![[logo] US EPA](https://webarchive.library.unt.edu/eot2008/20090513082404im_/http://www.epa.gov/epafiles/images/logo_epaseal.gif){kind=logo}
Issues and Challenges in Modeling Children's Longitudinal Exposures: An Ozone Case Study
Halûk Özkaynak; Jianping Xue; Valerin Zartarian
U.S.EPA, ORD National Exposure Reesearch Laboratory, RTP, NC, USA
Modeling children’s exposures is a complicated, data-intensive process. Modeling longitudinal exposures, which are important for regulatory decision making, especially for most air toxics, adds another level of complexity and data requirements. Because it is difficult to model inter- and intra- personal variability for exposure model inputs, there is potential for inaccurate estimation of upper percentiles of longitudinal exposure distributions. In order to develop a scientifically sound exposure prediction model, we need to resolve how to do the following: (1) obtain longitudinal data needs for time activity and pollutant measurements; (2) separate intra- and inter- person variability of model inputs; (3) link inputs from different data sources and fit those data as inputs for the model, which will preserve its variance-covariance structure; and (4) use cross-sectional data to simulate longitudinal data, which is difficult and expensive to collect.
In this presentation, we address these issues by applying both an existing and a new technique for estimating personal ozone exposures of school-age children living in two California communities. The first modeling analysis employed commonly used methods for estimating exposures using a microenvironmental exposure model that used independent distributions in simulations, which were fit to observed inputs for the model. In the new modeling methodology, various variance components derived from the underlying data were put back into model, so that proper variance-covariance relationships in model inputs were maintained. Contribution of intra-personal, inter-personal, seasonal, and area’s variances predicted by the new model, were: 38%, 19%, 39% and 4%, which were quite close to those derived from the original data while intra-personal variance is 91% for the old method without decomposition of the variance. The standard deviation and 99th percentile of overall personal ozone exposures from the model with the new method and the old method without decomposition of variance of intra, inter personal and other factors, were: 12 ppb, 53 ppb and 21 ppb, 89 ppb, respectively, and in comparison to 12 ppb, 52 ppb for the observed personal ozone data. Results show that this method not only can keep variance-covariance structures of inputs and output in a simulation, but also can accurately predict high percentiles of longitudinal exposures that are important for regulatory evaluations.
Disclaimer: Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy.