Improving Noise Exposure Assessment for Epidemiological Studies

Environmental noise has long been recognized as adversely affecting sleep and concentration, but evidence is emerging that it may affect other disease processes. The overarching goal of this research is to improve the assessment of noise exposures in multi-pollutant environmental epidemiology studies. To explore the state of epidemiological research about traffic-related pollutants, we used bibliometric analysis to characterize the peer-reviewed literature about the association of NO2, PM2.5 and noise exposures with cardiometabolic disorders. We retrieved references published 1994-2017 from Scopus and classified references with respect to exposure, health outcome and study design using index keywords. Temporal trend, top cited references, index keywords used and the number of hypothesis testing and non-hypothesis testing studies were identified for each group of exposures. We found that the study of simultaneous exposures to multiple pollutants (NO2 or PM2.5 and noise) is a current trend, and likely to continue. And we believe the transition to the use of hypothesis testing study designs to explore associations between noise and cardiometabolic outcomes may be supported by improved understanding of the mechanism of action, and/or improvements to the accuracy and precision of air pollution and noise exposure assessments. In contrast to traffic-related air pollutants, exposure assessment models for environmental noise are relatively scarce. We developed a land use regression (LUR) to predict noise levels in Chicago, Illinois, United States. Noise LUR models have been developed in a variety of cites around the world recently. An advantage of the LUR model relative to acoustic models is that it can reflect the complexity of environmental noise: The multivariable regression structure allows us to select the independent variables hypothesized to be related to noise, not just known to be sources of noise. A limitation of LUR models is that it may not accurately predict noise levels for conditions not represented by the observed data, an error termed prediction outside the range. We found that an existing monitoring network in Chicago, the Array of Things (AoT), was not sufficient to build a noise LUR model for Chicago because 27.4% of city area would require prediction outside of range of the AoT monitoring sites. This motivated a field noise sampling campaign, involving 5-minute daytime noise levels (Leq, 5-min) measured at 75 sites in March 2019. These sites were selected using a stratified random sampling approach. A simulation study demonstrated that random sampling approaches with 40-80 samples were sufficient to represent characteristics of Chicago, IL. To develop the noise LUR model, a series of geographically-based variables describing land use, aviation traffic, ground traffic, and natural environment, at sampling location were collected for use in supervised stepwise regression model building. The noise LUR model was built to estimate the urban daytime noise level, and the final model had the following independent variables: distance to the nearest CTA train track, total lengths of primary road within 100 meter, and mean NDVI within 100 and 500-100 meters. The adjusted R2 and RMSE for the noise LUR model with training and validation data sets were 0.60 and 4.67 dBA and 0.51 and 5.90 dBA, respectively. We found the mean noise level in Chicago (Leq, 5-min) was 60.6 dBA, and nearly 75% of city area is predicted to have daytime noise levels higher than 55 dBA, a WHO guideline value for daytime environmental noise to prevent annoyance. The noise LUR model we built is suitable for predicting daytime noise levels in the majority of the city area (85.3% of the area), and can be used as a noise exposure assessment tool for epidemiology studies. In conclusion, we developed an exposure assessment method for noise in Chicago – a LUR model – that can be used to support innovative multi-pollutant environmental epidemiology studies of cardiometabolic and other health outcomes. More generally, we also demonstrated the effectiveness of a relatively simple method for the selection of noise sampling sites to support development of LUR models. Future work will utilize this model in ongoing epidemiological studies in the Chicago area.