A variety of health effects are caused by chronic, cumulative exposure over time to pollutants. In these cases, to establish dose-response relationships for epidemiological and risk assessment purposes, it is vital to determine the exposures of individuals or cohorts as functions of time. Most existing occupational exposure databases, however, do not contain continuous records of historical exposures to airborne contaminants. These gaps in the historical record may be filled by using the knowledge-base that experts and professionals in the field possess. In this project, we present a new framework, based on Bayesian probablistic reasoning, for obtaining estimates of exposure histories for airborne particulates from limited historical measurements, using subjective expert judgment. The framework has great potential applications in instances where there is sparse information or missing data on past exposures. Traditional methods, using only sparely available historical measurements, result in estimates with large uncertainties. Limited information on estimates of airborne concentrations and worker exposures to airborne nickel aerosol are used to estimate the uncertainty in historical air monitoring data. This uncertainty arises from environmental variabilities systematic biases as well as uncertainty due to various measurement criteria used over a period of several decades. Retrospective exposure reconstruction based solely on such historical measurements leads to estimates with such large error bars as to be not useful for developing quantitative dose-response relationships for epidemiology. This is demonstrated in the first part of the project. However, estimation of the variance in historical measurements enables the calculation of a likelihood function which is then used to obtain posterior probability distributions for exposures. Additional information, in the form of expert judgments informed by knowledge of historical plant conditions, is brought to bear on this process. The experts are provided with an information packet that contains historical process information, process throughput levels for each year, the dimensions of the workplace, ventilation records, and task descriptions for each job category. Based on this information, the experts provide subjective prior probability distributions for input parameters to a general ventilation model that predicts building concentrations. These priors can be synthesized with the historical measurements using the Bayes formalism. The prior distributions of exposures are updated using the average measured exposures (historical measurements) and their associated variances to obtain the posterior probability distributions for building concentrations as well as concentrations at specific locations in the building. Expert input was also obtained from a plant industrial hygienist, in the form of probability distributions, regarding the amounts of time spent by each job category in different locations in the building. Monte-Carlo sampling, from the posterior probability distributions of concentrations in different micro-environments and the probability distributions of time spent by each job category in those micro-environments, was used to obtain worker exposures using a time-weighted averaging model. Results of this analysis are compared with those obtained using a more traditional methodology for retrospective exposure assessment. The methodology has been applied to obtain the concentration history of nickel aerosol in a smelter, and the exposure history of different job codes. A comparison of the concentration history obtained using only the historical measurements with that obtained using Bayesian methods shows that while the median levels obtained using the two methods are comparable, the uncertainties are greatly reduced for the Bayesian method. The same is true for the exposure history as well. The approach used here emphasizes the need for detailed information about the industrial operations, materials tasks, and other environmental variables obtained from company archives and site visits. Without such information, quantitative exposure assessment with manageable uncertainties would be impossible.
Gurumurthy Ramachandran, PhD, CIH, Division of Environmental and Occupational Health, School of Public Health, University of Minnesota, Box 807, Mayo, 420 Deleware St. SE, Minneapolis, MN 55455
School of Public Health, University of Minnesota, Minneapolis, Minnesota