Associations between birthweight and metals: A real world example of bias amplification in a North Carolina birth cohort
On this page:
Background and aim: Air pollution comprises a mixture of gases and small particles, including metals, and has been associated with changes in mean birthweight (BW). Often, studies estimate exposure for a single metal in isolation or multiple metals simultaneously, but rarely evaluate both or examine the potential bias amplification. Methods: We evaluated census tract level concentrations of 8 metals (arsenic, cadmium, chromium, cobalt, lead, manganese, mercury, and nickel) from the 2011 National Air Toxics Assessment linked with 431,929 infant-mother pairs with births between 2012-2015. Correlations were low to moderate (r<0.62) between the metals. The change in BW (grams, g) and 95% confidence interval (CI) per one log-transformed unit of change in concentration was estimated with generalized estimation equation models, adjusted for race/ethnicity, age, marital status, education, and Medicaid status. Exposure to each metal was considered in isolation (IPM) and part of a multiple pollutant model (MPM) of 8 metals (mutually adjusted). Results: We observed the largest decrease in BW for manganese [-8.0g(95% CI:-10.0,-6.1)] in the IPM and for arsenic [-11.5g(-19.0,-4.0)] in the MPM. The greatest increase in BW was observed with cadmium for both models [isolated: 16.5g(11.5,21.56); multiple: 13.7g(6.2,21.2)]. The direction of association changed with arsenic, chromium, and lead between the IPM and MPM. All metals, except manganese, were associated with more than a 10% change in BW per one log-transformed unit change in metal concentration when comparing the results from the IPM to the MPM, suggesting possible bias amplification. Conclusions: We observed changes in directionality and substantial changes in the magnitude of the associations between IPM and MPM. Evaluating MPM may over or under emphasize associations due to residual confounding or collinearity, leading to bias amplification. Considering the potential for bias amplification and its magnitude is useful in evaluating MPM.