Traditionally, environmental epidemiologic studies have focused on assessing risks related to a single pollutant at a time. This, however, does not reflect reality, since we are constantly exposed to multiple pollutants at once. It is very important, therefore, to be able to assess exposure to pollutant mixtures when conducting environmental epidemiologic methods. Doing so, however, is especially challenging, mainly due to the high dimension of the multi-pollutant exposure matrix (if the exposure of interest includes more than e.g. 5 or 10 chemicals) and because these pollutants are usually very highly correlated with each other. Although some methods are available to address these issues, they usually require strong assumptions and have severe limitations. With this study we propose to bypass most of these limitations by adapting and extending a novel and robust method to assess exposure to multiple pollutants, called Principal Component Pursuit (PCP). We will assess the performance of PCP synthetic datasets representing multiple potential scenarios and study designs, and compare our results to those obtained by existing methods. Subsequently, we will apply PCP to three important Public Health issues, i.e. to evaluate the associations between (i) in utero exposure to a mixture of PCBs and neurodevelopment, (ii) exposure to a metals mixture and cardiovascular health, and (iii) exposure to an air pollution mixture and emergency cardiovascular admissions. Finally, we will develop and share software so other researchers can freely use this novel, robust and flexible tool across a plethora of study designs and research questions. Our proposed work will be significant as it will provide epidemiologists with a novel and robust tool to assess exposure to environmental pollutant mixtures.
We are constantly exposed to a mixture of environmental pollutants at once, but current epidemiologic methods either assess each pollutant separately, not capturing reality, or have severe limitations. With this study we propose to adapt a wildly popular method used in computer vision applications, called Principal Component Pursuit (PCP), and develop flexible extensions for many epidemiologic settings. We propose to assess the performance of this method, compare with existing methods and employ in real-life applications, as well as develop and share software so other research can also use this novel, robust and flexible tool.
|Wasserman, Gail A; Liu, Xinhua; Parvez, Faruque et al. (2018) A cross-sectional study of water arsenic exposure and intellectual function in adolescence in Araihazar, Bangladesh. Environ Int 118:304-313|