Associations between long-term air pollution exposure and mortality and cardiovascular morbidity: A comparison of mobility-integrated and residential-only exposure assessment

Epidemiological studies investigating the health effects of long-term air pollution exposure typically only consider the participants’ residential addresses when determining exposure. Neglecting mobility may introduce measurement error, potentially leading to bias or reduced precision of exposure-health relationships in epidemiological studies. In this study we compared the exposure-health associations between residential-only and mobility-integrated air pollution exposures. We evaluated two major pollutants, NO2 and PM2.5, and four health outcomes, natural and cause-specific mortality and coronary and cerebrovascular events. Agent-based modeling (ABM) was used to simulate the mobility patterns of the participants in the EPIC-NL cohort in the Netherlands and the Swiss National Cohort (SNC) in Switzerland, based on travel survey information. To obtain mobility-integrated exposures, hourly air pollution surfaces were developed and overlaid with the time-dependent location data from the ABM. We used Cox proportional hazards models within each cohort separately to evaluate the association between residential-only and mobility-integrated exposure and mortality and cardiovascular events, adjusting for major individual and area-level covariates. The mobility-integrated exposure and the residential exposure showed very high correlations for both pollutants and cohorts (R2 > 0.97). The mean exposure was 1–2 % and the exposure contrast 10–20 % lower for the mobility-integrated exposure. For all health outcomes, both pollutants and both cohorts, there were only small differences between residential-only and mobility-integrated exposure effect estimates. For the SNC, Hazard ratios (HRs) for natural mortality were 1.04 (1.03 – 1.04) and 1.03 (1.03 – 1.04) per interquartile range (IQR) increase in NO2 for residential and mobility-integrated exposure, respectively. For PM2.5 the corresponding estimates were 1.01 (1.01 – 1.02) per IQR increase for both approaches. Our findings support the growing evidence that assessment of long-term air pollution exposure at the residential address only in epidemiological studies may not lead to substantial bias and loss of precision in health effects estimates.