Annals of Occupational Hygiene Advance Access originally published online on August 26, 2005
Annals of Occupational Hygiene 2006 50(1):55-64; doi:10.1093/annhyg/mei041
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© 2005 British Occupational Hygiene Society Published by Oxford University Press
Original Article |
Are Variance Components of Exposure Heterogeneous Between Time Periods and Factories in the European Carbon Black Industry?
1 Centre for Occupational and Environmental Health, University of Manchester, UK; 2 Institute of Occupational Health, University of Birmingham, UK; 3 Department of Public Health Sciences, University of Alberta, Edmonton, Canada; 4 Institute for Risk Assessment Sciences, Utrecht University, The Netherlands; 5 The Medical School, The University of the Witwatersrand, Johannesburg, South Africa
* Author to whom correspondence should be addressed. Tel: +44-(0)161-275-8500; fax: +44-(0)161-275-5595; e-mail: martie.van-tongeren{at}manchester.ac.uk
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Occupational exposure to chemical agents can vary enormously within- and between-workers, even when carrying out the same jobs. When repeated measurements are available, the variance components can be estimated using random- or mixed-effects models. Pooling the variance components across the fixed effects, in mixed-effects models, reduces the complexity of the models; especially, when there are a large number of fixed effects. The analyses presented in this paper tested the assumptions of homogeneity in the variance components between factories and surveys for inhalable dust exposure in the European carbon black manufacturing industry. In total, 5296 measurements from 1771 workers were available collected during two surveys carried out between 1991 and 1995. Workers were grouped into eight job categories, and for each of these separate mixed-effects models were developed, including factory, survey and in some cases the interaction term as the fixed effects. The likelihood ratio test was used to test the assumptions of homogeneity of the variance components. Statistically significant heterogeneity of the variance components was observed for two of the eight job categories, ‘Fitter/Welder’ and ‘Warehouseman’. The heterogeneity was due mainly to differences in variance between the factories. When estimating the probability of overexposure for all the factories combined, there was little difference between the models with and without heterogeneous variance components for ‘Fitters/Welders’. For the ‘Warehousemen’ the probability of overexposure in the last survey changed marginally from 4% in the pooled model to 6% in the heterogeneous model. Larger differences between the models were observed when estimating the probability of overexposure for individual factories, which was due to over- or under-estimation of the variance components in the pooled models. In conclusion, for most job categories pooling of the variance components appears to be justified in this database. In addition, no large differences were found when determining the industry-wide probability of ‘overexposure’ when comparing the pooled with the heterogeneous models. However, when evaluating the factory-specific probability of ‘overexposure’ or when using the models to provide exposure estimates for epidemiological studies heterogeneity in the variance components should be investigated.
Keywords: variance compounds • exposure modelling • mixed effects models
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
In the occupational environment exposure concentrations (intensities) vary enormously. Workers within the same department performing the same tasks and sharing the same working environment can have considerable differences in average exposure levels (between-worker variance) and definitely experience varying exposure concentrations from day-to-day (within-worker or day-to-day variance) (Rappaport, 1991
; Kromhout et al., 1993; Rappaport et al., 1993; Kromhout and Vermeulen, 2001). The actual magnitude of these variance components can be estimated from log-transformed exposure concentrations using a random-effects ANOVA (ANalysis Of VAriance) model when the same workers are sampled on more than one occasion. In this model, exposure is expressed by the overall geometric mean of the group, with the variability in exposure expressed as within- (or day-to-day) and between-worker variances. The latter means that the individual workers are allowed to have distinct personal mean exposures. Kromhout et al. (1993) estimated that the variation in results of 8 h shift long measurements of airborne exposure could be as large as 3–4000 fold. Generally, within a group of workers in the same location, the day-to-day component of exposure variability is larger than the between-worker differences in average exposure.
The estimates of central tendency of the group (geometric mean) and variance components (within- and between-worker) can subsequently be used to estimate the probability of ‘overexposure’ (chance that a randomly selected worker's mean exposure is greater than the regulatory occupational exposure standard) (Tornero-Velez et al., 1997) or can be used to determine estimates of exposure in epidemiological studies (Burstyn et al., 2000). When using mixed-effects models, fixed effects can be introduced to take into account any differences in the mean exposure levels between factories or job categories, while simultaneously adjusting these mean differences for random between- and within-worker variability in exposure intensity (Schlunssen et al., 2004).
Rappaport et al. (1999) showed in a study of particulate exposure amongst construction workers that the between-worker variance components can be significantly different between jobs; although, the results suggested that the within-worker variance components could be pooled across the jobs. In contrast, Symanski et al. (2001) found no significant heterogeneity in the variance over time and between workers amongst groups of workers exposed to inorganic mercury. Estimating the variance components for each level of the fixed effects will increase the complexity and computer time required for estimating the parameters for the models. Nonetheless, if there are significant differences in the covariance structure between groups defined by fixed effects, the advantage of using one model to describe all the exposure data across an industry will be lost.
In a previous paper, trends over time in personal inhalable dust exposure in the European carbon black manufacturing industry were investigated (van Tongeren et al., 2000). Carbon black is a very fine powdered form of elemental carbon that is produced predominantly in a furnace by the controlled vapour phase pyrolysis of liquid or gaseous hydrocarbons (Gardiner et al., 1992). After cooling, carbon black is separated from the production gases in bag filters. Subsequently, the fine carbon black is generally processed to produce pelletized carbon black, which is packed in large bags or containers, or stored in silos for bulk transport. Carbon black is used mostly in the production of tyres (Gardiner, 1995; IARC, 1996).
For the analyses presented by van Tongeren et al. (2000), it was assumed that the variance components were constant across time and factory, but not across the various job categories (as separate models were developed for each job category). This paper will investigate if this assumption was valid or if there was any evidence that variance components differed among factories and surveys. Finally, we will investigate if heterogeneity in the variance components alters conclusions about estimated probabilities of overexposure.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
In total, 8015 inhalable dust measurements were carried out in 19 factories in seven countries (UK, France, Germany, The Netherlands, Spain, Italy and Sweden) during three surveys between 1987 and 1995. All workers were classified into one of eight job categories (Table 1) and from each category a stratified random sample of workers was selected for monitoring. In the second and third surveys, repeated measurements were taken on the same workers. Personal inhalable dust exposure measurements were carried out using the IOM sampling head (Mark and Vincent, 1986
) with a flow-rate of 2.0 l min–1. For further details of the sampling and analytical methods see Gardiner et al. (1992, 1996) and van Tongeren et al. (2000). The limit of detection for inhalable dust was determined at 0.025 mg per volume of air (m3) (van Tongeren et al., 1994); values below the limit of detection were replaced by a value of 2/3 x 0.025 mg per volume of air. For the analyses presented here, data from the first survey (1987–1989) (n = 1316) were excluded because no repeated measurements on the same individual were collected during this period. In addition, data from three factories were excluded as these were closed down prior to or during the final survey (n = 590). Finally, when no repeated measurements were available for a job category in a factory in either of the two surveys, all data for this particular job category in this particular factory were also excluded (n = 863).
|
For each job category, separate mixed-effects models were developed using Proc Mixed in SAS for Windows (Release 8.02). Random (worker identity and residual error) and fixed effects (survey, factory and interaction term) were estimated, using the restricted maximum likelihood algorithm. The following model was used to describe the data for each job category separately:
 | (1) |
g represents the fixed effect of the gth survey; ßh represents the fixed effect of the hth factory; 
gh(i) represents the random effect of the ith worker; and 
gh(ij) represents the random within-worker variation. It is assumed that 
, 
and 
; that gh(i) and gh(ij) are normally distributed with zero means, and variances of 
and 
, respectively, representing the between- and within-worker variance components (estimated for each survey and factory separately); and that the random worker gh(i) and residual error (gh(ij)) effects are statistically independent. It was assumed that any two measurements on the same worker have equal correlation irrespective of the time interval between the measurements, whilst measurements on different workers are uncorrelated (compound symmetry covariance structure) (Peretz et al., 2002). An interaction term for survey and factory was included in the model ([ß]gh) only if this significantly improved the model (using P < 0.05).
The description of the mixed-effects model in equation (1) estimates the variance components for each combination of survey and factory separately ( and ). Several other reduced models were also fitted, including models with pooled variance components (
and 
) across the fixed-effects, and models with heterogeneity in the between-worker variance for survey (
) or for factory (
). Likelihood ratio tests were used to determine which model had the best fit, using the model described in equation (1) as a reference. The degrees of freedom (df) for the likelihood ratio tests were determined by the difference in the number of variance parameters [i.e. when estimating distinct between-worker variance components for the two surveys but with pooled within-worker variance components (three variance estimates), the df when comparing this model with the one with pooled variance components (2 variance parameters) is 1].
The probability of ‘overexposure’ (
gh), the chance that a randomly sampled worker's long-term mean exposure exceeds the UK Workplace Exposure Limit (WEL) of 3.5 mg m–3 (HSE, 2005), was calculated using the following equation (Tornero-Velez et al., 1997):
 | (2) |
{t} denotes the probability that a standard normal variate falls below the value of t (Rappaport et al., 1995; Lyles et al., 1997; Tornero-Velez et al., 1997). Here, µx,gh represents the mean exposure of the ith worker in the gth survey and the hth factory; 
, with µy,gh(i) = µy,gh + gh(i) and µy,gh = µy + g + ßh + [ß]gh. Variance components specific for a given factory and survey are represented by and . In case the estimates of the variance component are pooled across factory and survey, and are replaced by and . |
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
After all exclusions described in the methods section, 5296 inhalable dust measurements from 1771 workers were available for the analyses (which equates to 66% of the total of 8015 measurements). Table 1 provides the number of measurements, median exposure and the range in exposure by job category and survey. The overall median exposure was reduced from 0.4 mg m–3 in the second to 0.3 mg m–3 in the third survey. In each job category the median exposure was reduced, although differences were small.
Table 2 shows the –2 log-likelihood estimates for the various models with different covariance structures by job category. Factory and survey were included as fixed effects in the models for all job categories, whilst for the job categories ‘Laboratory Assistant/Control Room Operator’, ‘Process Foreman/Furnace Operator’, ‘Fitter/Welder’, ‘Warehouseman’ and ‘Site Crew/Cleaner’ the interaction term between factory and survey was also included, indicating that the change in exposure between the two surveys varied between the factories. (As a number of workers participated in both surveys, the total number of workers with measurements as indicated in Table 2, is not the same as sum of the number of workers in the 1st and the 2nd survey as given in Table 1).
|
For a number of models, the model parameters could not be estimated owing to infinite likelihood or too many likelihood estimations. Generally, this was the case when attempting to fit the full model including heterogeneous within- and between-worker variance components for factory or for a combination of survey and factory. Only for the job categories ‘Fitter/Welder’ and ‘Warehouseman’ was it possible to estimate the full model with the heterogeneous within- and between-worker variance components for both survey and factory in the same model.
Only for the job categories ‘Fitter/Welder’ and ‘Warehouseman’ was there clear evidence that reducing the model, by pooling the variance components significantly reduced the fit of the model. For both job categories the most appropriate model appeared to include heterogeneous within- and between-worker variance components for factory. When comparing this model to the fully reduced model the likelihood ratio test statistic (
2) for ‘Welders/Fitters’ was 1938.5 – 1889.6 = 48.9 with 30 – 2 = 28 df (P = 0.0086), whilst for the ‘Warehousemen’ the 2 was 2315.8 – 2244.5 = 71.3 with 50 – 26 = 24 df (P < 0.0001). Including heterogeneity in the variance components for the survey, did not improve the model further for the ‘Welders/Fitters’ (2 = 1889.6 – 1875.7 = 13.9, with 30 df; P = 0.9946) nor for the ‘Warehousemen (2 = 2244.5 – 2222.6 = 21.9, 26 df; P = 0.6942).
Next, the probability of overexposure was estimated for the ’Fitters/Welders' and ‘Warehousemen’ using the reduced models and the models including heterogeneous variance components for factory to determine the sensitivity of the results to the specification of the models (Table 3). Table 3 provides the (range of) within- and between-worker variance, and the range in the probabilities of overexposure for the factories. The number of workers with mean exposures expected to exceed 3.5 mg m–3 was calculated by multiplying the probability of overexposure by the total number of workers in the job categories ‘Fitters/Welders’ and ‘Warehousemen’ during each survey. For ‘Fitters/Welders’ the estimated probability of overexposure was low for both models and for both surveys, and the choice of model would not have changed the overall conclusions. For the ‘Warehousemen’ the probability of overexposure in the second survey was estimated to be 7% when using the pooled model, and 8% when using the model with heterogeneous variance components. In the last survey this changed from 
4% (11 out of 285 workers) for the model with pooled variance components to 6% (16/285) in the model with heterogeneous variance components.
|
When comparing the probabilities of overexposure for the individual factories, considerable differences were observed (Figs 1 and 2). For ‘Fitters/Welders’, the main differences occurred in factory 15 for Survey II and in factory 3 for Survey III (Fig. 1). For factory 15, the probability of overexposure in Survey II changed from 13% in the model with pooled variance components to 0% in the model with heterogeneous variance. This difference was due to an overestimation of the between-worker variance in the pooled analyses (
: 0.32) compared to the analyses with heterogeneous variance components (
: 0.01). For factory 3, the probability of overexposure was increased in Survey III from 6 to 13% when allowing for heterogeneous variance components, due to underestimation of the within-worker variance in the fully reduced model (
: 0.90 versus 
: 1.37).
|
|
For the ‘Warehousemen’, the estimated probabilities of overexposure increased for factories 1, 3 and 6, whilst for other factories, such as factories 9, 16 and 18, the estimated probabilities were reduced when comparing the models with heterogeneous variance components with the fully reduced model. In the fully reduced model, the within-worker variance was underestimated in factory 1 and 6 (
: 0.96 compared to : 1.31 and 1.22, respectively) and overestimated in factories 9 and 18 (: 0.48 and 0.03, respectively), whilst the between-worker variance was underestimated for factory 3 (: 0.24 versus : 0.60) and overestimated for factories 9, 16, 17 and 18 (: 0.08, 0.00, 0.03 and 0.12, respectively). |
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Increasingly, mixed-effects models are used to investigate trends in and determinants of occupational exposure (see for example Symanski et al., 1996
; Nylander-French et al., 1999; Rappaport et al., 1999; Burstyn et al., 2000; van Tongeren et al., 2000; Vermeulen et al., 2000; Burstyn and Kromhout, 2002; Peretz et al., 2002) as this type of model allows for the estimations of fixed effects and random effects simultaneously. Failure to model random effects correctly has been shown to lead to erroneous conclusions about the importance of various determinants of exposure (Peretz et al., 2002) and can affect conclusions about compliance with regulatory exposure limits (Rappaport et al., 1999). Typically, the within- and between-worker variances are pooled across the fixed effect. Part of the reason for this is that the models become much more complex when allowing for heterogeneous variance components across all the fixed effects, requiring large datasets as well as considerable amount of computing time on standard desktop computers. In addition, increasing the complexity of the models may also lead to lack of convergence. The advantage of pooling the variance components (and fixed effects) is that this will lead to more precise estimates of the model parameters. This is important because most datasets in occupational hygiene have only a very limited number of repeated measurements per workers that can be used to estimate within-worker variance. In this study, the average number of repeated measurements varied from 1.2 for ‘Administrative Staff’ to 2.9 for ‘Instrument Mechanic/Electrician’, both in the second survey. However, pooling must be justified lest it leads to erroneous conclusions due to failure to account for important differences in covariance structure among groups of workers defined by fixed effects. In a sense, the requirement to form monomorphic (stationary) exposure groups (i.e. use of the model that fits the data) in assessment of overexposure (Rappaport et al., 1995) must be satisfied in application of mixed-effects models to exposure data from large studies with participation of multiple companies. The analyses presented in this paper showed that there was evidence in two job categories (‘Fitter/Welder’ and ‘Warehouseman’) that the variance components were heterogeneous across the factories but not over time (i.e. not across survey). It was assumed that the variance components would be different between the different job categories due to the different exposure levels and patterns, and therefore, separate models were fitted for each job category.
The reasons why different groups of workers have distinct variance components are only partly understood. Kromhout et al. (1993) investigated the determinants of within- and between-worker variance components in a large database of occupational exposure to various chemicals. They found that the within-worker (or day-to-day) variance was greater in groups working outdoors and those working without local exhaust ventilation. Mobile workers, workers involved in intermittent processes and working with local or mobile sources also appeared to have large day-to-day variance in exposure. For the between-worker variance only type of process (whether continuous or intermittent) had a significant effect. Major changes in the process and environmental conditions, working patterns and tasks can therefore lead to changes in both the intensity and within- and between-worker variance. As there was only a 2–3 year interval between the second and third survey, there was limited opportunity for such major changes.
Differences in the variance components between the factories could be due to differences in the level of outdoor work (‘Fitters/Welders’), mobility of workers (‘Fitters/Welders’ and ‘Warehousemen’), and use and effectiveness of local exhaust ventilation systems (‘Warehousemen’). However, we have no actual data on these exposure-variability-affecting factors to substantiate this.
When using the parameters of the models to estimate the probability (across all the factories) that the mean individual exposure exceeds an exposure limit of 3.5 mg m–3, we found only minor differences between the models. Thus, just like Symanksi et al. (2001), we have observed that, in general, pooling of the data for estimating of variance components appeared to be justified. However, when looking at the results for the factories separately, we observed some large differences between the models, which were caused by the over- or underestimation of within- and/or between-worker variance when using the pooled variance estimates.
Unfortunately, it was not possible to fit all the models due to issues such as infinite likelihood estimates. If the models suggest that there is heterogeneity in the variance components and if there are problems with infinite likelihood in the full model (i.e. including interaction terms and heterogeneous variance components), then it may be advisable to analyse all the subgroups separately.
The models applied for these analyses assumed that there was no correlation between the measurements in the two surveys, even though there were workers who were sampled in both surveys. To assess whether this assumption was valid we tested for autocorrelation in the data between the two surveys. Results of these analyses showed very little evidence of autocorrelation between the surveys, ranging between 0.00 and 0.21.
The dataset available to us was very large and comprehensive (prospective multi-year exposure survey with repeated exposure measurements) compared to most other exposure datasets. In order to understand how variance components behave in other industries, other large exposure databases are required. This is especially important since estimates of exposure distributions are essential for risk assessment and regulatory purposes (Rappaport et al., 1995) and epidemiological studies (Steenland et al., 2000; Loomis and Kromhout, 2004).
In conclusion, it appears that our previous assumption regarding the homogeneity of the variance components across time periods and factories for the time trends analyses of inhalable dust exposure in the carbon black industry were largely justified. No large differences were found when determining the industry-wide probability of overexposure using models with pooled and with heterogeneous variance components. However, some caution is advised when evaluating the factory-specific probability of overexposure, as there were significant differences when allowing for heterogeneity in the variance components. In addition, when using mixed-effects models to provide estimates of exposure for epidemiological studies, it is also advised to investigate whether there is heterogeneity of the variance components across the fixed effects. Results provided here are only from one particular industry, and therefore more insight is required into the behaviour of variance components over time and between plants and jobs for other industries.
Received May 27, 2004; in final form July 5, 2005
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Burstyn I, Kromhout H. (2002) Trends in inhalation exposure to hydrocarbons among commercial painters in the Netherlands. Scand J Work Environ Health; 28: 429–38.[Web of Science][Medline]
Burstyn I, Kromhout H, Kauppinen T et al. (2000) Statistical modelling of the determinants of historical exposure to bitumen and polycyclic aromatic hydrocarbons among paving workers. Ann Occup Hyg; 44: 43–56.
Gardiner K. (1995) Effects on respiratory morbidity of occupational exposure to carbon black: a review. Arch Environ Health; 50: 44–60.[Web of Science][Medline]
Gardiner K, Trethowan NW, Harrington JM et al. (1992) Occupational exposure to carbon black in its manufacture. Ann Occup Hyg; 36: 477–96.
Gardiner K, Calvert IA, van Tongeren MJA et al. (1996) Occupational exposure to carbon black in its manufacture: data from 1987 to 1992. Ann Occup Hyg; 40: 65–77.
HSE. (2005) EH40/05 workplace exposure limits. United Kingdom: Health and Safety Executive.
IARC. (1996) IARC monographs on the evaluation of carcinogenic risks to humans. Volume 65 Printing processes and printing inks, carbon black and some nitro compounds. Lyon, France: International Agency for Research on Cancer, World Health Organization.
Kromhout H, Vermeulen R. (2001) Temporal, personal and spatial variability in dermal exposure. Ann Occup Hyg; 45: 257–73.
Kromhout H, Symanski E, Rappaport SM. (1993) A comprehensive evaluation of within- and between-worker components of occupational exposure to chemical agents. Ann Occup Hyg; 37, 253–70.
Loomis D, Kromhout H. (2004) Exposure variability: concepts and applications in occupational epidemiology. Am J Ind Med; 45: 113–22.[CrossRef][Web of Science][Medline]
Lyles R, Kupper LL, Rappaport SM. (1997) A lognormal distribution-based exposure assessment method for unbalanced data. Ann Occup Hyg; 41: 63–76.
Mark D, Vincent JH. (1986) A new personal sampler for airborne total dust in workplaces. Ann Occup Hyg; 30: 89–102.
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: 99–109.
Peretz C, Goren A, Smid T et al. (2002) Application of mixed-effects models for exposure assessment. Ann Occup Hyg; 46: 69–77.
Rappaport SM. (1991) Assessment of long-term exposures to toxic substances in air. Ann Occup Hyg; 35: 61–121.
Rappaport SM, Kromhout H, Symanski E. (1993) Variation of exposure between workers in homogeneous groups. Am Ind Hyg Assoc J; 54: 654–62.[Web of Science][Medline]
Rappaport SM, Lyles R, Kupper LL. (1995) An exposure-assessment strategy accounting for within- and between-worker sources of variability. Ann Occup Hyg; 39: 469–95.
Rappaport SM, Weaver M, Taylor D et al. (1999) Application of mixed models to assess exposures monitored by construction workers during hot processes. Ann Occup Hyg; 43: 457–69.
Schlunssen V, Sigsgaard T, Schaumburg I et al. (2004) Cross-shift changes in FEV1 in relation to wood dust exposure: the implications of different exposure assessment methods. Occup Environ Med; 61: 824–30.
Steenland K, Deddens J, Zhao S. (2000) Biases in estimating the effect of cumulative exposure in log-linear models when estimated exposure levels are assigned. Scand J Work Environ Health; 26: 37–43.[Web of Science][Medline]
Symanski E, Kupper LL, Kromhout H et al. (1996) An investigation of systematic changes in occupational exposure. Am Ind Hyg Assoc J; 57: 724–35.[Web of Science][Medline]
Symanski E, Sällsten G, Chan W et al. (2001) Heterogeneity in sources of exposure variability among groups of workers exposed to inorganic mercury. Ann Occup Hyg; 45: 677–87.
Tornero-Velez R, Symanski E, Kromhout H et al. (1997) Compliance versus risk in assessing occupational exposure. Risk Analysis; 17: 279–92.[CrossRef][Web of Science][Medline]
van Tongeren MJA, Gardiner K, Calvert IA. (1994) An assessment of the weight-loss in transit of filters loaded with carbon black. Ann Occup Hyg; 38: 319–23.
van Tongeren MJA, Kromhout H, Gardiner K. (2000) Trends in levels of inhalable dust exposure, exceedance and overexposure in the European carbon black manufacturing industry. Ann Occup Hyg; 44: 271–80.
Vermeulen R, de Hartog J, Swuste P et al. (2000) Trends in exposure to inhalable particulate and dermal contamination in the rubber manufacturing industry: effectiveness of control measures implemented over a nine-year period. Ann Occup Hyg; 44: 343–54.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  C.-C. Chen, C.-L. Chuang, K.-Y. Wu, and C.-C. Chan Sampling Strategies for Occupational Exposure Assessment under Generalized Linear Model Ann. Hyg., July 1, 2009; 53(5): 509 - 521. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Tielemans, D. Noy, J. Schinkel, H. Heussen, D. Van Der Schaaf, J. West, and W. Fransman Stoffenmanager Exposure Model: Development of a Quantitative Algorithm Ann. Hyg., August 1, 2008; 52(6): 443 - 454. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Spaan, J. Schinkel, I. M. Wouters, L. Preller, E. Tielemans, E. T. Nij, and D. Heederik Variability in Endotoxin Exposure Levels and Consequences for Exposure Assessment Ann. Hyg., July 1, 2008; 52(5): 303 - 316. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. MEIJSTER, E. TIELEMANS, N. D. PATER, and D. Heederik Modelling Exposure in Flour Processing Sectors in The Netherlands: a Baseline Measurement in the Context of an Intervention Program Ann. Hyg., April 1, 2007; 51(3): 293 - 304. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||