Statistical methods for the evaluation of effects of environmental chemical mixtures on adverse pregnancy outcomes

Gestational exposure to environmental chemicals presents an interesting challenge for epidemiological analysis of pregnancy outcomes and fetal development due to concurrent exposure of multiple chemicals. The analysis of chemical biomarkers presents several interesting challenges such as analysis of biomarkers below the limit of detection and handling repeated measures from the same chemical class. Additionally, the issue of co-linearity among the chemicals often results in unreliable effect estimates in traditional analysis approaches such as linear regression. Recent development in statistical methodology demonstrate promising opportunities for novel methods on mixture analysis. We explored the existing methods in the literature and propose to a novel method that combines Bayesian statistics and factor analysis for mixture analysis. We presented two different applications of this method in both regression and mediation analysis of birth weight. We demonstrated that both of the applications show strengths in precision of the estimate and interpretation of the results.