county rates. The techniques in GeoDA also produce smoothed map distributions, but are based on spatial context and contiguity (i.e., spatial weighting). Preliminary results suggest that the clusters and regions constructed by these techniques are meaningful and can provide risk “surfaces” for less common causes of mortality.

Conclusion: Regionalizing can be useful in understanding geographical associations and distributions of mortality. However, process scales, spatial structure are also important considerations.