Introduction: The association between modifiable built environment factors and diabetes prevalence may vary across space. Global regression models may mask underlying relationships occurring at more granular geographic levels. This census tract-level analysis explores such associations, and their potential spatial nonstationarity, in the state of Georgia.

Methods: Spatial autocorrelation in 2022 tract-level diabetes prevalence was tested using Global Moran’s I. Using geographically weighted regression (GWR), the association between diabetes prevalence and walkability, park land coverage, retail food environment and average distance to the nearest clinic was estimated, while controlling for several demographic and socioeconomic characteristics. GWR allows for coefficients to vary locally so we can assess whether the significance and magnitude of these associations changed over space.

Results: There is strong evidence of clustering of values in diabetes prevalence in Georgia (Moran’s I=0.61, p<0.001). In both the adjusted global and geographically weighted model, walkability was the only statistically significant predictor of interest associated with adult diabetes prevalence (mean [SE] OLS: β = −0.74[S.E. = 0.05], p<0.01; mean [SE] GWR β, -0.57 [SD= 0.23]). Results from the GWR suggest that, on average, a one SD increase (3.4) in national walkability index score is associated with an average 0.57 SD (2.22%) decrease in diabetes prevalence, and that this association is locally significant in 58% of census tracts across the state.

Conclusion: There is evidence that the distribution of diabetes prevalence in Georgia is clustered, and that the association between walkability and diabetes prevalence varies in magnitude and significance across the state. This exploratory study lends evidence to look further into improved neighborhood walkability as a policy intervention to improve the health of Georgians.