Using decision trees to understand the influence of individual- and neighborhood-level factors on urban diabetes and asthma.

Abstract

To determine the influence of individual and neighborhood factors that combined are associated with asthma and diabetes in a sample of urban Philadelphians using data mining, a novel technique in public health research. We obtained secondary data collected between May 2011 and November 2014 on individual's health and perception of neighborhood characteristics (N = 450) and Philadelphia LandCare Program data that provided relevant environmental data for the analysis (N = 676). RapidMiner open access data mining software was used to perform decision tree analyses. Individual- and neighborhood-level environmental factors were intricately related in the decision tree models, having varying influence on the outcomes of asthma and diabetes. The decision trees had high specificity (95-100) and classified factors that were associated with an absence of disease (diabetes/asthma). Improved neighborhood-level conditions related to social and physical disorder were consistently found to be associated with an absence of both asthma and diabetes in this urban population. This study illustrates the potential utility of applying data mining techniques for understanding complex public health issues.