M24 - Uncertainty Analysis in Exposure AssessmentVentura 3:30 - 5:00 pm
|
| Chair(s): Paul Price
|
M24.1 Evaluation and Recommendation of Sensitivity Analysis Methods Applied to EPA/SHEDS Models . J.* Zheng, A. Mokhtari, H.C. Frey; North Carolina State Univ. jzheng3@eos.ncsu.edu
Abstract: Sensitivity analyses of exposure or risk models can help identify the most significant exposure or risk factors, and aid in identifying the important variability and uncertainty sources for prioritizing additional measurement needs or research in order to improve the uncertainty in estimates of exposure or risk. The U.S. Environmental Protection Agency’s Stochastic Human Exposure Dose Simulation (SHEDS) models are probabilistic and physically-based human exposure models to simulate variability and uncertainty in cumulative human exposure and dose for pollutants of interest. There is a need to develop and implement appropriate and rigorous methods for sensitivity analysis of the SHEDS models in order to identify key sources of uncertainty and provide insight regarding risk management questions. The purposes of this project are to: (1) evaluate selected sensitivity analysis methods based upon the use of a simplified version of SHEDS models, (2) make recommendations regarding the selection of appropriate sensitivity analysis methods, and (3) provide procedures regarding how the recommended methods should be incorporated into SHEDS model frameworks. The main characteristics and structures of SHEDS models are investigated. The sensitivity analysis methods evaluated including correlation coefficients, regression analysis, analysis of variance (ANOVA), Fourier Amplitude Sensitivity Test (FAST), and Sobol’s method. These methods are evaluated based upon a simplified SHEDS model that captures the same characteristics of nonlinearity, interaction, monotonicity, and thresholds as contained in the various SHEDS models. The methods are evaluated with respect to their capability to identify the contribution of individual inputs to the variance in selected outputs. The evaluation results are discussed, and recommendations regarding appropriate sensitivity analysis methods for the use of SHEDS models and implementation strategies for the recommended methods are made.
|
M24.2 Predicting the Need for Aggregate Risk Assessments. P.S. Price, C.F. Chaisson, M.A. Jayjock; The LifeLine Group, Inc. psprice@thelifelinegroup.org
Abstract: Considerable effort is being spent to assess the risks from aggregate chemical exposures (concurrent exposures to multiple sources of a single chemical). The concern that drives such assessments is that while no single source of exposure may pose an unacceptable risk to any member of a population; the aggregate exposure may result in unacceptable risks to some individuals. This potential for underestimating exposure was investigated by constructing a simple Monte Carlo model of aggregate doses across a hypothetical population exposed to multiple sources of a chemical. Underestimation of dose was modeled using the ratio of an individual’s total dose from all sources to the highest dose received from a single source (the Total to Max Ratio or TMR). Where the values of TMR are close to one, the aggregate dose is dominated by a single exposure and an aggregate assessment may not be required. The interindividual variation in TMR and the correlation of the values of TMR with the magnitude of the total dose were investigated as a function of 1) the variation of the mean dose in the population from each source, 2) the interindividual variation of the doses from a given source, and 3) correlation between doses from multiple sources. Values of TMR are highest when the mean doses received from multiple sources are similar in magnitude. Values of TMR are increased when the doses from multiple sources are correlated. Where the variation of the mean doses is large, the value of TMR is smaller and is inversely correlated with the magnitude of the total dose. Values of TMR are further reduced if the variation in the interindividual variation in dose from a single source increases. This finding suggests that the need to perform aggregate assessment is greatest when multiple sources of exposure have similar average doses across a population and when the interindividual variation doses are correlated in the population.
|
M24.4 Two-dimensional Monte Carlo analysis versus probability bounding: do they give the same answer?. D.S. Myers, W.T. Tucker, S. Ferson; Applied Biomathematics scott@ramas.com
Abstract: When analysts conduct two-dimensional Monte Carlo analyses, the results are presumed to robustly characterize the full extent of the variability and uncertainty in the forecast variables(s). This presumption may often be incorrect. We studied both simple numerical examples of exposure analyses under uncertainty and variability, as well as a more complex case study of sport fisherman exposed to organochlorines through contaminated catches. In parallel exposure assessments performed using two-dimensional Monte Carlo analysis (2MC) and probability bounds analysis (PBA), the results differed, and the differences tended to grow as the number of uncertain variables in the calculation increased. Probabilistic risks from the 2MC are consistently lower than those from PBA. This observation is disturbing because these particular PBA results are known to be mathematically optimal (as tight as possible). We have traced the discrepancies to three kinds of implicit assumptions: (i) use of distributional models that treat uncertainty as though it were variability, (ii) use of precise dependence assumptions between variables without justification, and (iii) assumptions of independence between parameters characterizing a (single) distribution. Both methodologies have considerable flexibility, and they can accommodate a variety of different analytical situations. When analysts are made aware of these implicit assumptions, they can often correct the respective analyses to make them conform to each other. However, the sampling strategy employed by 2MC does not seem capable of producing a correct characterization of the uncertainty that arises when the nature of the dependence between variables or their parameters is unknown. On the other hand, in other situations when parameters from a single distribution should be considered as independently varying, PBA will not be able to produce best-possible results.
|
M24.5 Cumulative Exposure: Implications of Benzene Exposure Assessment Uncertainty. S.L.* Collie, R.W. Bienert; Tetra Tech EM Inc. shanna.collie@ttemi.com
Abstract: Cumulative exposures ---- including workplace, lifestyle, and ambient environmental---- to a rather ubiquitous and potent carcinogen (benzene) will be evaluated using a case study approach. Among the information to be reviewed are implications of at least three main elements on final conclusions regarding chemical potency and risk: (1) retrospective exposure estimates and assumptions, (2) proxy data for “censored” or nondetect data, and (3) other missing measures of quantitative exposure. In effect, each missing exposure component contributes greater aggregate uncertainty to potency estimates and epidemiological findings. Specifically, critical review of industrial hygiene panel rating methodology and exposure assumptions covering a large timeframe will be presented. A comparison and contrast of foreign findings (such as those in Great Britain and Australia) with U.S. estimates will be presented. Statistical implications of proxy data decisions will be explored. Poorly understood uncertainty (or large aggregate uncertainty) may spotlight risk control (represented by reduced benzene concentrations) via specific U.S. environmental programs (such as U.S. EPA’s RCRA permitting and/or OSHA rulemaking) rather than on more difficult targets, such as control of nonpoint automobile emissions and tobacco smoking. This realization will become increasingly important as the U.S. EPA forges ahead with a congressionally-mandated approach that will provide a more holistic assessment of cumulative exposure and risk.
|
