Methods/Statistics
Revisiting the modifiable areal unit problem in the era of exposome-wide association studies: Assessing the performance of the CDC/ATSDR Social Vulnerability Index at privacy-protecting spatial scales Jonathan V. Lewis* Jonathan Lewis Nathaniel MacNell Gary J. Larson Anna J. Jones Ian D. Buller Farida S. Akhtari Kyle P. Messier Alison A. Motsinger-Reif
Background: Exposome-wide association studies (ExWAS) may be sensitive to environmental exposure assessments across geographic and temporal scales. Linking de-identified geomarkers to participant locations facilitates data sharing and improves privacy protection. However, conducting analyses at different spatial scales and boundaries may yield varying estimates of epidemiologic associations.
Methods: We replicated the CDC/ATSDR Social Vulnerability Index (SVI) and its n=16 components and n=4 themes using 2015-2019 U.S. Census Bureau American Community Survey data across North Carolina census tracts (TR; n=2,195) and four aggregations: county subdivision (CS; 1,041), 5-digit Zip Code Tabulation Area (Z5; 808), county (CT; 100), and 3-digit Zip Code Tabulation Area (Z3; 20). We descriptively compared individual components, themes, and overall SVI at chosen scales and fit 67,392 logistic regression models of theoretical associations using a simulated epidemiologic cohort of one million locations.
Results: Fifteen individual components (94%; p-value<0.05) showed statistically significant differences across scales by a Kruskal-Wallis test. Moran’s I (one lag) for SVI became increasingly variable and showed less evidence of spatial clustering at larger scales (ITR=0.53; ICS=0.30; IZ5=0.25; ICT=0.44; IZ3=-0.09). In models fit to simulated data, odds ratios for SVI at larger scales attenuated towards null with decreased empirical coverage as the simulated population effect size increased.
Discussion: SVI became less informative when aggregated at larger administrative units. The epidemiologic efficacy of a simulated ExWAS study to detect associations was reduced when using aggregated geomarkers.
Conclusion: Our results emphasize the importance of using data at spatial resolutions that align with hypothesized exposure-phenotype mechanisms and anticipating the potential loss of epidemiologic efficacy when participant data are limited to larger spatial scales to protect privacy.
