You are here:
EFFECTS OF CORRELATED PROBABILISTIC EXPOSURE MODEL INPUTS ON SIMULATED RESULTS
Xue, J, V Zartarian, AND A H. Ozkaynak. EFFECTS OF CORRELATED PROBABILISTIC EXPOSURE MODEL INPUTS ON SIMULATED RESULTS. Presented at International Society of Exposure Analysis 14th Annual Conference, Philadelphia, PA, October 17-21, 2004.
The primary objective of this research is to produce a documented version of the aggregate SHEDS-Pesticides model for conducting reliable probabilistic population assessments of human exposure and dose to environmental pollutants. SHEDS is being developed to help answer the following questions:
(1) What is the population distribution of exposure for a given cohort for existing scenarios or for proposed exposure reduction scenarios?
(2) What is the intensity, duration, frequency, and timing of exposures from different routes?
(3) What are the most critical media, routes, pathways, and factors contributing to exposures?
(4) What is the uncertainty associated with predictions of exposure for a population?
(5) How do modeled estimates compare to real-world data?
(6) What additional human exposure measurements are needed to reduce uncertainty in population estimates?
In recent years, more probabilistic models have been developed to quantify aggregate human exposures to environmental pollutants. The impact of correlation among inputs in these models is an important issue, which has not been resolved. Obtaining correlated data and implementing a correlated input structure into a simulation model is difficult. This paper assesses the impact of various correlated inputs on EPA's Stochastic Human Exposure and Dose Simulation (SHEDS) model for pesticides.
The object-mouth (OM) nondietary exposure pathway in SHEDS is used to evaluate effect of two and three correlated inputs on the daily and annual averaged OM exposure. The distribution-free technique was used to generate various correlated input distributions for dislodgeable surface residues (V1,ug/cm2), ratio of residue on object mouthed to residue on floor surface (V2), object surface area mouthed per mouthing event (V3,cm2/mouthing event), frequency of object-mouth activity (V4,mouthing events/hr) and saliva removal efficiency(V5). Available published data were used to fit the original distribution.
The relevant equation is: OM exposure (ug/event)=V1*V2*V3*V4*V5*duration of macroactivity (hr/event).
Results with enforced paired and multiple correlated inputs in the model simulations show:
1) If the correlation coefficient of paired inputs is less than 0.5, the relative error of the 95th percentile of daily averaged OM exposure is less than 10%. Any degree of paired correlation will have no impact on any percentile of the annual averaged OM exposure.
2) The spread of the input distribution has an impact on the effect of correlated inputs for daily averaged OM exposure. The bigger the standard deviation of the original distribution, the larger the effect of correlated inputs on daily averaged OM exposure.
3) Correlated inputs affect standard deviation of the OM exposure more than its 95th and 99th percentile.
4) Positive and negative correlations have a cancellation effect; Multiple correlated inputs have an additive effect.
The conclusions are that big correlated input will affect central values and profile of short-term exposure and multiple correlated inputs will have joint impact on the results but correlated inputs will have no impact on predicted annual averaged exposures, and small correlated inputs (correlation coefficient < 0.5) will be insignificant in exposure model predictions even for daily averaged exposures.
Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy.