Skip to main content
eScholarship
Open Access Publications from the University of California

Using Birth Cohort Data to Estimate Prenatal Chemical Exposures for All Births around the New Bedford Harbor Superfund Site in Massachusetts.

  • Author(s): Khalili, Roxana
  • Bartell, Scott M
  • Levy, Jonathan I
  • Fabian, M Patricia
  • Korrick, Susan
  • Vieira, Verónica M
  • et al.

Published Web Location

https://doi.org/10.1289/EHP4849Creative Commons 'BY' version 4.0 license
Abstract

BACKGROUND:Children born near New Bedford, Massachusetts, have been prenatally exposed to multiple environmental chemicals, in part due to an older housing stock, maternal diet, and proximity to the New Bedford Harbor (NBH) Superfund site. Chemical exposure measures are not available for all births, limiting epidemiologic investigations and potential interventions. OBJECTIVE:We linked biomonitoring data from the New Bedford Cohort (NBC) and birth record data to predict prenatal exposures for all contemporaneous area births. METHODS:We used prenatal exposure biomarker data from the NBC, a population-based cohort of 788 mother-infant pairs born during 1993–1998 to mothers living near the NBH, linked to their corresponding Massachusetts birth record data, to build predictive models for cord serum polychlorinated biphenyls (expressed as a sum, [Formula: see text]), [Formula: see text] (DDE), hexachlorobenzene (HCB), cord blood lead (Pb), and maternal hair mercury (Hg). We applied the best fit models (highest pseudo [Formula: see text]), with multivariable smooths of continuous variables, to predict exposure biomarkers for all 10,270 births during 1993–1998 around the NBH. We used 10-fold cross validation to validate the exposure models and the bootstrap method to characterize sampling variability in the exposure predictions. RESULTS:The 10-fold cross-validated [Formula: see text] for the [Formula: see text], DDE, HCB, Pb, and Hg exposure models were 0.54, 0.40, 0.34, 0.46, and 0.40, respectively. For each exposure model, multivariable smooths of continuous variables improved the fit compared with linear models. Other variables with significant effects on exposure estimates were paternal education, maternal race/ethnicity, and maternal ancestry. The resulting exposure predictions for all births had variability consistent with the NBC measured exposures. CONCLUSIONS:Predictive models using multivariable smoothing explained reasonable amounts of variance in prenatal exposure biomarkers. Our analyses suggest that prenatal chemical exposures can be predicted for all contemporaneous births in the same geographic area by modeling available biomarker data for a subset of that population. https://doi.org/10.1289/EHP4849.

Many UC-authored scholarly publications are freely available on this site because of the UC's open access policies. Let us know how this access is important for you.

Main Content
Current View