|
Abstract:
|
Health data from large cohorts are frequently aggregated by administrative areas such as counties and zip codes. However, data on air pollutant concentrations are typically available at point locations from monitors, which results in spatial misalignment between the health and exposure data. For large air pollution epidemiology cohorts, area-level exposures are frequently computed by directly assigning monitor observations to the containing area. An alternative approach is to fit a point-level exposure prediction model and average predictions from that model together. Even when the prediction model is mis-specified and not able to fully represent the underlying pollution concentration surface, the prediction approach may result in less measurement error in the area-wide average than using monitor values directly. We compare these approaches using data on particulate matter and childhood asthma prevalence from the United States Medicaid cohort.
|
