Keywords: diabetes, health disparities, propensity score analysis, geographic confounding, spatial data analysis
Motivated by exploring racial disparities in care and control between non-Hispanic black and non-Hispanic white veterans with type 2 diabetes, we develop spatial propensity score analysis (PSA) methods to reduce bias associated with geographic confounding, which occurs when measured or unmeasured confounding factors vary geographically, leading to imbalanced group comparisons. We augment PSA with spatial random effects, which are assigned conditionally autoregressive priors to improve inferences by borrowing information across neighboring geographic regions. In simulation, we show that ignoring spatial variation results in increased absolute bias and mean squared error, while spatial estimators perform well under various levels of spatial heterogeneity and sample sizes. In the motivating application, we construct a global hierarchical spatial estimate of risk difference and observe a gradual attenuation in effect compared to unadjusted and patient-level-adjusted estimates. Smoothed maps indicate areas of poor care and control across the southeastern United States, suggesting the need for community-specific interventions to target diabetes in geographic areas of greatest need.