Statistical Models to Assess Associations between the Built Environment and Health: Examining Food Environment Contributions to the Childhood Obesity Epidemic.

Models are developed and applied to examine the associations between built environment features and health. These developments are motivated by studies examining the contribution of features of the built food environment near schools, such as availability of fast food restaurants and convenience stores, to children’s body weight. The data used in this dissertation come from a surveillance database that captures body weight and other characteristics for all children in 5th, 7th, and 9th grades enrolled in public schools in California during 2001-2010 and a commercial data source that contains the locations of all food establishments in California for the same time period. First, we develop a hierarchical multiple informants model (HMIM) for clustered data that estimates the marginal association of multiple built environment features and formally tests if the strength of their association differs with the outcome. Using this new model, we establish that the contribution of the availability of convenience stores to children’s body mass index z-scores (BMIz) is stronger than that of fast food restaurants. Second, we propose to use a distributed lag model (DLM) to examine whether and how the association between the number of convenience stores and children’s BMIz decays with longer distance from schools. In this model, distributed lag (DL) covariates are the number of convenience stores within several contiguous “ring”-shaped areas from schools rather than circular buffers, and their coefficients are modeled as a function of distance, using smoothing splines. We find that associations are stronger with closer proximity to schools and vanish by about 2 miles from school locations. Third, we develop a hierarchical distributed lag model (HDLM) to systematically examine the variability of the built environment association across regions to help address a yet unanswered question in the built environment literature: whether and how activity spaces relevant to health vary across regions. We find DL coefficients vary across regions, implying that variation in activity spaces also exists. We also identify areas where children’s BMIz is more vulnerable to built environment factors. This dissertation provides novel methods with which to study how built environment factors affect health.PhDBiostatisticsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/110362/1/jongguri_1.pd