CFE 2019: Start Registration
View Submission - CMStatistics
B1121
Title: Bayesian analysis of correlated exposure biomarkers subject to a limit of detection Authors:  Lawrence McCandless - Simon Fraser University (Canada) [presenting]
Abstract: Biomarkers are widely used in perinatal epidemiology to examine the health effects of environmental chemical exposures during pregnancy. However a difficulty with biomarkers is that the chemical concentrations at low doses are often left-censored by a limit of detection. Further complicating the analysis, chemical biomarkers are often highly correlated because the exposure occur in mixtures and they have the same underlying exposure source (eg. polychlorinated biphenyl mixtures). The pattern of missing data and correlated predictor variables complicates efforts to measure the causal effects of individual and combined exposure to multiple chemical agents on reproductive and child health outcomes. We describe a novel Bayesian approach to examine the role of correlation between chemical biomarkers to enhance imputation of missing data, and we use this information to improve estimation of the health effects of low level chemical exposures.