Methods/Statistics
A measurement error analysis of prenatal phthalate effects on an infant outcome Jennifer Adibi* Jennifer Adibi Hai-Wei Liang Jiebaio Wang Nathaniel Snyder Donald DeFranco Janet Catov
Background: In epidemiology, confounding is rigorously addressed while measurement error is often overlooked. In the case of prenatal phthalate exposure, measurement error is likely when using a proxy (maternal urinary concentrations) instead of the gold standard (placental concentrations).
Objective: The aim was to evaluate measurement error in prenatal phthalate exposure when relying on maternal urinary concentrations, and to estimate the unbiased effects of prenatal phthalates on newborn anogenital distance (AGD).
Method: Sixty-eight pregnancies from the TIDES Study had data available on maternal urinary (3 trimesters) and placental tissue (3 tissue types villous (CF), chorion smooth (CS), basal plate (BP)) phthalate concentrations (at birth) from a full cohort (N=811). Intraclass correlation coefficients (ICC) were calculated to assess reproducibility. Linear associations between urinary and placental tissue concentrations were evaluated, with and without adjustment for confounding and heterogeneity by maternal characteristics. Measurement error correction in the effect of phthalates on infant AGD was evaluated and compared, using DAGs to identify sources of potential bias, and analytical approaches including regression calibration (RC), multiple imputation for measurement error (MIME) and machine learning.
Results: 12 phthalate metabolites were quantitated in over 70% of placental tissues and 100% of urine samples. Associations between urinary and placental phthalates were mostly negative or null. Reproducibility within maternal urine was low to moderate (ICC range 0.20-0.50, mean 0.27) and low within 3 placental tissue types (ICC range 0-0.11, mean 0.03). RC did not perform well due to lack of correlation between urinary and placental measures. The machine learning approach did not result in accurate prediction. Analysis is ongoing.
Conclusion: In the setting of weak correlation between the proxy and gold standard measures, standard approaches to measurement error correction do not perform well. In this scenario, we infer that urine and placental tissue phthalates are separate and distinct exposure constructs. Findings are presented, and generalizable recommendations are made on how to evaluate measurement error across tissue matrices as a crucial step in the application of biomarkers in epidemiology.