Assessing the impact of exposure to bisphenol A and triclosan during early pregnancy on birth weight outcomes in Canadian studies
MetadataShow full item record
It is of interest to know whether early pregnancy exposure to chemicals such as bisphenol-A (BPA) and triclosan (TCS) has negative impacts on birth outcomes. These chemicals are rapidly excreted from the body and therefore using a single exposure measurement for each individual, as a proxy for their average exposure, can induce large amounts of measurement error. It is well known that this will generally attenuate any exposure-outcome relationships that may exist. In this work, we consider data from two recent Canadian studies to assess the effect of BPA and TCS on low birth weight (LBW), small for gestational age (SGA) and other birth outcomes. The majority of participants in the studies have one chemical exposure measurement, while a small subset have 3 to 20 repeated chemical exposure measurements. We estimate the effect of BPA and TCS by using two regression calibration methods to handle measurement error. One method divides naive estimates obtained from an uncorrected linear or logistic exposure-outcome model by an estimated intra-class correlation coefficient (ICC). The other method calculates the best linear unbiased predictor of average exposure and uses it in place of the observed exposure in linear and logistic exposure-outcome models. Specific gravity, time of day, and time since last urine void are investigated for their ability to explain some of the variation in observed exposure within a study participant. The results suggest decreased odds of LBW (uncorrected odds ratio = 0.85, 95% confidence interval: 0.75−0.98; corrected odds ratio = 0.81, 95% confidence interval: 0.68 − 0.97) and SGA (uncorrected odds ratio = 0.87, 95% confidence interval: 0.76 − 0.98; corrected odds ratio = 0.83, 95% confidence interval: 0.70 − 0.99) birth outcomes with increased logTCS exposure, but the relationship is not highly significant. Measurement error in BPA was so severe that regression calibration results were insignificant with wide confidence intervals. We also conducted a simulation study to investigate the performance of the estimation methods. The results suggest a few options for future studies to have more power to detect the effect of BPA and TCS on birth outcomes, such as more exposure measurements for each individual, or identifying covariates that explain more within-individual variation in exposure.