Relying on repeated biospecimens to reduce the effects of exposure measurement error in studies linking the exposome to health

Abstract

OPS 19: Chemicals: exposure assessment challenges, Room 110, Floor 1, August 28, 2019, 1:30 PM - 3:00 PM Background/Aim: The exposome concept encompasses all environmental exposures a human is exposed to from the prenatal period onwards, and calls for assessing numerous exposures, typically using biomarkers. The amount of measurement error may differ between exposures, possibly influencing the performance of variable selection models in exposome-health studies. We evaluated this impact, and the efficiency in this context of two measurement error correction methods relying on the collection of repeated biospecimens. Methods: In a simulation study, we generated 237 exposures with different amounts of measurement error, assuming that 1 to 10 exposures linearly influenced a continuous health outcome and that up to 10 biospecimens were available per subject. We applied within-subject biospecimens pooling or regression calibration (RC), both followed by the deletion/substitution/addition algorithm to estimate the exposure-health associations, and compared their performances. Results: In the absence of measurement error, the average sensitivity to identify exposures influencing health was 75% and false detection proportion (FDP) was 26%. Measurement error decreased sensitivity to 46%, increased FDP to 49% and caused 52% attenuation bias in dose-response functions. When repeated biospecimens were available, within-subject pooling and RC improved sensitivity (average, 63%), FDP (average, 37%) and attenuation bias (average, 48%), and performance increased with the number of available biospecimens. Conclusions: Exposome studies relying on spot exposure biospecimens suffer from decreased sensitivity and increased bias and FDP, with greater amplitude for the exposures with the largest amount of measurement error. Study performances can be improved by collecting repeated biospecimens per subject.