Environment International 2001-01-01

Changes in epidemiologic associations with different exposure metrics: A case study of phthalate exposure associations with body mass index and waist circumference

Krista Christensen, Jon Sobus, Martin Phillips, Todd Blessinger, Matthew Lorber, Yu-Mei Tan

Index: Environ. Int. 73 , 66-76, (2014)

Full Text: HTML

Abstract

The use of human biomonitoring data to characterize exposure to environmental contaminants in epidemiology studies has expanded greatly in recent years. Substantial variability in effect measures may arise when using different exposure metrics for a given contaminant, and it is often not clear which metric is the best surrogate for the ‘causal’ or ‘true’ exposure. Here we evaluated variability and potential bias in epidemiologic associations resulting from the use of different phthalate exposure metrics in the 2009–2010 National Health and Nutrition Examination Survey (NHANES). We examined associations between urinary phthalates metabolites and the outcomes of body mass index (BMI) and waist circumference (WC). We examined each of the following NHANES-derived exposure metrics for metabolites of individual phthalates: molar excretion rate (nmol/min), molar amount (nmol), molar concentration (nmol/mL, with and without additional model adjustment for creatinine), creatinine corrected molar concentration (nmol/g creatinine), and reconstructed daily phthalate intake (nmol/kg/day). In order to investigate potential biasing effect of each metric, we first assumed that daily intake of the parent phthalate is the causal exposure. We then constructed a simulated population based on the 2009–2010 NHANES, and randomly assigned each individual a di-2-ethylhexyl phthalate (DEHP) intake dose based on a published distribution, but independent of any other factor. Accordingly, all associations between these randomly assigned intake doses and individuals' BMI and WC should be null. Next, demographic data in NHANES were incorporated into a pharmacokinetic model to predict urinary molar excretions of five DEHP metabolites based on the randomly assigned DEHP intake. The predicted molar excretions were then used to calculate the same exposure metrics listed above. Three exposure metrics (randomly generated intake, excretion rate, urine concentration) showed no significant associations with BMI, which supports the null hypothesis stated above. In contrast, metrics adjusted for creatinine showed a significant negative correlation, and reconstructed daily intake showed a significant positive correlation, indicating the introduction of bias away from the true (i.e., null) association. Interestingly, trends in the simulation analysis were similar to those seen in the observed NHANES data. Our findings show that, at least in this example case, the choice of exposure metric can introduce significant bias of varying magnitude and direction into the calculation of epidemiologic associations.