Application of hybrid modelling techniques in the assignment of estimates of personal exposure to ambient air pollution: the MELONS study

Keywords: Air pollution, Personal exposure Background/Aim Hybrid modelling of ambient air pollution has the potential to improve the accuracy of personal exposure estimates assigned to individuals in cohort studies. Traditional modelling techniques neglect the movement of individuals and the gradients of exposures observed when indoors and in transport microenvironments. The present study aimed to assess the changes in assigned personal exposures after adjustment by estimated time-activity patterns derived from a representative sample. Methods Travel survey data for ~19,000 respondents aged 50 and older, in addition to hourly residential CMAQ-urban dispersion model estimates of NO₂, PM₁₀ and PM₂.₅, were applied to each individual. The difference between these estimates and those adjusted for estimated location, transport mode and microenvironment were then calculated before aggregation by age and spatially across London (LHEM concentration factors). Ninety members of the COPE study were included in the present analysis and annual average concentration estimates from the nearest monitor were assigned before adjustment via the aforementioned LHEM concentration factors. Results Annual average concentrations (μg/m³) from the nearest monitor were assigned to each COPE individual for NO₂ (mean: 43.8; SD: 17.3), PM₁₀ (21.4; 4.7) and PM₂.₅ (12.7; 2.37). After adjustment for estimated time-activity by LHEM concentration factors, assigned personal estimates reduced to 15.5 (6.14), 6.42 (1.47) and 7.61 (1.34), respectively. Before- and after- adjustment estimates were highly correlated for NO₂ (r: 0.95) and PM₂.₅ (0.96); suggesting that the ranking of estimates showed little change after adjustment. However, for PM₁₀ (0.82) there was some change in the ranking of exposures after adjustment. The determinants of this change are explored. Conclusions Leveraging previously collected representative data through methods similar to those presented here, epidemiological studies may have the ability to incorporate complex population-level time-activity information into personal exposure estimates; potentially enhancing assignment using fixed-site monitors or traditional modelling techniques.