Adjusting for unmeasured spatial confounding with distance adjusted propensity score matching

摘要

Propensity score matching is a common tool for adjusting for observedconfounding in observational studies, but is known to have limitations in thepresence of unmeasured confounding. In many settings, researchers areconfronted with spatially-indexed data where the relative locations of theobservational units may serve as a useful proxy for unmeasured confounding thatvaries according to a spatial pattern. We develop a new method, termed DistanceAdjusted Propensity Score Matching (DAPSm) that incorporates information onunits' spatial proximity into a propensity score matching procedure. We showthat DAPSm can adjust for both observed and unobserved confounding and evaluateits performance relative to several other reasonable alternatives forincorporating spatial information into propensity score adjustment. The methodis motivated by and applied to a comparative effectiveness investigation ofpower plant emission reduction technologies designed to reduce populationexposure to ambient ozone pollution. Ultimately, DAPSm provides a framework foraugmenting a "standard" propensity score analysis with information on spatialproximity and provides a transparent and principled way to assess the relativetrade offs of prioritizing observed confounding adjustment versus spatialproximity adjustment.