Can network analysis capture area-level expression of disease? An Exploratory Analysis

For over thirty years small area studies examined regional variations in socioeconomic characteristics and health outcomes. More recently, health researchers have incorporated social network into behavioral models of infectious disease, health risks associated with obesity, and the influence of peers on adolescent smoking (Luke & Harris, 2007). However, network analysis has far greater potential to enhance small area studies than has been demonstrated in the literature. Network modeling focuses on the relationship between observational units, whereas, thus far, the research has generally constrained observational units to patient level detail. By shifting the observational unit of analysis to the diagnostic code level network analysis may provide insights into underlying sources of co-morbidity such as environmental and social factors. The objective of this presentation is to inform researchers how to define a graph theoretic data structure, describe the network properties, evaluate how the network characteristics vary across space and time, then compare and contrast the corresponding health outcome across counties. The goal of this analysis is to incorporate a network based approach to data analysis with more traditional epidemiological spatial analysis and GIS mapping tools to study relationships between ICD 9 diagnostic codes across counties in North Carolina using Healthcare Cost and Utilization Project (HCUP) data from 2000 and 2009. Our findings demonstrate how co-morbidity relationships vary across space and time, revealing patterns of human disease not readily apparent from studying individual disorders, offering a visualization and statistical metrics to help inform more cost effective public health interventions.