Investigating connectivity of interventions for poor birth outcomes through network interpolation

Common spatial statistical methods that quantify patterns in individual outcomes often rely on point or kernel density measures, while patients needing and accessing resources depend almost entirely upon infrastructure that is not considered by the analysis.   We proposed and are developing network interpolation methods using ArcGIS and R software that examine the connectivity of address matched individuals with respect to community structure. As a pilot study for these methods in targeting local public health interventions, poor birth outcomes from vital statistics were examined at the county scale for Kalamazoo and Calhoun in Michigan.  Records of poor birth outcomes were initially examined as densities within areas of overall high births, via point density and kernel density algorithms, and as aggregations by census units, controlling for demographic factors.  When compared with census demographic data and low income housing locations, these results were used by community leaders who were assessing locations for maternal and child health interventions. Preliminary analyses found high rate clusters, some of which crossed census tract lines. In addition to those established geospatial analysis methods, we further examined community neighborhood connectivity and network linkages to standard health services via network interpolation algorithms. We found that because counties have very different infrastructure patterns standard density measures can produce misleading results when connectivity, and therefore access, is not consistent directionally across urban or urban-suburban-rural interface. Targeted intervention could benefit when the unique connectivity and neighborhood structure that is specific to each county is considered during analysis of individual level data.