From eyeballing to statistical modelling : methods for assessment of occupational exposure

In this thesis methods for assessment of occupational exposure are evaluated and developed. These methods range from subjective methods (qualitative and semiquantitative) to more objective quantitative methods based on actual measurement of personal exposure to chemical and physical agents.In chapter 2, data from a general population cohort of 878 men from the town of Zutphen, the Netherlands, were used to evaluate the performance of two general job- exposure matrices. Exposures, generated by the job-exposure matrices on the basis of job histories, were compared. The validity of those exposures was measured against exposures reported by the participants in 1977/1978. The performance of the different exposure measures was assessed in proportional hazards analyses of lung cancer morbidity incidence. The two general job-exposure matrices generally disagreed with regard to exposure classification because of differences in exposure assessment and the level of detail of the job axis. When compared with self-reported exposures, the sensitivity of both job-exposure matrices was low (on average, below 0.51), while the specificity was generally high (on average, above 0.90). Self-reported exposures to asbestos, pesticides, and welding fumes showed elevated risk ratios for lung cancer, which were absent for exposures generated by the two job- exposure matrices. A population-specific jobexposure matrix was proposed as an alternative to general job-exposure matrices developed elsewhere. Such a matrix can be constructed from the results of indepth interviews of a job-stratified sample of cohort members. Sound validation and documentation of exposure assessment methods used in job-exposure matrices were recommended.In chapter 3 a study is described in which a method for semi-quantitative estimation of the exposure at task level was used and validated with actual measurements in five small factories. The results showed that occupational hygienists were in general the most successful raters. Plant supervisors and workers handled the estimation method less successfully because of more misclassification of the tasks. The method resulted, in general, in a classification of tasks in four exposure categories ranging from no exposure to high exposure. The exposure categories correlated positively with mean concentrations, but showed overlapping exposure distributions. This resulted in misclassification of the exposure for individual workers when a relatively large inter-individual variability in exposure levels within an exposure category was present. The results showed that this method can be used for workplace exposure zoning, but that the usefulness of the estimates for epidemiological purposes was not clear-cut and depended strongly on the actual exposure characteristics within a workplace. A combination of the semiquantitative exposure estimation method together with assessment of the exposure levels by measurements makes a rearrangement of tasks or individual workers possible and could improve the validity of this method for epidemiological purposes.In chapter 4 the performance is studied of nine occupational hygienists, who semiquantitatively estimated the exposure to methylene chloride and styrene in a small polyester factory. They ranked the jobs from low to high exposure, and subsequently classified them into three exposure categories (0-½TLV, ½TLV-TLV, and>TLV). The influence of quantitative exposure data on the results of the estimations was studied. Therefore, three estimations were performed. The first estimation was made after a visit to the workplace; the second and third were made after limited exposure data were presented. The ranking of styrene exposure was, in general, poor compared to the ranking of methylene chloride exposure. Physical properties, such as perception of smell, application in the process, and level of exposure might be the reasons for this striking difference. Classification of exposure into quantitative exposure categories was poor without knowledge of actual exposure data. No differences in the performance of the occupational hygienists between the two solvents were present. The results suggested that the success of an exposure estimation method depends on the type of exposure (kind of chemical, use, appearance), the available information on jobs and process, and the kind of estimate (ranking or classification). Semiquantitative classification of exposure by occupational hygienists appears to be better if they have a limited set of air sampling data at their disposal. Ranking of jobs can be performed successfully without exposure data, but a detailed description of the workplace and tasks is needed. More insight is needed concerning the influence of the chemical type, exposure pattern(s), and raters' experience on the results of semiquantitative ranking methods.Chapter 5 describes an exposure survey in 10 rubber-manufacturing plants. Personal exposures to airborne particulates, rubber fumes and solvents, and also dermal contamination, were measured. To identify factors affecting exposure the personal exposure levels and information on tasks performed, ventilation characteristics, and production variables were used in multiple linear regression models. The exposure was generally very variable. The specific circumstances in each department of each plant determined the actual levels of exposure to a large extent. The factors affecting exposure turned out to be different for each of the types of exposure considered. The model for exposure to airborne particulates explained 40% of the total variability and incorporating the actual time spent on a task only slightly improved the model ( R2=0.42) . The handling of chemicals in powder form was the main factor affecting exposure, forced ventilation having a negligible effect. The model for exposure to curing fumes (measured as the cyclohexane-soluble fraction of the particulate matter) explained 50% of the variability. Both curing temperature and pressure determined the level of rubber fumes. Local exhaust ventilation showed a significant exposure reducing effect. The effect of curing different elastomers was not statistically significant. Dermal exposure to cyclohexane- soluble matter could only be explained to a limited extent ( R2=0.22) . Tasks with frequent contact with (warm) compound and maintenance tasks in the engineering services departments resulted in high dermal exposure. Tasks in which solvents were directly used explained 56% of the variation in solvent exposures. Exposure data together with information on tasks, methods of work, ventilation and production throughout a branch of industry, can be used to derive empirical statistical models which occupational hygienists can apply to study factors affecting exposure. These determining factors are of crucial importance, whenever hazard control or epidemiologic research is the ultimate goal. In chapter 6 the implications of exposure variability are examined for the design of occupational epidemiology studies in the rubber industry. The efficiency of different grouping schemes for exposure to particulates, dermal exposure to cyclohexane-soluble contaminants, and exposure to solvents was assessed. Statistical parameters for contrast in average exposure and precision of average exposure were developed to enable comparison of different grouping schemes. Groupings based on job title, plant, factors affecting exposure, published classifications, and the ISCO-ILO classification were compared. Grouping of exposure to particulates and dermal exposure appeared to be less efficient than grouping of exposure to solvents. Grouping of solvent exposure using either occupational title groups, existing classification schemes, and schemes based on factors affecting exposure showed comparable high resolution in exposure levels. Even the most detailed grouping schemes based on the combination of plant and occupational title group showed relative modest resolution in particulate and dermal exposure levels. Groupings based on factors affecting exposure showed for these exposures similar resolution, but were more efficient because of a higher precision due to a smaller number of groups. It was concluded, that application of optimal exposure grouping strategies will benefit new research on cancer among rubber workers. Eventually, this might resolve the unwanted situation in which a complete industry was included on the list of proven human carcinogens.Chapter 7 focuses on within- and between-worker exposure variability. A database of approximately 20,000 chemical exposures was constructed in close co-operation between the School of Public Health of the University of North Carolina at Chapel Hill and the Department of Air Pollution of the Wageningen Agricultural University. A special feature of this database was that only multiple measurements of exposure from the same workers were included. This enabled estimation of within- and between-worker variance components of occupational exposure to chemical agents throughout industry. Most of the groups were not uniformly exposed as is generally assumed by occupational hygienists. In fact only 42 out of a total of 165 groups (25%), based on job title and factory, had 95% of individual mean exposures within a two-fold range. On the contrary, about 30% of the groups had 95% of individual mean exposures in a range which was greater than 10-fold. Environmental and production factors were shown to have distinct influences on the within-worker (day-to-day) variability, but not on the between-worker variability. Groups working outdoors and those working without local exhaust ventilation showed more day-to-day variability than groups working indoors and those working with local exhaust ventilation. Groups consisting of mobile workers, those working with an intermittent process and those where the source of contamination was either local or mobile also showed great day-to-day variability. In a multivariate regression model, environment (indoors- outdoors) and type of process (continuous-intermittent) explained 41 % of the variability in the within-worker component of variance. Another model, in which only type of process (continuous-intermittent) had a significant effect, explained only 13% of the variability in the between-worker component of variance.In chapter 8 the results are reported of a large survey of occupational exposure to 60 Hz magnetic fields conducted among randomly selected workers in five electric power companies. The design of the study facilitated the examination of exposure variability and provided the base for a job-exposure matrix (JEM) for linking health outcomes and occupational magnetic field exposures in the epidemiological study of employees of these companies. Almost 3.000 successful measurement attempts indicated average exposures ranging from 0.11 μT for 'Senior Managers' to 1.50 μT for 'Cable Splicers'. The differences among the five companies were relatively small with the more urban companies showing somewhat higher average exposures. The day-to-day component of variance exceeded the within- and between- group components of variance. The final JEM consisted of five groups with average exposure levels of 0.12, 0.21, 0.39, 0.62, and 1.27 μT, respectively. Given the variance in exposure, even this optimal grouping showed considerable overlap in exposure between adjacent groups. Nevertheless, the JEM incorporated the differences in exposure level within occupational categories between companies in the most efficient way and provides an objective and statistically based method for estimation of cumulative magnetic field exposure.Finally, in chapter 9 a general discussion and conclusions are given. Through validation and methodological studies, as described in the thesis, some light has been shed on the science of occupational exposure assessment. Although improvement of subjective methods is feasible to some extent, the inherent pitfalls can lead to exposures estimates not accurate enough to be used in epidemiological exposure-response relationships. Statistical models, as developed in this thesis, to unravel factors affecting exposure and to estimate variance components will contribute to more accurate ways of exposure assessment. Application of the developed statistical methods to optimize the grouping of exposure will result in less misclassification and bias and therefore in better exposure-response relationships. Consequently, this will lead to more protective occupational exposure limits. Hopefully, more randomly collected quantitative exposure data will become available to make use of the developed tools. Only then, the widely criticized art of retrospective guessing of occupational exposures will become obsolete.