Selecting Spatial Scale of Covariates in Regression Models of Environmental Exposures

• Activity Number: 73
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Lauren P. Grant* and Chris Gennings and David Wheeler
• Companies: Virginia Commonwealth University and Icahn School of Medicine at Mount Sinai and Virginia Commonwealth University
• Keywords: model selection ; spatial scale ; nitrate ; environment ; cancer risk factors

Abstract

Environmental factors or socioeconomic variables used in regression models to explain health outcomes or environmental chemical exposures are often in practice modeled at the same buffer distance or spatial scale. In this paper, we present four model selection algorithms that select the best spatial scale for each buffer-based or area-level covariate. Contamination of drinking water by nitrate is a growing problem in agricultural areas of the United States, as ingested nitrate can lead to the endogenous formation of N-nitroso compounds, which are potent carcinogens. We applied our methods to model nitrate levels in private wells in Iowa. We found that environmental variables were selected at different spatial scales and that a model allowing spatial scale to vary across covariates provided the best goodness-of-fit. Our methods can be applied to investigate the association between environmental risk factors available at multiple spatial scales or buffer distances and measures of disease, including cancers.