Statistical approaches to evaluating the association between mother's exposure to perfluoroalkyl substances (PFASs) and birth outcome in North-Rhine Westphalia

摘要

Background: Developmental toxicity of perfluoroalkyl substances (PFASs) has been proven in animal studies. This raises concern over the potential effects of PFAS on human reproductive health. However, results of epidemiologic studies of exposed populations are inconsistent.As reported earlier, more than 4 million residents in several areas of North-Rhine Westphalia (NRW), Germany, have been exposed to PFASs in drinking water. PFASs' concentrations in drinking water were used as a surrogate marker of exposure. Aim: Two statistical approaches (linear regression, spatial Bayesian method) are compared to evaluate the association between mother's exposure to PFASs via drinking water and birth outcome in NRW, Germany. Methods PFASs' drinking water concentrations were derived from an extensive monitoring programme of the North Rhine-Westphalia State Environment Agency. Birth outcome was assessed on the basis of the quality control datasets of the respective Medical Association. For a spatially correlated effect, Markov random field (MRF) priors were chosen, which reflect spatial neighborhood relationships. To estimate the posterior inference the Integrated Nested Laplace Approximation (INLA) was used. Results Drinking water analyses from 33892 measurements provided PFOA-concentrations for 260 water supply areas. PFOA-concentrations ranged from below LOQ to 650 ng/l (37 % ＞ LOQ). Birth outcome, maternal smoking status, medication and other confounders were derived on postcode level for 1,176,488 newborns and their mothers between 2003 and 2010.Matching of birth outcome and PFASs' exposure data was successfully achieved by geoprocessing methods in 382 postcode areas, resulting in a total of 281,019 datasets. Strength of effects and potential confounders will be compared. Conclusion Spatial models might account better for the surrogate drinking water contamination, although MRFs only consider confounders on an aggregate basis. Data will be followed on a longitudinal basis.