AMBIENT AIR POLLUTION MODELS: IMPACT ON ESTIMATED EXPOSURE AND RELATIONSHIP WITH FETAL GROWTH, THE EDEN MOTHER CHILD COHORT

Abstract

Background and Aims: We previously reported that, although moderately correlated, nitrogen dioxide (NO2) atmospheric levels estimated by air quality monitoring stations (AQMS) and a geostatistical model, showed similar associations with birth weight (BW). We extended this comparison by also considering a dispersion model, and PM10 levels. Methods: We focused on women from the Eden cohort recruited in two French maternity hospitals, living less than 5km away from a permanent monitoring station in two areas of about 500km2 in which we implemented dispersion and geostatistical models. NO2 and PM10 pregnancy average levels were available for 770 and 452 women, respectively. We estimated Kappa coefficients of concordance (K) to compare the exposure estimates (categorized in tertiles), and used adjusted linear regression to estimate the association between exposure levels and BW. Results: Mean ±sd exposure levels (μg/m3) during pregnancy were 29±10 for NO2 and 19±2 for PM10, as estimated by the AQMS model, 24±8, and 18±3 respectively for the dispersion model, and 24±5 for NO2 as estimated by the geostatistical model. The dispersion model showed a stronger agreement with the geostatistical (KNO2=0.52) than with the AQMS model (KNO2=0.37 KPM10=0.41). Agreement between dispersion and AQMS models improved within 2 and 1km of an AQMS. Each PM10 increase by 10μg/m3 during the pregnancy was associated with an adjusted BW decrease of 123g (95% confidence interval, -273; 27g) as estimated by the AQMS model and 79g (-202; 44g) as estimated by the dispersion model. Results for NO2 showed statistically weaker associations with BW, but were consistent among the three exposure models. Conclusion: Dispersion model NO2 estimates were in moderate agreement with the geostatistical model but exhibited similar associations. PM10 levels from the AQMS and the dispersion models were fairly concordant, and associations with BW were somewhat statistically stronger for the AQMS model.