Exposure Assessment for Cohort Based on General Population Living Near Industrial Complex

Abstract

ISEE-753 Objective: To construct the outside exposure levels of air pollutants including PM10, O3, CO, NO2, and SO2, for a general population and to test the validity of spatial interpolation estimate method. Materials and Methods: We used the kriging interpolation to estimate the concentration on unmeasured areas (grid: 500 m × 500 m). Based on the semivariogram (lag distance: 2956 m) patterns of time periods over 1 hour, exponential model was adopted to estimate the spatial relationship. For the final validation, cross-validation method was applied and correlation coefficients between measured and estimated levels of air pollutants were calculated. Results: Daily variations of monitored concentration levels of each pollutant were considerably high. Daily outside air pollutant concentration of 443 subjects over 7 months were estimated. Geometric mean concentrations for 24 hours based on a 1 day lag period were 38.0 μg/m3, 12.6 ppb, 21.8 ppb, 4.0 ppb, 421.5 ppb for PM10, O3, NO2, SO2, CO, respectively. Estimated daily PM10 levels were strongly correlated with other pollutants indicating that PM10 varies concurrently with other pollutants, most probably due to the similarity in pollution sources or sharing of common causal relationships with other pollutants. Correlation based on daily variation was much stronger than the ones based on periods of a month or 7 months, partially suggesting the necessity of high time resolution of exposure assessment. Correlation coefficients of measured and estimated level of ozone was around 0.83, indicating decent validity of our exposure assessment. Conclusions: Even though residential areas have fewer monitoring stations, these results of decent validity with high correlation coefficients suggest the feasibility of spatial interpolation approaches for the exposure assessment of subjects living near industrial complexes. Meteorological factors still need to be considered to improve the estimations. The results also show that prediction is improved much stronger with high time resolution approaches.