51
DEVELOPMENT OF A JOB DICTIONARY FOR EXPOSURE RECONSTRUCTION IN A RETROSPECTIVE
EPIDEMIOLOGICAL STUDY.
H. Basara, P. Jones, T. Hall, The University of Oklahoma, Oklahoma City, OK; R. Leonard, Dupont, Wilmington, DE; N. Esmen, University of Illinois at Chicago, Chicago, IL.
The development of a coherent job dictionary is a critical component of reconstructing exposures in retrospective epidemiological studies. To construct a job dictionary, medical and personnel records must be reconciled with industrial hygiene (IH) exposure records so that estimated exposures are correctly assigned. In this study there were three plants producing the same products, with 3573 distinct job titles identified from medical and personnel records. A detailed review of each entry resulted in the condensation of the original job title listing to approximately half of those initially identified. The remaining distinct job titles were further categorized into 10 basic groupings, based on similar tasks and work areas, each with a specific job code. Job dictionary codes were assigned to exposure records by comparing the job and task descriptions on IH reports of exposure measurements to those used to create each job code. However, IH records were often recorded using a variety of job titles or descriptions of activities which did not match the official job titles provided by personnel and medical records. Differences were resolved by comparisons with process/work area and task information. Other challenges included variations between plants; changes in ownership and management structure through plant history; and changes in production processes that resulted in differing terminologies used to describe work areas, production processes, and occupations. Each facility studied had its own job dictionary to account for the variety of job titles observed between the plants. Thus, the development of an accurate job dictionary, which is essential for a representative and accurate exposure reconstruction, is not as straightforward and precise as reported in many retrospective epidemiologic studies. The construction of a job dictionary is a subjective, decision laden process that has the potential to contribute a large component of the observed exposure misclassification inherent in occupational epidemiological studies.
52
THE LINKAGE OF A JOB EXPOSURE MATRIX WITH EXPOSURE MEASUREMENTS FOR OCCUPATIONAL
EXPOSURE ASSESSMENT IN AN EPIDEMIOLOGY STUDY.
J. Coble, R. Vermeulen, S. Xue, B. Ji, W. Chow, M. Dosemeci, NCI, Rockville, MD.
The retrospective assessment of exposures for occupational epidemiology studies is a difficult task that, based on knowledge and experience, industrial hygienists may be asked to conduct. For studies of chronic disease, the exposure of interest may have occurred several decades in the past, and can change over time. For population-based studies, in particular, a wide variety of occupations and industries must be assessed. In this type of study, an epidemiologic analysis based on occupation or industry titles has limited power to detect associations with disease due to small numbers when detailed job titles are used, and the heterogeneity and low specificity of exposure within broader occupational groups. One approach for improving occupational exposure assessment is the development and use of job exposure matrices (JEMs) that contain semiquantitative ratings for the intensity, probability, and frequency of exposure for specific agents by occupation and industry. Along with JEMs, occupational exposure measurement data can be also be useful for conducting retrospective exposure assessments, however, measurement data are often available for only a limited number of occupations and industries. The combination of a JEM with exposure measurement data can provide more comprehensive quantitative exposure estimates, as well as improve the reliability of the JEM ratings. In this presentation, methods are described for the linkage of a JEM with a database of historical exposure measurements that was then used to develop quantitative estimates for exposure to benzene by industry, occupation, and time period for a population-based cohort study of women in Shanghai, China. Estimates for the overall prevalence of exposure to benzene by intensity level are presented, along with a regression analysis of time trends in exposure levels.
53
THE EFFECT OF EXPOSURE BAND WIDTH ON MISCLASSIFICATION RATES IN OCCUPATIONAL
EXPOSURE ASSESSMENT.
M. Phillips, University of Oklahoma, Oklahoma City, OK.
In retrospective occupational exposure assessment and other applications, a common practice is to define quantitative exposure classes or “bands” that are of uniform width on a log scale. Due to between-worker variability, a fraction of individual workers will be misclassified if they are assigned to an exposure class based on the between-worker median exposure for their job category. If individual median exposures are lognormally distributed, and the between-worker median exposure for a particular job category happens to fall near an exposure class boundary, the probability of misclassifying individual workers will be close to 50%. The probability of the between-worker median falling near a class boundary decreases with increasing band width. The expected misclassification rate can be calculated as function of the band width and the between-worker geometric standard deviation (GSD). The “Solver” tool in Microsoft Excel was used to determine the critical value of the band width relative to the between-worker GSD that would result in an expected misclassification rate equal to a predetermined level for all job categories. It was found that the band width would have to be approximately four times the GSD to yield an expected misclassification rate of 20%, eight times the GSD for a misclassification rate of 10%, and 16 times the GSD for a misclassification rate of 5%. Thus, very broad band widths would typically be needed to ensure that no job categories could have an expected misclassification greater than 10%. This is, however, a very stringent criterion. Alternatively, the class boundaries could be chosen so that the broadest distribution among the important job categories is symmetrically placed between the boundaries of the applicable exposure class. The band width could then be set such that the misclassification rate for that job would not exceed an acceptable level.
54
EXPOSURE RECONSTRUCTION DATA FOR A COHORT MORTALITY STUDY OF JET ENGINE
MANUFACTURING WORKERS.
S. Lacey, N. Esmen, K. Kennedy, University of Illinois at Chicago, Chicago, IL.
A large occupational epidemiologic investigation in collaboration between the University of Pittsburgh and the University of Illinois at Chicago is underway to determine cause-specific cohort mortality among the employees of a jet engine manufacturing company. The subjects comprise all workers (approximately 250,000) who ever worked at any of the seven distinct manufacturing facilities between 1952 and 2002 inclusive. One component of the investigation is the exposure reconstruction, which is designed to quantitatively estimate occupational exposures possibly related to potential health outcomes. The examination of the company archives, extensive meetings with management, human resources, and engineering personnel, union representatives, and both hourly and salaried employees have revealed massive amounts of data that could be used in the exposure reconstruction. These data sources include, but are not limited to: employment records, part manifest and manufacturing process documentation, manufacturing operation locations in space and time, detailed operation procedures, worker time-analysis studies, process machinery brochures, and standard operating procedures that go back to the beginning of the study, as well as more than 60,000 pages of exposure measurement or characterization documentation. Since the estimated amount of available data exceeds 23 million pages, rational examination of all the documents and abstracting the necessary information is unrealistic. Therefore, while ideally all data is important to exposure reconstruction, the practical limitations superimposed by the amount of data requires detailed examination of the relative importance of each type of information in exposure reconstruction. The details of an analytical framework developed are presented that describe the connectivity among pieces of data and how a data set can be organized in a hierarchy to provide the information necessary for exposure reconstruction.
55
RECONSTRUCTION OF SILICA AND NON-SILICA EXPOSURES AND LUNG BURDENS USING A
BAYESIAN APPROACH.
G. Ramachandran, S. Banerjee, R. Hoffbeck, B. Alexander, University of Minnesota, Minneapolis, MN; C. Rice, University of Cincinnati, Cincinnati, OH.
The North Carolina Dusty Trades program, which is a large data set of silica exposures and silicosis cases has been re-analyzed to estimate historical exposures for (a) silica and (b) non-silica dust for each job-code in the program for each workplace for the time period 1935–1980. The re-analysis of the exposure assessment was done for a case-control epidemiological study of the dose-response relationship between particulate-containing crystalline silica and silicosis (216 cases and 672 matched controls). Our hypothesis was that a cumulative exposure metric that includes both silica and non-silica dust is more related to the development of silicosis (for which the mechanism is still undetermined) than cumulative silica exposure alone. The exposure data set contained task exposures (impinger samples from 1930s–1970s) and job exposures (cyclone samples from late 1970s to early 1980s). An empirical factor was used to convert impinger data from count to mass concentration. The exposure reconstruction accounted for uncertainties in exposure data and lung clearance mechanisms that affect the cumulative lung burden using pharmacokinetic data on clearance of inhaled dusts. Pharmacokinetic modeling for estimating the cumulative lung dose of silica and non-silica dust allowed estimation of the time to macrophage overload. Due to the sparseness of exposure measurements, a novel Bayesian methodology that synthesized expert judgment and historical measurements was used for exposure reconstruction. Four experts were provided with available information on workplace conditions and worker activities, and they provided estimates of task exposures in probabilistic terms. These “prior” exposure judgments were updated with the data to provided Bayesian “posterior” distributions. The exposure distributions and classification of worker exposures using traditional exposure assessment methods were compared with the results obtained using the Bayesian approach. A companion abstract will present details on the epidemiological findings from the analysis of the case-control study.
56
EFFECT OF NON-SILICA RESPIRABLE PARTICULATE MATTER ON DEVELOPMENT OF SILICOSIS.
B. Alexander, G. Ramachandran, R. Hoffbeck, University of Minnesota, Minneapolis, MN; C. Rice, University of Cincinnati, Cincinnati, OH.
The objective of this study was to determine whether nonrespirable silica particulate matter encountered in jobs with varying silica exposure influences the development of silicosis by contributing to lung overload, and whether the timing of the lung overload is associated with development of silicosis. A nested case-control study of the North Carolina Dusty Trades cohort compared the silica and non-silica respirable particulate exposure history of silicotics and non-silicotics. Cases identified through the Dusty Trades Program were hired after 1930 or actively working in 1935. Controls were matched to cases by age, race, year of hire, and time in surveillance program. Up to four controls were selected for each case. Exposure models, aided by expert judgement and a Bayesian framework, estimated monthly cumulative silica and respirable non-silica particulate exposure. A companion abstract describes the exposure assessment methodology. A lung deposition model was used to estimate cumulative lung burden and time (months from first employment) at which the lung clearance mechanisms are overloaded, and accounted for periods of nonexposure, such as military leave and removal from exposure. Risk of silicosis was modeled with conditional logistic regression. The 216 cases and 673 controls had median cumulative lung burdens of 567 and 427 ug/kg lung weight, respectively. The median estimated time to lung overload was 133 months for cases and 198 for controls. Adjusted for cumulative silica exposure and cumulative exposure to other particulates, the risk of silicosis increased as the number of months from first employment to lung overload decreased (OR = 0.9957, 95% CI = 0.9941–0.9974); this corresponds to an OR of 0.95 (95% CI = 0.93, 0.97) for each year of employment without reaching overload. Silica and non-silica respirable particulate exposure contributes to lung defense mechanism overload, which makes the lung more susceptible. Moreover, intense exposures early in a career may be particularly influential in developing silicosis.
Posted May 30, 2005