Investigating the effects of multiple exposure measures to traffic-related air pollution on the risk of breast and prostate cancer
Résumé
Background: Traffic-related air pollution is associated with a range of health effects. Exposure to Nitrogen dioxide (NO2), a known marker of traffic-related air pollution, has been traditionally estimated using surfaces generated through land-use regression (LUR) models. Recently, air pollution dispersion models, with refined spatial resolution, have been used to derive NO2 exposures in urban areas. With either LUR development or dispersion modelling, there is evidence that data collection protocols and modelling assumptions can have a large effect on the resulting NO2 spatial surface. This study investigates the effects of various NO2 exposure surfaces on the resulting risk estimates of breast cancer (BC) and prostate cancer (PC), both of which have already been associated with traffic-related air pollution.
Methods: We derived exposures for individuals in two case control studies (BC and PC) in Montreal, Canada using four different surfaces for NO2. Two of the surfaces were developed using LUR but employed different data collection protocols (LUR-1 and LUR-2), and the other two surfaces were generated using dispersion modelling; one with a regional model (dispersion-1) and another with a street canyon model (dispersion-2). Also, we estimated separate Odds Ratios (ORs) using concentrations of NO2 as measures of exposure, both for the BC and PC case–control studies.
Results: While the range of NO2 concentrations in dispersion surfaces (4-26 ppb) is lower than the range in LUR surfaces (4-36 ppb), the four surfaces were found to be reasonably correlated, with Pearson correlation coefficients ranging between 0.52 and 0.79. The ORs for BC were estimated to be 1.26 (CI: 0.97, 1.63), 1.10 (CI: 0.89, 1.37), 1.07 (CI: 0.94, 1.23), and 1.05 (CI: 0.90, 1.22) based on LUR-1, LUR-2, dispersion-1, and dispersion-2. In contrast, the ORs for PC were estimated to be 1.39 (CI: 1.14, 1.69), 1.30 (CI: 1.08, 1.56), 1.13 (CI: 1.03, 1.24), and 1.04 (CI: 0.94, 1.16) based on LUR-1, LUR-2, dispersion-1, and dispersion-2.
Conclusions: Our analysis sheds light on the use of four different measures of exposure to traffic-related air pollution. We observed higher mean ORs based on the LUR surfaces but with overlapping CIs. Since LUR models capture all sources of NO2 and dispersion models only capture traffic emissions, it is possible that this difference is due to the fact that non-road sources also contribute to the spatial distribution in NO2 concentrations.