Spatial uncertainty is shown to bias epidemiological results through: (1) geocoding error, (2) boundary problems (e.g., mis-classification near boundaries, inappropriate boundaries), and (3) spatial autocorrelation in cohorts and exposures. These spatial errors are particularly problematic for clinical trials - especially where cohorts ar clustered (e.g., near recruitment centers), or spatially stratified in multi-center trials - becaus spatially-distributed social and environmental exposures may influence treatment response. We hypothesize that: (1) Spatial error arising from geocoding error, boundary effects, or spatial autocorrelation likely varies by city; (2) Spatially-distributed social and environmental exposures not randomized in clinical trials, may partially explain or modify treatment effects; (3) The influence of social and environmental factors on observed treatment effects may be influenced by spatial error - thus, treatment effect estimates may be refined by accounting for each. In collaboration with AsthmaNet, a national multi-site network of clinical trials for asthma, we will develop and analyze a multi-region database of GIS-based data on social and environmental exposures previously associated with asthma. We will quantify the impact of spatial errors on exposure estimates and clinical trial results - using this unique resource of clinical trials recruted and conducted using exactly the same protocols in nine very different urban areas (Albuquerque, Atlanta, Boston, Chicago, Denver, Milwaukee, Pittsburgh, San Francisco, Winston-Salem). Specifically, we will: (1) examine and ground-truth multiple geocoding techniques, with attention to differential performance by study site; (2) explore impacts of boundary error on social and environmental exposure assignment; (3) explore the effect of accounting for spatial autocorrelation and spatial error on exposure assignment and clinical effect estimates; and (4) derive best practices for spatially-informed examination of clinical tria results. To our knowledge, this will be the first study leveraging GIS-based information to inform upon the validation and interpretation of clinical trials. Spatial techniques developed and validated under this project will improve our understanding of spatial error and its influence on effect estimates for multiple exposures in asthma. It will also improve our ability to account for multiple exposures and spatial error in clinical trials, thus better tailoring clinical interventios.
Epidemiological results can be biased through spatial error, such as errors in geocoding, mis- classification near boundaries, or clustering in cohorts or exposures. These errors may be particularly problematic in clinical trials with clustered participations, or when merging data across multi-site trials. We propose the first investigation (to our knowledge) on the role of spatial error and spatially-distributed exposures (e.g., violence air pollution) on the observed efficacy of asthma Rx treatments in clinical trials implemented identically across nine U.S. cities.
