Annals of Occupational Hygiene Advance Access originally published online on January 7, 2005
Annals of Occupational Hygiene 2005 49(2):155-165; doi:10.1093/annhyg/meh088
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© 2005 British Occupational Hygiene Society Published by Oxford University Press;
Company-Level, Semi-Quantitative Assessment of Occupational Styrene Exposure when Individual Data are not Available
1 Aarhus University Hospital, Department of Occupational Medicine, Aarhus, Denmark; 2 Utrecht University, Institute of Risk Assessment Sciences, The Netherlands; 3 University of Alberta, Department of Public Health Sciences, Canada
* Author to whom correspondence should be addressed. Tel: +45 89 49 42 90; fax: +45 89 49 42 60; e-mail: hkols{at}as.aaa.dk
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
In epidemiological research, self-reported information about determinants and levels of occupational exposures is difficult to obtain, especially if the disease under study has a high mortality rate or follow-up has exceeded several years. In this paper, we present a semi-quantitative exposure assessment strategy for nested case–control studies of styrene exposure among workers of the Danish reinforced plastics industry when no information on job title, task or other indicators of individual exposure were readily available from cases and controls. The strategy takes advantage of the variability in styrene exposure level and styrene exposure probability across companies. The study comprised 1522 cases of selected malignancies and neurodegenerative diseases and controls employed in 230 reinforced plastics companies and other related industries. Between 1960 and 1996, 3057 measurements of styrene exposure level obtained from 191 companies, were identified. Mixed effects models were used to estimate expected styrene exposure levels by production characteristics for all companies. Styrene exposure probability within each company was estimated for all but three cases and controls from the fraction of laminators, which was reported by a sample of 945 living colleagues of the cases and controls and by employers and dealers of plastic raw materials. The estimates were validated from a subset of 427 living cases and controls that reported their own work as laminators in the industry. We computed styrene exposure scores that integrated estimated styrene exposure level and styrene exposure probability. Product (boats), process (hand and spray lamination) and calendar year period were the major determinants of styrene exposure level. Within-company styrene exposure variability increased by calendar year and was accounted for when computing the styrene exposure scores. Exposure probability estimates based on colleagues' reports showed the highest predictive values in the validation test, which also indicated that up to 67% of the workers were correctly classified into a styrene-exposed job. Styrene exposure scores declined about 10-fold from the 1960s–1990s. This exposure assessment approach may be justified in other industries, and especially in industries dominated by small companies with simple exposure conditions.
Keywords: case control study • epidemiological method • epidemiology • exposure assessment • mixed effects models • occupational exposure • risk assessment • styrene
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
In epidemiological research, self-reported information about determinants and levels of occupational exposures is difficult to obtain, especially if the disease under study has a high mortality rate or follow-up has exceeded several years. Living relatives of deceased subjects may provide qualitative information on the determinants of exposure, but the sensitivity and specificity of such data are often questionable (Boyle and Brann, 1992
; Hansen et al. 1997). Proxy measures of exposure obtained from other sources (e.g. co-workers, employers) are therefore often used. Job-exposure matrices provide links between job titles and probabilities and levels of exposure and can be used if job titles are available (Olsen, 1988). However, in all instances quantitative exposure estimates based on actual measurements of exposure concentrations are preferred (Seixas and Checkoway, 1995). In this paper, we present a semi-quantitative exposure assessment strategy for styrene among workers of the reinforced plastics industry for which neither qualitative nor quantitative exposure information was readily available on an individual level. The strategy takes advantage of the variability in styrene exposure level and styrene exposure probability across companies and calendar year and provides independent estimates of styrene exposure level and styrene exposure probability with company as the basic unit of analysis.
The exposure assessment depends on historical airborne exposure measurements, in combination with reports on the fraction of laminators among all workers within a company provided by colleagues, employers and dealers of production raw materials. Separate exposure estimates are provided that reflect peak exposure (based on measurements lasting <1 h) and full-shift average exposure (1–8 h measurements).
The exposure assessment is part of several nested case–control studies of myeloid and lymphatic malignancies, pancreatic cancer and degenerative nervous system diseases among workers potentially exposed to styrene. The studies are an extension of earlier reports that suggested increased occurrences of these diseases in the Danish reinforced plastics industry (Kolstad et al., 1994, 1995, 1996).
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
The nested case–control studies included 1522 cases and controls (index subjects). They were identified among 78 908 workers employed between 1964 and 1997 in 552 Danish companies producing reinforced plastics or related products. The follow-up period extended from the start of employment to the end of 1997. Information on the workers' personal identification numbers, identity of employing companies and duration of employment in these companies were ascertained from the national pension fund; health effects in the national registries of cancer and hospitalisations; and vital status in the national population registry (Kolstad et al., 1993
). At the time of data collection 48% of the index subjects were dead. Among 781 living index subjects, 427 (55%) provided information about their work in the reinforced plastics industry. This information was also obtained from the next-of-kin of 186 dead index subjects. We thus had no access to individual exposure information for 909 index subjects (60%). The index subjects were employed in 230 index companies out of the 552 identified companies.
Styrene exposure measurements
The Danish National Institute of Occupational Health has collected a total of 2454 personal measurements of airborne styrene levels between 1960 and 1996 at the request of the labour inspection service. Personal measurements were also identified from several internal company surveillance programs (603 measurements). Company name, sampling time and calendar year were recorded in addition to the styrene level (mg/m3) for each sample. The type of task done was recorded for 2411 out of all 3057 measurements and it was found that 87% of these measurements were taken during manual lamination. Only measurements from companies with at least two samples were considered. Stationary samples and Dräger tube samples were not considered. Some repeated measurements among individual workers were obtained, but they could not be systematically identified in the database. Until 1970, measurements lasting more than 15 min were based on air collected close to the breathing zone of an employee through impingers containing ethanol. Analysis was made by either spectrophotometry or gas chromatography. For samples with a brief sampling period (
1 min), air from close to the breathing zone of an employee was collected in gas-tight pipettes and analysed by gas chromatography. From 1970, a personal sampling technique was applied. Air was drawn through an absorbent layer of charcoal and analysed by gas chromatography. Detailed descriptions of sampling and analytical methods are no longer available.
Until the 1970s, the labour inspection service focused on worst-case situations, while after 1980 measurements were made mainly to evaluate their compliance with the exposure limits in an exposure surveillance program. Differences in sampling strategies between samples were adjusted for in subsequent statistical analyses, and all predictions made with statistical models were standardised to samples collected in the surveillance program, judged to be more representative of the working conditions. Descriptive data for a sub-sample of the dataset has been published previously (Jensen et al., 1990).
Company production characteristics
Descriptions of product, process and, for a subset of companies, annual amount of polyester used and surface area of products, were obtained from current and former employers and from two dealers of plastic raw materials, who possessed knowledge of the Danish industry dating back to the early 1960s or from the styrene measurement reports.
Styrene exposure probability
Because of the lack of individual exposure information for as much as 60% of the index subjects, we estimated the styrene exposure probability for index subjects during employment in the index companies. We provided these estimates for index subjects with and without individual exposure information obtained directly from them or from their next-of-kin. Five different estimates of the fraction of exposed workers within each company (classified at three levels: 0%, 1–49% and 50–100%) were achieved based on information from living colleagues of the index subjects (Colleagues I and Colleagues II), employers of index companies (Employers I and Employers II) and the dealers of plastic raw materials (Dealers).
The colleagues selected were as comparable as possible with the index subjects with respect to styrene exposure status. We considered the employing company, calendar year of employment, duration of employment and sex to be relevant determinants of styrene exposure and furthermore, this information was available for the complete study population from the national pension fund. We therefore stratified the index subjects by company (230 companies), calendar year (<1970, 1970–1979, 1980–1989, 
1990), duration of employment (<1 year, 1–4 years, 5 years) and sex. Within each stratum we randomly selected two living colleagues. We identified a total of 1384 colleagues and 945 of them (68%) participated in a telephone interview (n = 714) or completed a questionnaire (n = 231). They were asked if they worked as (i) a laminator; (ii) if not, did they work in a room where lamination was conducted by others (bystander exposure); (iii) if not, did they do other work.
Based on the colleagues' information each stratum was classified according to the relative number of laminators [(i/i + ii + iii), Colleagues I]. If none of the colleagues were laminators, that stratum was classified with an exposure probability of 0%, if there were one laminator and one non-laminator, 1–49%; if both were laminators, 50–100%. If only one worker was identified that stratum was classified as either 0% or 50–100%. The scientific ethical committee did not allow us to ask colleagues about the individual index subjects' work.
In addition, the colleagues were asked to provide their own estimate of the proportion of the total workforce engaged in lamination tasks in the company (Colleague II). For each stratum we computed one estimate by averaging. These two measures of the likelihood that any worker within each stratum was exposed to styrene were subsequently linked by stratum with 1245 of the 1522 index subjects (82%).
The dealers and the employers also provided estimates on the proportion of the total workforce engaged in lamination tasks (Dealer and Employer I). The employers, in addition, provided estimates of the proportion of all working hours of the company spent on lamination, to take account of companies mainly employing part-time laminators (Employer II). The employers provided exposure estimates for 1234 index subjects and the dealers for 1421 index subjects.
The five estimates of exposure probability were validated among the 427 living index subjects who were interviewed or filled in a questionnaire and, like the colleagues, reported if they had worked as (i) a laminator; (ii) if not, did they work in a room where lamination was conducted by others or (iii) if not, did they do other work. The reports were obtained independently of the reports obtained from the colleagues, employers and dealers and were regarded as the gold standard. We computed the sensitivity of the exposure probability estimates among the laminators (i) and the specificity among the non-laminators [(ii) + (iii)]. For these computations we dichotomised the exposure probability estimates and regarded a value of 1% or above as a ‘positive test’. We also assessed the ability of the exposure probability estimates to classify correctly a laminator (positive predictive value).
For the epidemiological analyses we classified index subjects by the exposure probability estimate that showed the highest predictive value. Index subjects with missing information for this estimate were classified by the exposure probability estimate ranked with the second highest predictive value, and so forth. By this strategy an exposure probability estimate was available for all but three of the 1522 index subjects. Only one exposure probability estimate was obtained for an individual. We computed an average exposure probability estimate for the 142 workers employed in more than one company (with possibly different exposure probabilities).
We also validated this final exposure probability classification among the 427 living index subjects. In this validation we used the original three levels of exposure probability (0%, 1–49% and 50–100%) and assessed the ability to classify correctly an index subject as laminator (i) or laminator or bystander exposed production worker (ii + iii).
Statistical methods
Styrene exposure level
Short (<1 h) and long (1–8 h) term measurements were examined separately using statistical analyses. The reason for such segregation of the data was that measurements that lasted <1 h were more likely to cover only one task (lamination) where the highest exposures were expected, whilst the measurements that spanned longer time periods were likely to encompass several tasks, and thus are more representative of exposure concentrations averaged over the full duration of a work-shift.
The frequency distribution of exposure levels was examined using frequency histograms and normal probability plots to determine whether logarithmic transformation was warranted prior to application of parametric statistical models. Right skew in frequency distribution was taken as evidence that log-normal frequency distribution provided a better fit to the data. Exposure levels were summarised in terms of number of measurements, number of non-detectable measurements, geometric and arithmetic means, corresponding standard deviations, and ranges.
Mixed effects models were used to identify determinants of styrene exposure. These models considered exposure concentrations as the dependent variable, and company production characteristics (product, production process and calendar year period) and sampling characteristics, as predictor variables. All predictor variables were a priori expected to affect exposure levels and were included in the models. The models had the following general form:
 | (1) |
l is the natural logarithm of the exposure concentration measured on the jth day of the ith company in lth time period, belonging to a group defined by the predictor variables ß1,...,ßn; µ is the true mean of log-transformed exposure averaged over all groups; ß1,...,ßn are the predictor variables not dependent on time (fixed effects); l is the fixed effect of the lth time period; 
l(i) is the random effect of the ith company in lth time period; 
l(ij) is the random effect of the jth day in ith company during lth time period.
The model assumes that for each time period (l), l(i) and l(ij) are normally distributed with zero means and variances 
and 
, respectively, which are mutually independent. These variances are estimated as between-company (
) and sample-to-sample within-company (
) variance components for lth time period. Simpler models were also considered in which we assumed that 
and 
were homogeneous across all time periods l, but such models had considerably worse fit to the data, as judged by likelihood ratio tests (Weaver et al., 2001), and therefore were not considered further. Plotting standardised residuals against predicted values tested assumptions of the mixed effects models.
Mixed effects models were evaluated using PROC MIXED of SAS version 6.12 (SAS Institute, Cary, NC), employing restricted maximum likelihood algorithm.
Styrene exposure scores
Two sets of semi-quantitative exposure scores for styrene were calculated for each index subject included in the epidemiological studies by multiplying either the short-term or the long-term styrene level estimated in the mixed models by the positive predictive values obtained for the final exposure probability classification in the validation study in a manner analogous to that described by Burstyn et al. (2003). Briefly, this two-stage procedure is as follows:
- Calculate Xl(k) from the model built using equation (1), which represents the median value of the long-term means of individual exposures of a group of workers (Tornero-Velez et al., 1997) in a company that experienced kth exposure scenario, Sl(k), in lth time period [defined by a combination of {ß1,...,ßn, 1}: e.g. hand lamination in boat making during 1975–1979 period, standardised to long-term samples (thus, ‘scenario’ is a situation in a company that is described by a combination of fixed effects)]:
where Yl(k) and
(2) were estimated directly from equation (1).
- Estimate the median of mean exposure (Ml) for a worker in a company that experienced k exposure scenarios (Sl(1),S1(2),...Sl(k)...,Sl(K)) in lth time period according to the following formula:
where P(Sl(k)) was the probability of kth scenario during lth time period, such that
(3) 
P(Sl(k)) = 1. Probabilities of exposure scenarios for index subjects were the positive predictive values obtained for the final exposure probability classification as shown in Table 6. Thus, in this paper we simply multiplied Xl(k) from the first year of exposure (based on job histories) by the exposure probability estimate. [When computing cumulative exposure (mg/m3 years), we took into account changes in exposure level by calendar year (by each 5-year period). In this computation, we did not consider possible changes in product or process for the workers who were employed in more than one company. The cumulative exposure scores are not presented in the paper.]
|
In our data, no long-term samples existed prior to 1970 and we used a WCS2-value of 0.04 (based on subjective judgment of Table 4) in calculation of exposure scores.
|
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
The companies produced boats, containers, windmill wings, panels, etc. (Table 1). Some companies did not produce reinforced plastics but, for example, metal products or thermoplastics. Hand or spray lamination was the preferred production process and few companies used other processes (e.g. pultrusion or bulk mold compound). An average of 16.0 styrene measurements (range 2–447) were recorded in 191 companies with at least two repeated measurements; exposure measurements from these companies were retained for modelling of the determinants of styrene exposure level. The overall geometric mean of styrene level was 137.5 mg/m3 for short-term samples [geometric standard deviation (GSD), 3.4] and 98.5 mg/m3 for long-term samples (GSD, 3.2) as seen in Table 2 that also presents average styrene exposure levels by product, production process and calendar year for short and long-term measurements respectively.
|
|
Production characteristics determining styrene exposure level
Preliminary examination (by the Duncan multiple range test) had shown that the production of large and small boats were associated with comparable styrene exposure levels. Consequently, product was classified as boats versus other products in the multivariate analysis. Likewise, hand and spray lamination production processes showed comparable styrene exposure levels in preliminary analyses, and process was classified as lamination versus other processes. According to the mixed effects models, styrene exposure levels were elevated by a factor of 1.6 [exp(0.46)] to 1.7 [exp(0.55)] in companies producing boats (Table 3). Hand and spray lamination was also associated with an average factor of 1.5–1.8 increase in styrene exposure. Calendar year was a strong and consistent predictor of the styrene exposure level. Between the 1960s and 1990s, levels of styrene exposure declined by 7% annually (from models with year as a continuous variable, data not shown). The relative influence of product, process and calendar year period on the styrene exposure levels was comparable for short- and long-term samples. Long sampling duration was associated with overall lower exposure concentrations. Worst-case sampling tended to be linked to elevated exposure levels only in long-term samples. Graphs of residuals indicated that assumptions of normality and homoscedacity of residuals were not violated.
|
Styrene exposure variability
Table 4 presents the estimates of the variance components for short-term and long-term samples corrected for production and sampling method characteristics. The within-company variability was higher than the between-company variability for short-term samples and increased by calendar year period for both short-term and long-term samples. The between-company variability showed no consistent pattern by calendar year. A model that included neither time trend (calendar year period) nor production characteristics (process and product), and assumed homogeneity of covariance structure between time periods showed a between-company variance component of 0.97 for short-term samples. Thus, it appears that the combination of time trend and production characteristics explain between 94% [(0.97–0.06)/0.97] and 17% [(0.97–0.81)/0.97] of the between-company variance of short-term samples, depending on time period. For long-term samples the between-company variance component of a model without time trend and production characteristics was 0.66. The complete model then explained between 6 and 85% of this variability.
Styrene exposure probability
Table 5 provides information on sensitivity, specificity and positive predictive value for the five different exposure probability estimates. The exposure probability estimate based on the colleagues' report on their own work as laminator (Colleagues I) showed the highest specificity (72%). Low specificity was seen for the other estimates. The highest sensitivity was seen for the dealers' (Dealers) and the employers' estimate based on the proportion of company working hours spent on lamination (Employers II) (92%).
|
The colleagues' report on their own work as laminators (Colleagues I) showed the highest predictive value of 37%, while the Employers II estimate showed the lowest value of 20%.
Table 5 also presents the proportion of missing estimates among the 427 index subjects included in the validation study. The lowest value was seen for the Dealers estimate (7%). The table also presents the cumulative contribution of the individual exposure probability estimates to the final exposure probability classification that provided information for all but three of the 1522 index subjects.
Table 6 shows that the final exposure probability classification correctly categorised up to 53% of the laminators and up to 67% of workers directly or indirectly exposed to styrene (laminators and bystander exposed production workers). The final classification was unrelated to styrene exposure level when assessed in a multiple linear regression model for short-term samples (estimate –0.02, P-value 0.84) and long-term samples (estimate –0.05, P-value 0.65) but showed an inverse relation with the within-company variance components of short-term measurements (WCS2: 0% 1.12; 1–49% 0.69; 50–100% 0.57, data not shown), but no such trend was seen for long-term measurements.
Styrene exposure scores
The styrene exposure scores computed for 1519 of the index subjects based on the short-term and the long-term measurements, respectively, are presented in Figs 1 and 2. Average styrene exposure scores declined about 10-fold from the 1960s to the 1990s reflecting a decline in styrene exposure levels. However, the proportion of workers potentially exposed to styrene (an exposure probability of 1% or more) increased by calendar year from 36% in the 1960s to 46% in the 1990s, partially counteracting the effect of declining styrene exposure levels.
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Styrene exposure levels were higher when measured by short-term samples as compared with measurements by long-term samples. This probably reflects that long-term samples more often included low-exposure tasks occurring during a shift than short-term samples. Long-term measurements therefore better reflected full-shift average exposure and should be used for classifying workers in epidemiological studies of health effects following cumulative styrene exposure. If the interest is in the health effects of peak styrene exposure, short-term measurements may be a better choice.
Calendar year, product and process determined short-term and long-term styrene exposure level. These findings are in agreement with several earlier reports on styrene exposure in the reinforced plastics industry (Lemasters et al., 1985; Kogevinas et al., 1994; Crandall and Hartle, 1985; Nylander-French et al., 1999; Lenvik et al., 1999). The 7% annual decline in styrene level is within the range of 4–14% reduction observed in a systematic review of long-term trends of occupational exposures (Symanski et al., 1998, 2000).
The within-company variability increased by calendar year and it was necessary to specify mixed models that took account of this. The trend in within-company variability may be explained by increasing diversity of tasks within the companies, which was not captured by main product and process in the models. One may speculate that such diversity was related to the proportion of laminators. This was supported by the observation that within-company variability of short-term measurements decreased, as the proportion of laminators rose, possibly in response to the laminators employed at the same time performing tasks associated with different exposures. The trend in within-company variability may also be related to changes in sampling strategy not captured by the inclusion of a sampling strategy variable in the models, since the measurements were not taken at random. Variability in analytical methods was a less likely cause, since it was expected that analytical methods became more standardised during later years. However we did not have access to information that would have made it possible to evaluate these considerations further.
Heterogeneity in sources of exposure variability is an emerging issue in occupational exposure assessment (Symanski et al., 2001a), and it remains to be seen if the patterns we observed hold true for other industries. The valid estimate of variance components is essential, since it was used in the calculation of the median of between-company arithmetic means of styrene exposure scores (Tornero-Velez et al., 1997; Burstyn et al., 2003).
Measured companies were not sampled at random and measurements were not sampled at random within the companies. They probably included more worst-case situations, because they were obtained to control compliance with the exposure limits. This possibly biased styrene exposure level estimates towards higher values, even though we did attempt to standardise predictions of exposure scores to samples collected during exposure surveillance programs (as opposed to worst-case sampling by the labour inspectors).
We disregarded dermal exposure to styrene, even if dermal exposure may be considerable (Eriksson and Wiklund, 2004), since dermal absorption of styrene vapour or liquid is low compared with inhalation exposure (Berode et al., 1985).
Within the reinforced plastics industry, laminators and lamination tasks are known to be significant determinants of styrene exposure level (Crandall and Hartle, 1985) and levels of urinary styrene metabolites (Kolstad et al., 1999; Symanski et al., 2001b). However, in this study we were unable to take into account job, task or other individual factors and thus were unable to make a distinction between directly exposed workers (laminators), bystander-exposed workers and non-exposed workers, with expected considerable contrast in styrene exposure level. As an alternative, we defined groups of workers (by classifying them by the proportion of laminators) according to comparable styrene exposure probabilities. The estimates of the dealers were significantly easier to obtain than the others, and furthermore they were provided for the highest number of companies. However, the estimates based on the colleagues' reports on their own work showed the highest specificities and predictive values. This probably reflects the fact that these colleagues were individually matched with the index workers by calendar year of employment, duration of employment and sex, which were in turn associated with the job in this industry. Thus, it seems advisable in other similar studies to use matched colleagues' reports of their own work in the estimation of exposure probability for index subjects.
The final exposure probability classification took account of the positive predictive value and thus the magnitude of misclassification of the individual estimates (Olsen, 1988). For companies with a high exposure probability according to this classification, 53% of the workers were correctly classified as laminators (the predictive value) and 67% as laminators or bystander exposed production workers. This relatively low degree of misclassification probably reflects that 50% of the reinforced plastics companies employed less than five workers and the simple exposure conditions dominated by one airborne contaminant: styrene.
A major limitation of this exposure assessment strategy, is still the misclassification of index subjects with respect to exposure status. This inevitably will lead to bias towards the null hypothesis in epidemiological studies of styrene related health effects. Since the exposure assessment was independent of reports from the index subjects', epidemiological analyses will, on the other hand, not be flawed by differential misclassification.
The styrene exposure score was the product of estimated exposure probability and exposure level, which were not mutually related, and putting them together will dilute the information obtained from each of them. Therefore, in the epidemiological study, subjects should also be classified separately by exposure probability and exposure level, or by a cross tabulation of these two determinants of exposure. Still, the exposure score provides one aggregated exposure estimate, which is appealing because it provides a parsimonious presentation of the data.
The styrene exposure score only provides an indirect measure of individual styrene exposure intensity. However, we believe that, when combined with individual employment histories (e.g. duration of employment in the index companies) it will be a sound basis for valid semi-quantitative grouping of styrene-exposed workers in future epidemiological risk assessment.
Such model-based exposure assessment is analogous to group-level exposure assessment (Wameling et al., 2000), which can be expected to yield unbiased relative risk estimates at the expense of inflated variance of risk parameter (Preller et al., 1995; Armstrong, 1998; Tielemans et al., 1998). This is so, because grouping leads to a predominance of Berkson-type error in exposure assessment (Armstrong, 1998). However, caution is advised when combining predictions of regression models with estimates of exposure probability, since errors in estimates of exposure probability can negate any gains made due to modelling of exposure intensity (Wameling et al., 2000). It is for this reason that we have invested a considerable amount of effort in obtaining valid exposure probability estimates, and advise other researchers in the field to pay an equal amount of attention to validating exposure intensity and probability estimates.
Another strength of our approach lies in the use of inhalation exposure to classify subjects for the epidemiological study, which was shown to be superior to the use of biological monitoring in a similar industry (Liljelind et al., 2003).
In conclusion, we developed a semi-quantitative exposure classification of workers for whom no individual exposure information was available. Such exposure classification is often needed because retrospective individual exposure information may be difficult to obtain and is subject to recall bias. This exposure assessment approach may be justified in other industries, and especially in industries dominated by small companies with simple exposure conditions, like the Danish reinforced plastics industry. It should be noted that large-scale epidemiological studies of such small-scale industries could be expected to dominate occupational epidemiology in the future (Blair et al., 1999).
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
This study was supported by a grant from the Danish Working Environment Fund (1997–1942).
Received February 27, 2004; in final form August 10, 2004
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Armstrong BG. (1998) Effect of measurement error on epidemiological studies of environmental and occupational exposures. Occup Environ Med; 55: 651–6.[Abstract]
Berode M, Droz PO, Guillemin M. (1985) Human exposure to styrene. VI. Percutaneous absorption in human volunteers. Int Arch Occup Environ Health; 55: 331–6.[CrossRef][ISI][Medline]
Blair A, Rothman N, Zahm SH. (1999) Occupational cancer epidemiology in the coming decades. Scand J Work Environ Health; 25: 491–7.[ISI][Medline]
Boyle CA, Brann EA. (1992) Proxy respondents and the validity of occupational and other exposure data. The Selected Cancers Cooperative Study Group. Am J Epidemiol; 136: 712–21.
Burstyn I, Boffetta P, Kauppinen T et al. (2003) Estimating exposures in the asphalt industry for an international epidemiological cohort study of cancer risk. Am J Ind Med; 43: 3–17.[CrossRef][ISI][Medline]
Crandall MS, Hartle, RW. (1985) An analysis of exposure to styrene in the reinforced plastic boat-making industry. Am J Ind Med; 8: 183–92.[ISI][Medline]
Eriksson K, Wiklund L. (2004) Dermal exposure to styrene in the fibreglass reinforced plastics industry. Ann Occup Hyg; 48: 203–8.
Hansen J, Boffetta P, Andersen A et al. (1997) Comparison of information on occupation and lifestyle habits obtained from European man-made vitreous fibre production workers and their relatives. Int J Epidemiol; 26: 1009–16.
Jensen AA, Breum NO, Bacher J et al. (1990) Occupational exposures to styrene in Denmark 1955–88. Am J Ind Med; 17: 593–606.[ISI][Medline]
Kogevinas M, Ferro G, Andersen A et al. (1994) Cancer mortality in a historical cohort study of workers exposed to styrene. Scand J Work Environ Health; 20: 251–61.[ISI][Medline]
Kolstad HA, Pedersen B, Olsen J et al. (1996) Clonal chromosome aberrations in myeloid leukemia after styrene exposure. Scand J Work Environ Health; 22: 58–61.[ISI][Medline]
Kolstad HA, Bonde JP, Spano M et al. (1999) Change in semen quality and sperm chromatin structure following occupational styrene exposure. ASCLEPIOS. Int Arch Occup Environ Health; 72: 135–41.[CrossRef][ISI][Medline]
Kolstad HA, Juel K, Olsen J et al. (1995) Exposure to styrene and chronic health effects: mortality and incidence of solid cancers in the Danish reinforced plastics industry. Occup Environ Med; 52: 320–7.[Abstract]
Kolstad HA, Lynge E, Olsen J. (1993) Cancer incidence in the Danish reinforced plastics industry. IARC Sci Publ; 301–8.
Kolstad HA, Lynge E, Olsen J et al. (1994) Incidence of lymphohematopoietic malignancies among styrene-exposed workers of the reinforced plastics industry. Scand J Work Environ Health; 20: 272–8.[ISI][Medline]
Lemasters GK, Carson A, Samuels SJ. (1985) Occupational styrene exposure for twelve product categories in the reinforced-plastics industry. Am Ind Hyg Assoc J; 46: 434–41.[ISI][Medline]
Lenvik K, Osvoll PO, Woldbaek T. (1999) Occupational exposure to styrene in Norway, 1972–1996. Appl Occup Environ Hyg; 14: 165–70.[CrossRef][Medline]
Liljelind I, Rappaport S, Eriksson K et al. (2003) Exposure assessment of monoterpenes and styrene: a comparison of air sampling and biomonitoring. Occup Environ Med; 60: 599–603.
Nylander-French LA, Kupper LL, Rappaport SM. (1999) An investigation of factors contributing to styrene and styrene-7,8-oxide exposures in the reinforced-plastics industry. Ann Occup Hyg; 43: 99–105.
Olsen J. (1988) Limitations in the use of job exposure matrices. Scand J Soc Med; 16: 205–8.[ISI][Medline]
Preller L, Kromhout H, Heederik D et al. (1995) Modeling long-term average exposure in occupational exposure–response analysis. Scand J Work Environ Health; 21: 504–12.[ISI][Medline]
Seixas NS, Checkoway, H. (1995) Exposure assessment in industry specific retrospective occupational epidemiology studies. Occup Environ Med; 52: 625–33.[Abstract]
Symanski E, Bergamaschi E, Mutti A. (2001b) Inter- and intra-individual sources of variation in levels of urinary styrene metabolites. Int Arch Occup Environ Health; 74: 336–44.[CrossRef][ISI][Medline]
Symanski E, Chang CC, Chan W. (2000) Long-term trends in exposures to nickel aerosols. Ann Ind Hyg Assoc J; 61: 324–33.
Symanski E, Kupper LL, Rappaport SM. (1998) Comprehensive evaluation of long-term trends in occupational exposure: Part 1. Description of the database. Occup Environ Med; 55: 300–9.[Abstract]
Symanski E, Sallsten G, Chan W et al. (2001a) Heterogeneity in sources of exposure variability among groups of workers exposed to inorganic mercury. Ann Occup Hyg; 45: 677–87.
Tielemans E, Kupper LL, Kromhout H et al. (1998) Individual-based and group-based occupational exposure assessment: some equations to evaluate different strategies. Ann Occup Hyg; 42: 115–9.
Tornero-Velez R, Symanski E, Kromhout H et al. (1997) Compliance versus risk in assessing occupational exposures. Risk Anal; 17: 279–92.[CrossRef][ISI][Medline]
Wameling A, Schaper M, Kunert J et al. (2000) Individual toluene exposure in rotary printing: increasing accuracy of estimation by linear models based on protocols of daily activity and other measures. Biometrics; 56: 1218–21.[CrossRef][ISI][Medline]
Weaver MA, Kupper LL, Taylor D et al. (2001) Simultaneous assessment of occupational exposures from multiple worker groups. Ann Occup Hyg; 45: 525–42.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. G. M. Van Rooij, A. Kasper, G. Triebig, P. Werner, F. J. Jongeneelen, and H. Kromhout Trends in Occupational Exposure to Styrene in the European Glass Fibre-Reinforced Plastics Industry Ann. Hyg., July 1, 2008; 52(5): 337 - 349. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. LAVOUE, D. BEGIN, C. BEAUDRY, and M. GERIN Monte Carlo Simulation to Reconstruct Formaldehyde Exposure Levels from Summary Parameters Reported in the Literature Ann. Hyg., March 1, 2007; 51(2): 161 - 172. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B Serdar, R Tornero-Velez, D Echeverria, L A Nylander-French, L L Kupper, and S M Rappaport Predictors of occupational exposure to styrene and styrene-7,8-oxide in the reinforced plastics industry Occup. Environ. Med., October 1, 2006; 63(10): 707 - 712. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I. BURSTYN, H-M. KIM, N. CHERRY, and Y. YASUI Metamodels of bias in Cox proportional-hazards and logistic regressions with heteroscedastic measurement error under group-level exposure assessment Ann. Hyg., April 1, 2006; 50(3): 271 - 279. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||