Measurement Errors in Environmental Epidemiology

摘要

Abstract Exposure measurement error is a likely source of bias in nearly all environmental and epidemiological studies, typically leading to under-estimation of relative risks and loss of statistical power to detect effects. In the previous cycle of this project, we extended the regression calibration method for adjustment for measurement error in multivariate regression models, including Cox models and logistic models, to accommodate the study designs and data structures encountered in environmental epidemiology. We featured these new methods in a number of publications on the health effects of environmental exposure to endotoxin, methyl tert-butyl ether, lead, and indoor NO2, and developed publicly available software. In the next cycle of this project, we will focus on issues in air pollution epidemiology - in particular, the chronic effects of particulate exposure and elemental carbon on all-cause mortality, cardio-vascular mortality and lung cancer mortality. Having assembled an inter-disciplinary team of leading statisticians, environmental scientists and environmental epidemiologists from around the world, we will develop methods to adjust for measurement error in Cox regression models suitable for the prospective cohort designs of the Six Cities Study, Nurses' Health Study, Netherlands Cohort Study, and MESA-Air in relation to cumulative and 12 month running average exposure metrics. Careful attention will be paid to the important issues surrounding timing of exposure and potential non-linearity of the dose-response curves of the exposure metrics, in particular, removing bias in the quantification of these features due to exposure measurement error. The biomarker data available in NHS and MESA-AIR will be used to improve the exposure validation. User- friendly software will be posted on the web, facilitating widescale application of the new methods.