Influence of spatial heterogeneity in socioeconomic deprivation on birth outcomes and air pollution epidemiology in New York City, USA

摘要

Background: Numerous studies have linked area-level socioeconomic deprivation with adverse birth outcomes. Deprivation is often operationalized as a composite index, which efficiently combines diverse components of socioeconomic position (SEP), but also may obscure spatial heterogeneity in variable distributions, potentially biasing epidemiological models. Aims: We aimed to build (and test) a deprivation index reflecting spatial heterogeneity in New York City (NYC), to include with fine-scale air pollution in birth outcomes models. Methods: We explored intra-urban spatial variability across contextual SEP factors previously associated with birth outcomes (n=20). We applied spatially-stratified Principle Components Analysis (borough-level) to reflect spatially-varying prevalence and combinations of SEP factors in a composite deprivation score. For comparison, we calculated a crude citywide index following the same data reduction process. Results: The spatial deprivation index retained seven area-based SEP variables (college degree, unemployment, management/professional occupation, residential crowding, poverty, households receiving public assistance, non-white racial composition), explaining 55% of overall variance in SEP factors. The crude index included fewer and somewhat different SEP variables, explaining 41% of overall variance, and correlated with the spatial index at r=0.70. Residential outdoor PM2.5 exposure was differentially correlated with the spatial (r=-0.11) and crude (r=0.02) indices. After adjusting for individual covariates and exposure to PM2.5, spatial and crude deprivation indices were associated with 5.32g and 6.56g decrements in birth weight per IQR change in SEP index - small changes, yet within the range of concern for air pollution effects on birth outcomes. Conclusions: Spatially-informed data reduction techniques can improve estimation of area-level deprivation, and elucidate potential confounding and bias in epidemiological analyses.