Quantifying the Predictive Value of Categories of Neighborhood-Level Risk Factors to Predict Health Outcomes

A person’s environmental context is well known to impact health outcomes.  However, this information is rarely available in an acute clinical care setting.  Although a growing body of literature combines the information available in the Electronic Medical Record (EMR) with environmental and place-based data to better examine the environmental impact on health, these studies generally focus on a single index or category of environmental data, thus failing to take into account the richness of geo-spatial data that are available, as well as the underlying interactions of multiple community risk factors.  Recent improvements in access to geo-spatial data at multiple layers make big data approaches to use of geo-spatial data for clinical care increasingly feasible.  However, the value of such approaches has not been demonstrated.  I’ll present a scoping review of machine learning approaches to combining place-based data with EMR data.  I’ll then propose a metadata framework for understanding the quality of geo-spatial data for health and compare this to a traditional index approach.  I’ll further propose a study design to examine if increased availability of different categories of geo-spatial data improves the accuracy of a hypertension risk model.