Development and Evaluation of a Model for Estimating Long-term Average Population Ozone Exposures of Children Jianpmg Xue1, Haluk Ozkaynak1, Valerie Zartarian1, John Spengler2 1. t'SI.l'A National Exposure Research Laboratory, RTP, NC, USA 2. Harvard University School of Public Health, Boston, MA, USA Introduction Ozone is an oxidant gas that has been shown to exert a variety of adverse effects on the human respiratory system. Accurate estimates of personal ozone exposure are important for human health risk assessments. Because personal ozone measurements are ideal but expensive to collect, modeled estimates of population ozone exposure can be used to assess its importance. A hierarchical regression model was used to estimate long-term (over one year) population ozone exposure. It was found that a simple model with easily accessible data can reasonably predict long-term population personal ozone exposure and help assess related health effects. Time (date) Figure 1. Temporal profile of ozone concentrations and ratios of personal to central, outdoor and indoor ozone concentrations Method 1) Data obtained from the Harvard California Chronic Ozone Exposure Study a) about 200 children ages 6-12 years followed for 1 year (6/95 - 5/96) b) detailed information on time activity and housing characteristics collected for each study subject; measurements of personal, indoor, and outdoor ozone concentrations collected for each subject c) personal ozone samplers worn for 6 consecutive days each month. d) indoor and outdoor ozone concentrations at participants' home monitored using passive ozone samplers 2) Central ozone concentrations derived from AIRS matched by GIS 3) Randomly-assigned two portions of data for Upland and Mountain areas, respectively: one for fitting models, other for model evaluation 4) Hierarchal-fitting regression models developed with time activity data, central outdoor ozone, outdoor ozone near children's homes, children's indoor ozone, etc. 5) Used fitted parameters and models independently to predict personal ozone exposure for the two geographic areas. 6) Used R2s and coefficients to check fits for the different models Table 1. Summary statistics of ozone concentration (ppb) by locations and sites Ozore site mean std p50 P5 p95 Outdoor Ozone Cone, at Central site Mountain Area 45.9 19.1 43.9 22.6 79.0 Outdoor Ozone Cone, at Central site upland 29.0 15.2 28.2 8.5 54.3 Inddoor Ozone Cone. rear Kid Home Ivbuntam Area 10.8 13.9 3.5 0.6 40.3 Inddoor Ozone Cone. rear Kid Home uplard 6.7 7.7 3.5 0.6 24.6 Outdoor Ozone Cone, rear Kid Home IVbuntain Area 46.2 18.3 43.8 22.5 75.1 Outdoor Ozone Cone, rear Kid Home uplard 32.5 17.5 30.3 8.3 63.3 Personal Ozone Cone, rear Kid Home Mountain Area 13.8 13.6 8.6 0.6 39.6 Personal Ozone Cone, rear Kid Home uplard 11.6 10.0 8.5 0.6 30.8 Fit R2: predicted personal ozone using parameters from one portion of data and observed personal ozone from the model evaluation data, its coefficient is Fit Coeff Table 2. Results of hierarchal-fitting regression models i . ± § i i ii ill 91 ± Figure 2. Ozone concentration variations in different measurements and by season. Upland Area Mountain Area Indep endent V ariable s Model R2 FitR2 FitCoeff. Model R2 FitR2 FitCoeff. percent time indoor 0.239 0.186 0.52 0.514 0.510 1.21 Central outdoor ozone 0.715 0.785 0.98 0.792 0.782 1.11 Time-weghted central outdoor ozone 0.705 0.574 0.87 0.772 0.758 1.18 Near-home outdoor ozone 0.784 0.848 0.98 0.727 0.682 1.07 Tim e-weghted Near-home outdoor ozone 0.751 0.550 0.79 0.753 0.707 1.13 Indoor ozone 0.713 0.888 1.02 0.889 0.918 1.06 Time-wegjhted indoor ozone 0.683 0.895 1.03 0.873 0.911 1.06 Outdoor and indoor ozone 0.833 0.909 0.95 0.917 0.932 1.07 Time-wei^it indoor and outdoor ozone 0.853 0.816 0.89 0.945 0.933 1.06 Central outdoor and indoor ozone 0.799 0.890 0.95 0.919 0.937 1.08 Time-wie^ited central outdoor and indoor ozone 0.828 0.840 0.93 0.939 0.939 1.08 Figure 3. Linear regression of personal ozone with other ozone parameters in Upland Figure 4. Linear regression of personal ozone with other ozone parameters in Lake area Results Conclusions • Outdoor and central outdoor ozone » personal ozone exposure and average outdoor and central ozone concentrations were very similar • Variability of ratios of personal ozone to outdoor or central ozone is small, while that with indoor ozone is much greater across whole year (see figure 1) • Concentrations were much higher in May-Sept. than other months (see figure 2) • Best models fit when indoor ozone used with outdoor or central ozone, R2s range from 0.8 to 0.9 with almost 100% accuracy • Worst model resulted when using only activity time data (in term of R2 and accuracy); time- weighted ozone concentrations do not help model prediction and accuracy • Results are consistent between Upland and Mountain areas (see table 2, figures 3 and 4) • Modeling with central outdoor ozone concentrations which are easily and cost-effectively obtainable can predict personal ozone exposure with reasonable prediction (R2 range from 0.7 to 0.9 and accuracy is about 90% for average personal ozone, about 10% over-prediction) References: Schwartz, J. Lung function and chronic exposure to air Pollution: A cross-section analysis of NHANES II; Environ. Res. 1989, 50, 309-321 J. Xue et al, Parameter evaluation and model validation of zone exposure assessment using Harvard Southern California chronic ozone exposure study data; J. Air & Waster Manage. Assoc.2005, 55:1508-1515 A model with only central ozone concentrations can predict long-term ozone personal exposures with reasonable accuracy. This model could help decision makers for controlling ozone concentrations at population level and reducing health risks from ozone exposure. The United States Environmental Protection Agency through its Office of Research and Development funded and managed the research described here. It has been subjected to Agency review and approved for publication. ------- |