2435

Background/Aim Effect estimates in air pollution epidemiology studies can vary depending on the exposure method used, due to error. We aimed to compare estimates for respiratory outcomes in a childhood cohort using three exposure methods. Methods We used: an intra-city land use regression (LUR) model; a national satellite-LUR (Sat-LUR) model; and a regional Bayesian blended model (BME); to estimate annual mean NO₂, a marker of traffic pollution, for 398 children (8 years) from the Childhood Asthma Prevention Study, Sydney. We collected questionnaire information on current asthma, wheeze, eczema, and rhinitis, and measured spirometry (FEV1, FVC, airway hyper-responsiveness (AHR)), exhaled nitric oxide (eNO), and atopy. We used logistic and linear regression to analyse binary and continuous variables respectively. Adjusted models included covariates chosen using directed acyclic graphs: sex; father’s education; gas cooking; and environmental tobacco smoke. Results The annual NO₂ means from LUR, Sat-LUR and BME models were 7.64 (SD 1.82), 8.77 (1.96) and 8.37 (1.81) ppb respectively. We generally found similar and non-significant (NS) effect estimates for most health outcomes across the three exposure methods after adjustment, with a few variations. We found significantly increased OR for any atopy (1.30 (1.02, 1.68)) and house dust mite atopy (1.35 (1.05, 1.74)) for LUR, and increased but NS effects for Sat-LUR and BME. Conversely, ORs were higher for current asthma for BME 1.31 (0.91, 1.69) than for Sat-LUR 1.24 (0.91, 1.68) or LUR 1.08 (0.82, 1.42) (all NS). Estimates for AHR were similarly increased with ORs varying from 24% (LUR) to 27% (BME), although NS. Conclusions Use of the three exposure measures resulted in similar effect estimates, albeit with subtle variations, illustrating the importance of exposure derivation. It is unclear which measure is most accurate and choice will depend on geography of models and cohorts and confidence in model inputs.