Superfund Research Program
A Holistochastic Approach to Human Exposure Assessment
Project Leader: George Christakos
Grant Number: P42ES005948
Funding Period: 1995 - 2006
The new Bayesian Maximum Entropy (BME) modelling approach developed at UNC has gained international recognition as an efficient and general space/time exposure analysis and mapping tool. It includes most existing mapping techniques as special cases and it can be applied in physical situations in which existing methods cannot be used. It has been already used in several human exposure studies, and has led to useful collaborations with other scientists in the environmental health field. Its range of applicability includes both microscopic investigations (mapping cell multiplication patterns, carcinogenesis, etc.) and macroscopic investigations (pollutant exposure maps, health effect assessment, population damage indicator characterization, etc.). Thus, it provides a superior tool for health risk assessment at multiple scales.
It is widely recognized that there is a need for a holistic framework that will offer an integrated assessment of environmental exposure to pollutants and the resulting ecological/health impacts. Project investigators originally proposed such a holistic framework, using the BME method, in "Spatiotemporal Environmental Health Modeling" (Kluwer Acad. Publ., 1998). During this past year important developments of the BME method in a statistical context were presented in a new book titled "Modern Spatiotemporal Geostatistics" (Oxford University Press, 2000). The researchers have essentially established the theoretical background for the processing of a wide range of knowledge bases relevant to human exposure.
Considerable effort was devoted this past year to finding contaminant data that may lead to a practical implementation of the BME exposure methods. This effort is critical in selecting important human exposure and risk assessment situations where BME can be used meaningfully and efficiently. This effort has resulted in the exploratory data analysis of Superfund contaminants from both the National Priority List (NPL) sites and from non-NPL sites. Currently the dataset encompasses networks consisting of 10-20 monitoring wells and about 5-10 measuring events.
Project investigators are in the process of building a numerical library of BME-based space/time variability analysis. This library is currently developed and tested to reach a high level of numerical efficiency and robustness, so that it can be used by a wide group of health scientists involved in human exposure analysis and risk assessment. Such a high-quality library allows scientists to work more efficiently in applying the theoretical BME tools, and it fosters collaborative efforts with other research groups interested in using these methods.
The BME numerical tools have been used in several studies of human exposure analysis and risk assessment, as documented in publications derived from this project. One example is the ongoing effort to analyze space/time mortality data from 58 counties in California. A modeling technique was developed for mapping mortality rates at a small scale on the basis of uncertain (soft) mortality counts at larger spatial scales. Space/time maps were obtained predicting the distribution of local death rates in California at any day of interest. This case study demonstrated the adequacy of BME to obtain health indicator maps at any space/time scale of interest, which helps develop understanding of the health impact of localized environmental contamination (such as that occurring in Superfund sites).
The researchers also investigated the computational aspects of incorporating physical and biological laws (in the form of differential equations) into the BME framework. In particular, the advection-reaction processes for pollutants in a realistic hydrologic system is being considered. Project investigators calculated the general knowledge-based probability density function by solving a non-linear system of integro-differential equations. The analysis included both Gaussian and non-Gaussian densities leading to very promising results. In certain cases, the method was successfully tested by comparing its numerical results to the analytical solutions available. However, the numerical complexity of the problem represents a challenging computational problem that will be further studied during the next year.