Reproductive
Community Violence and Adverse Maternal and Infant Outcomes: Census Tract Homicides and Preterm Birth in California 2007-2018 John Halifax* John Halifax Caitlin Chan Shelley Jung Jennifer Ahern
The community environment has strong connections with adverse maternal and infant health outcomes in the United States. Community violence may be an important aspect of the environment, with potential to influence biological stress response mechanisms or maternal behavioral changes. Between community confounding leads us to examine within-community variation in violence in relation to birth outcomes. This work investigates the within community variation in homicides at the census tract level in three periods of gestation (preconception, trimester 1, and trimester 2) and preterm birth, with two post-birth exposure periods included as negative controls by temporality. Data are comprehensive for births in California in census tracts with at least one homicide from 2007-2018 (n=4.3 million observations among 6,714 census tracts).
Five average treatment effects contrasting exposure in each of the five periods vs entirely unexposed were estimated using data-adaptive longitudinal targeted maximum likelihood estimation (LTMLE). We aim to include census tract fixed effects to facilitate within-community comparisons, but large sample size has thus far produced analytic challenges. For computationally-feasible preliminary results, we used partition around medoids to cluster census tracts based on community-level covariates and an LTMLE estimator with fixed effects on these clusters for within-similar-community comparisons. Preliminary results on a data subset of 1,000 randomly selected census tracts find small associations that are not statistically significant, with exposure to community-level homicide showing some indication of an association with preterm birth during trimester 1. Future work will estimate these associations in the full dataset and use simulation studies to evaluate bias, variance, and computational feasibility of alternative LTMLEs that incorporate fixed effects or pool community-specific LTMLEs to be robust to between-community confounding.
