2359 - Pareto Optimality for Assessing Multimodal Transportation Accessibility: Balancing Equity and Efficiency When Sitting Interventions

Abstract

Background Travel time as a proxy for spatial accessibility in health care location-allocation studies can be particularly relevant when spatially concentrated population have a repeating pattern of infections, or a shorter return interval for services. This travel time based accessibility study focuses on a population identified as having repeat sexually transmitted infections. This population has been shown to be strongly associated with low socioeconomic status, which often corresponds to substantive reliance on public transit. Public health intervention studies barely consider transit when quantifying accessibility. Moreover, common frameworks used in these studies mostly focus on efficient allocation of services and usually disregard equity issues. Methods This study proposes a Pareto optimality approach to recommend locations that are multimodal accessible and allow equitable and efficient access to services. A simulated at-risk population, drawn from known core areas of repeat sexually transmitted infections in Kalamazoo County, Michigan, is examined. A raster-based drive time and a transit time model were developed using ArcGIS. Bi-objective optimization models were then developed to balance efficiency (by minimizing average travel time of the cohort) and equity (by minimizing variations of travel time from all simulated households). Results This raster-based cost optimization approach successfully identifies potential locations that are efficiently and equitably accessible using different transportation modes. A range of alternative solutions (Pareto frontiers) spatially and statistically enable decision makers to select from multiple locations and assess acceptable limits to travel for a given population. In our example, choosing a location, T2 over T1 means reducing the average travel time by 2.19 minutes by conceding only 11.94% increase in standard deviation of travel time. Additionally, T2 reduces the average travel time by 8.91 minutes and standard deviation by 3.7 minutes when comparing to existing facility location. In general, the spatial area bounded by the minimum average and minimum standard deviation location for a frontier line would represent the constraints to intervention location. The area then becomes a usable metric to measure transit versus private vehicle accessibility. Conclusions Improving spatial access to health facilities is important in reducing the prevalence of disease. Therefore, multimodal accessibility should be emphasized in future intervention placement research. The analysis of individual's travel time from distinct households to facility locations helps to address the inherent mismatch between current statistical methods that require detection of significant densities and the reality of individuals located on a street network or constrained by a particular transportation modality.