Ann. occup. Hyg., Vol. 46, No. 1, pp. 69-77, 2002
© 2002 British Occupational Hygiene Society
Published by Oxford University Press
Article |
Application of Mixed-effects Models for Exposure Assessment
1Environmental and Occupational Health Group, Institute for Risk Assessment Sciences, Utrecht University, Utrecht, The Netherlands; 2The Department of Epidemiology, Sackler School of Medicine, Tel Aviv University, Tel Aviv 69978, Israel; 3EMGO Institute, Faculty of Medicine, Free University, Amsterdam, The Netherlands
Received 16 October 2000; in final form 20 June 2001.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
The benefits of using linear mixed-effects models for occupational exposure assessment were studied by re-analysing three data sets from two published surveys with repeated exposure measurements. The relative contributions of particular characteristics affecting exposure levels were assessed as in a multiple regression model, while controlling for the correlation between repeated measurements. While one-way ANOVA allows one only to estimate unconditioned variance components, a mixed model enables estimation of between- and within-worker variance components of exposure levels while accounting for the fixed effects of work characteristics. Consequently, we can identify the work characteristics affecting each variance component. Mixed models were applied to the data sets with repeated measurements and auxiliary information on work characteristics. The between-worker variance components were reduced by 35, 66 and 80%, respectively, in the three data sets when work characteristics were taken into account. The within-worker (day-to-day) variability was reduced only in the pig farmer data set, by 25%, when accounting for work activities. In addition, coefficients of work characteristics from the mixed model were compared with coefficients resulting from originally published multiple linear regression models. In the rubber manufacturing data, the coefficients of the mixed model showed similar relative importance, but were generally smaller than the coefficients from regression models. However, in the pig farm data, only the coefficients of work activities were somewhat reduced. The mixed model is a helpful tool for estimating factors affecting exposure and suitable variance components. Identifying the factors in the working environment that affect the between-worker variability facilitates a posteriori grouping of workers into more uniformly exposed groups. Identifying the factors that affect the within-worker variance is helpful for hazard control and in designing efficient sampling schemes with reference to time schedule.
Keywords: mixed-effects models; exposure assessment
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Ideally, exposure assessment of air pollutants at the workplace should be based on repeated measurements on randomly selected days of a randomly selected number of workers from a priori defined occupational groups (Rappaport, 1991; Rappaport and Smith, 1991; Burdorf, 1993; Kromhout et al., 1993; Boleij et al., 1995). Usually measurements done on the same worker are correlated.
Since exposure varies both within and between workers in a given exposure group (Rappaport, 1991; Rappaport and Smith, 1991; Kromhout et al., 1993; Rappaport et al., 1993; Peretz et al., 1997), these variance components should be taken into account in exposure assessment and for more effective hazard control, as well as in compliance testing and evaluation of exposure–response relationships (Buringh and Lanting , 1991; Heederik et al., 1991; Rappaport et al., 1995; Lyles et al., 1997).
Understanding the factors in the work environment that affect mean exposure levels enables the estimation of the between- and within-worker variance components conditioned on these factors. The identification of uniformly exposed groups of workers is essential for valid compliance testing and exposure–response evaluation. Identification of the factors in the work environment that are related to the between-worker variance component enables sub-grouping of workers into more uniformly exposed groups.
An understanding of the factors affecting within-worker variance assists in the identification of conditions in the work environment that cause concentrations to vary from day to day. This is a prerequisite for better protection of the individual worker from hazardous exposures.
So far, in studies with repeated measurements designs, most researchers have used either (i) a one-way random-effects model to estimate variance components, ignoring work characteristics (Burdorf et al., 1994; Woskie et al., 1994; Preller et al., 1995a,b; Nieuwenhuijsen et al., 1995; Kumagai et al., 1996; Symanski et al., 1996) and/or (ii) multiple linear regression to model the effect of work characteristics on observed exposure levels, ignoring the correlation between repeated observations from the same worker (Kromhout et al., 1994; Preller et al., 1995a,b). Mixed-effects models handling unbalanced data simultaneously estimate both the effects and the variance components in a more efficient way (Searle, 1988; Lindskey, 1993; Burton et al., 1998). For exposure groups those models can describe the influence of fixed and random work environment characteristics on the observed exposure levels, and estimate the within- and between-worker variance components controlled for work characteristics and other determinants of exposure. Recently, mixed-effects models were used by several researchers for different purposes (Symanski et al., 1996; Nylander-French et al., 1999; Rappaport et al., 1999; Burstyn et al., 2000). A time trend can be introduced into these models as a fixed effect. In this paper we present the benefits of using mixed-effects models for unbalanced data to estimate variance components while controlling for work characteristics. In addition, we present the coefficients of the work characteristics that affect exposure levels, controlled for the correlation between repeated measurements. Finally, we will show which work characteristics affect between- and within-worker variability in exposure concentrations.
For our analysis, we used three existing data sets with repeated personal exposure measurements, as described previously (Kromhout et al., 1994; Preller et al., 1995a,b). A common feature of the data sets was that they were from systematic surveys. Auxiliary information on the work environment and activities was collected during the measurements. The data sets stemmed from two industry-wide surveys among workers from the rubber manufacturing industry and one survey among pig farmers in The Netherlands. Detailed information on these studies, and the results from the one-way random-effects models and from the multiple linear regression models, can be found in the above-mentioned papers (Kromhout et al., 1994; Preller et al., 1995a,b).
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Study design, data collection and previous statistical analysis
First example: industry-wide survey of the rubber manufacturing industry
This study of the rubber manufacturing industry was performed in The Netherlands, to examine relationships between working conditions and chemical exposures. Personal exposures to airborne particulates, rubber fumes and solvents, as well as dermal contaminants, were measured in a representative sample of 10 factories producing an array of different rubber products. For each plant the measurements and observations took 4 days (Tuesday–Friday). Auxiliary data on tasks performed, use of personal protection devices, ventilation characteristics and process characteristics were collected through interviews of sampled workers. Workers were selected as stratified by production function and the job done, and surveyed on randomly chosen days during the course of the 4 day measurement period.
Multiple regression models were applied to evaluate the relationships between the collected auxiliary data and exposure levels, for two groups of workers: 234 workers with 620 measurements exposed to inhalable particulates, and a subgroup of 36 workers with 59 measurements exposed to rubber fumes (measured as the cyclohexane-soluble fraction of the inhalable particulates). Details of the study and the modelling can be found elsewhere (Kromhout et al., 1994).
Second example: survey on pig farmers’ exposure to inhalable endotoxin
In a study among 98 pig farmers from the south of The Netherlands, exposure to inhalable dust and endotoxin was monitored by personal sampling. Exposure was measured during one work shift on a randomly chosen day of the week; one day during the summer of 1991 and one day during the winter of 1992. Outdoor temperature was obtained from a monitoring station in the south of The Netherlands. Task activity patterns and farm characteristics were also recorded. Activities, which were represented by time spent in each activity, were based on daily averages during 12–14 days. For the purpose of this paper, only the exposure data on endotoxin will be used. Multiple linear regression analyses were applied to evaluate the relationship between farm characteristics, activities and outdoor temperature and log-transformed endotoxin concentrations. A one-way random-effects model was applied to estimate variance components. Details of the study and the modelling can be found elsewhere (Preller et al., 1995a,b).
Sources of exposure variability
We postulated that the variability of the exposure levels in an industrial hygiene (IH) group of workers arises from four sources.
1. Systematic between-worker variation: systematic differences in factors that define the work conditions of different workers. These factors are mostly spatially related (e.g. local ventilation), varying among workers but constant in time for each worker. Sometimes these factors are both temporal and spatial (e.g. process temperature), meaning that the levels differ among workers (the mean value) and within the same worker along time.
2. Random between-worker variation: differences among workers beyond what can be explained by specific factors. This additional variation may be associated with factors that are not measured due to time/money limations, inability to measure (e.g. workers’ habits) or lack of awareness.
3. Systematic within-worker variation: systematic differences in factors that define the work conditions of the same worker over time. These factors are temporal (e.g. burden, activities) and may be common to all workers in the IH group. Time itself is one possible systematic, within-worker factor (e.g. season, year). Usually these within-worker changes are related to the cycle of work, the production, seasonality etc.
4. Random within-worker variation: differences among measurements on the same worker at different time points beyond what can be explained by specific factors. This additional variation may be associated with further within-worker factors (e.g. changes of habits of a worker) that are not measured due to time/money limations, inability to measure or lack of awareness (e.g. measures taken by different hygienists, measurement errors).
Thus, within the same exposure group along time, the usual partition of the total exposure variance into two components—between workers and within workers (Rappaport, 1991; Rappaport and Smith, 1991; Kromhout et al., 1993; Rappaport et al., 1993; Peretz et al., 1997)—can be refined when work characteristics are taken into account. (i) Systematic variation (sources 1 and 3) accounts for differences in work characteristics. These differences can be included as explanatory variables in the model whose effects can be estimated. (ii) Random variation (sources 2 and 4), which is partitioned into two components: (a) between-worker variation, conditioned on the effects of the observed work characteristics (source 2). This variance component reflects additional variation among workers beyond differences due to work factors; and (b) within-worker variation (source 4), which reflects additional variation at the within-worker level.
The total random variance when work characteristics are taken into account is a conditional variance, so its value is less than the total variance when those factors are not taken into account, and is the sum of the conditioned random between-worker variance and the random within-worker variance.
Mixed-effects models
A mixed-effects model is a generalization of the standard linear model (a multiple regression model) that enables the analysis of data generated from several sources of variation instead of just one (SAS, 1996). It associates one continuous dependent variable (a response, an outcome) with several explanatory variables (categorical or continuous). The unique aspect of the mixed-effects model is the inclusion of both fixed and random factors. Fixed effects provide estimates of the average responses in the group, like in a common regression model, while random effects (e.g. subjects’ effects) account for the natural heterogeneity in the responses of different individuals and allow estimation of responses for each individual in the study. Since measurements done on the same subject are correlated, this correlation must be taken into account in modelling. The dependence among the repeated responses can be of different types, leading to specific covariance/correlation structures. The model allows the assumption of several covariance structures and enables estimation of the effects as well as variance parameters. The number of observations per subject can be either the same (a balanced design) or different (an unbalanced design). The time points can be either identical across subjects or not. The time interval between repeated observations can vary across repetitions (Lindskey, 1993).
Current application of the mixed-effects model
The application of this model for identifying the determinants of exposure and assessing variance components is presented in this paper using two examples: one from the rubber manufacturing industry and another from pig farming. In both examples, the dependent variable was the exposure level of a pollutant and the explanatory variables were workplace characteristics. These fixed effects were either time dependent (e.g. outdoor temperature) or fixed along time (e.g. stable flooring). The individual worker’s effect was taken as a random effect. We looked at two sources of random variance: that between workers and that among repeated measurements within workers. In the first example we had three repetitions, and in the second example we had only two repetitions per worker. We assumed that any two repeated measurements of the same worker have equal correlation irrespective of the time interval between them; a compound symmetry covariance structure. This is the covariance structure assumed in classical repeated measurements ANOVA. Furthermore, we assumed that the variance between workers is equal across the fixed factors of work characteristics (defined as homogeneity) and that the effects as well as the variance within workers are equal across all workers in the same group. The residuals were assumed to be independent of each other. Exposure concentrations were assumed to be log-normally distributed (Rappaport, 1991; Rappaport and Smith, 1991; Kromhout et al., 1993; Peretz et al., 1997).
Mixed models were applied with and without the fixed effects (without- is equivalent to the random-effects model), to determine the impact of the fixed effects upon the variance component associated with the random effects. The estimated variance components of both models were compared.
PROC MIXED from SAS System Software Version 6.1 (SAS, 1996) was used for the analysis. The procedure enables simultaneous estimation of the effects, their standard error and significance tests, as well as variance components and their confidence limits. Variance components were estimated using the restricted maximum likelihood (REML) method. Nested models were compared by likelihood ratio test.
The mixed-effects model is specified by the following expressions:
for I = 1,...,k (workers) and j = 1,...,ni (repetitions of the ith worker), where Yij is the log-transformed exposure level; ß0 is an overall intercept for the group that corresponds to mean background exposure (log-transformed) when all factors equal zero; ß1,...,ßp are fixed effects; xij1,...,xijp are values of the variables for the ith worker on the jth day; b1,...,bk are workers’ random effects; bi is the ith worker random effect, which corresponds to the discrepancy between his intercept and the group intercept ß0; and z1,...,zk are workers’ indicators (0/1).
It is furthermore assumed that
bi 
N(0, 
2b)
where bis are all independent;
ij N(0, 2w)
where ijs are all independent, and bis and ijs are all independent of each other;
2 = 2b + 2w
where 2b is the variance between-workers and 2w is the variance within-workers;
= 2b / 2
where is the correlation between any two repeated measures of the same worker.
compound symmetry structure
For every i, i = 1, . . .,k and for every j,l, j,l = 1, . . .,ni
(All yij of the same worker are correlated, and those of different workers are uncorrelated.)
To model the influence of work characteristics on the exposure levels they were considered as fixed effects in the above model; for example, ß1 is the process pressure effect; ß2 is the process temperature effect; ß3 is the local exhaust ventilation effect; xij1 is the process pressure value for the ith worker on the jth day; xij2 is the process temperature value for the ith worker on the jth day; and xij3 is the local exhaust ventilation indicator (0—absent/1—present) for the ith worker on the jth day.
To identify time trends in the above model, a time term was added to the model: ß4 is the period effect; and xij4 is the period indicator (0—first period/1—second period), for the ith worker on the jth day.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
First example: industry-wide survey of the rubber manufacturing industry
Results of the application of both the random-effects model (the model without the fixed effects) and the mixed-effects model (the model with the fixed effects) were compared in the two data sets with exposure to inhalable particulates and rubber fumes (see Tables 1 and 2). The effect of the day of the week on the exposure level was tested separately. Table 1 shows that for workers exposed to inhalable particulates, the factors affected the between-worker component of variance (
2b) considerably (35% reduction from 1.30 to 0.84), but did not alter the within-worker component of variance (2w). In the two models 2b is different from zero (P < 0.05) and the models were significantly different (P = 0.0023). The difference between the models suggests that there are no systematic changes in work tasks and production characteristics for individual workers during a week. The main difference between the mixed-effects model and the original multiple regression model (Kromhout et al., 1993) assuming independence between repeated measurements, was that fewer exposure-affecting factors were statistically significant or borderline statistically significant (P < 0.10) (17 versus 28 factors; see Table 3).
|
|
However, the coefficients had the same sign for all factors except one task (‘extruding–slicing’), whose coefficient was nearly 0 in the mixed-effects model. The tasks ‘punching powdered products’, ‘packing powdered products’, ‘tube inspection’ and ‘jointing’ affected exposure to inhalable particulates most dramatically, according to both the mixed-effects model and the original multiple regression model.
For workers exposed to rubber fumes (Table 2), the same phenomenon for the components of variance was observed. The three factors, one pure spatial between-workers factor (local ventilation) and the other two (process temperature and pressure) both spatial and temporal, affected the between-worker component of variance (66% reduction from 0.53 to 0.18). The within-worker component of variance (2w) was not affected. In the two models 2b is different from zero (P < 0.05) and the models were not significantly different.
|
The coefficients of the factors were almost identical when compared with the original multiple linear regression (Table 4). The effect of ‘day of the week’ (Tuesday–Friday) on the mean exposure and on the components of variance was non-significant for both exposure models (to inhalable particulates and rubber fumes), suggesting no systematic differences in exposure over the course of a week.
|
Second example: survey on pig farmers’ exposure to inhalable endotoxin
Table 5 presents results both the original one-way random-effects model and a mixed-effects model with 21 fixed effects for both farm characteristics and activities.
|
Furthermore, results from three additional mixed models with respectively only farm characteristics, only activities and outdoor temperature, and only a season term were elaborated, and are presented in the same table.
From the random-effects model (Table 5) it was clear that the within-worker variance component had a greater weight than the between-workers variance component (85 versus 15% of the total exposure variance). The between-worker variance component was low (0.11). Table 5 shows the extent of reduction in both the within- and between-worker variance components by including all the statistically significant factors from the original multiple regression model. The within-worker variance component was reduced by 0.16 (25%) and the two models are significantly different (P < 0.0001). The between-worker variance component was reduced by 0.09 (80%), and while in the random-effects model, 2b is different from zero (P < 0.05), in the mixed model with the farm characteristics, 2b 0 .
In Table 5 we see that farm characteristics appeared to be solely responsible for the reduction in the between-worker variance component (2b = 0.01). Outdoor temperature and farmers’ activities (pure within-worker factors) had no effect on the between-worker variance component (2b = 0.12), while they reduced the within-worker variance component by 25%. This clearly shows that different work environment characteristics contributed independently to the variance components. Farm characteristics, which are almost constant over the time period studied (1 yr), were responsible for differences in average endotoxin concentrations between farmers, while changes from day to day in temperature and activities performed led to temporal variability in exposure concentrations. The season factor was found to have a minor influence on the within-worker variance component. The two models (the random-effects model and the mixed-effects model with season effect) were found to be significantly different (P < 0.0001). When compared with the coefficients of the original multiple linear regression model (Preller, 1995a,b), the estimated coefficients from the mixed-effects model were almost exactly the same for the farm characteristics, but somewhat smaller for the activities (Table 6). Neither the relative order nor the P-values changed.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
The above examples illustrate the major contribution of the mixed-effects model in unbalanced designs to the investigation of exposure variance components and exposure-affecting factors. In contrast with the one-way random-effects model, the mixed-effects model deals with both fixed and random effects. It estimates the between- and within-worker variance while adjusting for fixed effects. Simultaneously, it assesses the linear relationships between the determinants of exposure (usually fixed factors) and exposure levels. Common multiple linear regression can be correctly applied when each worker has only a single measurement.
With repeated measurements of each worker, some dependence amongst repeated measurements will exist and the correlation between values for a given person must be taken into account when estimating the relationship between the determinants and exposure levels (Lindskey, 1993; Burton et al., 1998). The mixed-effects model is capable of taking this dependence into account in the modelling process.
In this study we used mixed-effects models to understand the relationship between specific work environment characteristics and between- and within-worker exposure variance components. Identification of specific work characteristics, which affect the between-worker variance component, will enable development of criteria for defining uniformly exposed groups of workers (Rappaport, 1991; Boleij et al., 1995; Rappaport et al., 1995; Symanski et al., 1996). Grouping workers into sub-groups is an inherent part of the work of an industrial hygienist in exposure surveys, as well as in compliance tests and epidemiological studies (Rappaport, 1991; Rappaport et al., 1993; Boleij et al., 1995). Despite the widespread use of grouping strategies, there is so far only limited experience with optimization of these strategies (Boleij et al., 1995; Seixas and Sheppard, 1996; Kromhout et al., 1997).
The analyses with the mixed-effects models forms the basis for the creation of uniformly exposed groups of workers. Reliance on observational factors such as a job title, which may lead to non-uniformly exposed groups, seems no longer to be necessary (Boleij et al., 1995). For instance, classifying curing workers in rubber manufacturing based on the determinant process temperature and pressure will lead to more uniform and distinctly different rubber fume exposure groups. It is further recommended that if the IH group cannot be split into sub-homogeneous groups due to the small number of workers, the testing of overexposure in IH groups, as suggested by Lyles et al. (1997), which relates to both within and between variance components, should be refined to account for the effects of work characteristics in these groups. As we have shown here, these effects can be estimated with mixed-effects models. Also, in this study, specific factors were identified which mainly influenced variability in exposure levels from day to day (within-worker). Hazard control should focus on these factors.
The two examples presented in this paper describe different work situations and measurement strategies. In the first example, 1–3 randomly chosen measurements were performed within a week on two groups of rubber workers in The Netherlands (Kromhout et al., 1994). In the second example, 1–2 measurements were collected in two different seasons of a particular year among Dutch pig farmers (Preller et al., 1995a,b). Different work characteristics were documented in each example. In the first example, for exposure to particulates, a reduction of 25% of the between-worker variance component was the effect of 17 factors affecting exposure. Classification of rubber workers into uniformly exposed groups will have to rely on those identified factors. In the second example, exposure to endotoxin among pig farmers, the reduction of 82% in between-worker variance was mainly an effect of the inclusion of 12 farm characteristics. Consequently, the classification of pig farmers into uniformly exposed groups will have to rely on those farm characteristics.
In the second example (pig farmers), there was a clear distinction between characteristics influencing each of the variance components. Eight time-dependent activities, as well as outdoor temperature, all pure within-worker factors, reduced the within-person (day-to-day) variability by 25%. From diary information collected among the pig farmers, it appeared that some of the activities followed a distinct temporal pattern, with some activities taking place only on particular days of the week (Preller et al., 1995a,b). The farm characteristics, pure between-worker factors, were responsible for reducing the between-worker variance component to zero.
In the mixed model, when the between-worker variance component is close to zero, the coefficients for the fixed effects would be very similar to those from a multiple regression model, assuming independence between repeated observations. In the model for the pig farmers, this was indeed the case (between-worker component = 0.02). For the rubber workers’ exposure to inhalable particulates the opposite situation, with a very large between-worker variance component, led to changes in the coefficients. In addition, fewer statistically significant exposure-affecting factors were found in the mixed model, since the significance tests of the coefficients in the mixed model are different.
In all the examples, we assumed that any two repeated measurements of the same worker had equal correlation, irrespective of the time interval between them; a compound symmetry covariance structure. This was because we had only two repetitions per worker in the second example and three in the first example. We also assumed independence across subjects. In general, the residuals can be analysed to check for departures from the independence assumption. However, such analyses will be relevant only when there are a reasonable number of observations within workers, say at least 8–10. Other models with different dependence structures may be applied when the time course of the exposure level of each person is of primary interest, that is, when the correlation itself has scientific relevance. The mixed-effects model was developed over the last decade, and were recently introduced in common statistical packages such as BMDP (5V), SAS (Proc Mixed) and S-Plus (BMDP, 1990; SAS, 1996; S-Plus, 1997). The computerized procedures enable simultaneous estimation of parameters, approximate standard errors and significance tests that have not been available before. One should take into account the fact that, in models for unbalanced data, the estimators are proxies, since the maximum likelihood estimates have approximately normal distribution. Balanced data (the same number of repeated measurements for each worker) are preferable whenever possible, in order to obtain more accurate estimators.
However, the mixed-effects model enables estimation of variance components of exposure levels that have been adjusted for workplace factors in order to improve the assessment of exposure. This statistical method can be used to improve future sampling strategies through the grouping of workers into more uniformly exposed groups, and for the identification of specific workplace conditions that should be controlled.
Acknowledgements—The authors are indebted to Dick Heederik for suggesting the idea for this paper. We would also like to acknowledge Liesbeth Preller for providing the pig farmers’ exposure data sets, Prof. D. Steinberg for his statistical review, P. Goldberg and Prof. S. M. Rappaport for their helpful comments, and the reviewers for improving the paper.
|
FOOTNOTES |
|---|
* Author to whom correspondence should be addressed. The Department of Epidemiology, Sackler School of Medicine, Tel Aviv University, Tel Aviv 69978, Israel. Tel: +972-3-6409867; fax: +972-3-6410555; e-mail: cperetz{at}post.tau.ac.il
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
BMDP. (1990) Statistical software, vol. 2. Berkeley, CA: University of California.
Boleij J, Buringh E, Heederik D, Kromhout H. (1995) Occupational hygiene of chemical and biological agents. Amsterdam: Elsevier.
Burdorf A. (1993) Shifting concepts in assessment of occupational exposures. Ann Occup Hygiene; 37: 447–50.
Burdorf A, Lillienberg L, Brisman J. (1994) Characterization of exposure to inhalable flour dust in Swedish bakeries. Ann Occup Hyg; 38: 67–78.
Buringh E, Lanting R. (1991) Exposure variability in the workplace its implication for the assessment of compliance. Am Ind Hyg Assoc J; 52: 6–13.[Web of Science][Medline]
Burstyn I, Kromhout H, Kauppinen T, Heikkila P, Bofefetta P. (2000) Statistical modelling of the determinants of historical exposure to bitumen and polycyclic aromatic hydrocarbons among paving workers. Ann Occup Hyg; 44: 43–56.
Burton P, Gurrin L, Sly P. (1998) Extending the simple linear regression model to account for correlated responses: an introduction to generalized estimating equation and multi-level mixed modelling. Statist Med; 17: 1261–91.
Heederik D, Boleij JSM, Kromhout H. (1991) Use and analysis of exposure monitoring data in occupational epidemiology. An example of an epidemiological study in the Dutch animal food industry. Appl Occup Environ Hyg; 5: 458–64.
Kromhout H, Symanski E, Rappaport RM. (1993) A comprehensive evaluation of within- and between-worker components of occupational exposure to chemical agents. Ann Occup Hyg; 37: 253–70.
Kromhout H, Swuste P, Boleij JSM. (1994) Empirical modelling of chemical exposure in the rubber manufacturing industry. Ann Occup Hyg; 38: 3–22.
Kromhout H, Loomis DP, Kleckner RC, Savitz DA. (1997) Sensitivity of the relation between cumulative magnetic field exposure and brain cancer mortality to choice of monitoring data grouping scheme. Epidemiology; 8: 442–5.[Web of Science][Medline]
Kumagai S, Kusaka Y, Goto S. (1996) Cobalt exposure level and variability in the hard metal industry of Japan. Am Ind Hyg Assoc J; 57: 365–9.[Web of Science][Medline]
Lindskey JK. (1993) Models for repeated measurements. Oxford: Clarendon Press.
Lyles RH, Kupper LL, Rappaport SM. (1997) A lognormal distribution-based exposure assessment method for unbalanced data. Ann Occup Hyg; 41: 63–76.
Nieuwenhuijsen MJ, Lowson D, Venable KM, Newman Taylor AJ. (1995) Flour dust exposure variability in flour mills and bakeries. Ann Occup Hyg; 39: 299–305.
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, Goldberg P, Kahan E, Grady S, Goren A. (1997) The variability of exposure over time: a prospective longitudinal study. Ann Occup Hyg; 41: 485–500.
Preller L, Heederik D, Kromhout H, Tielen M. (1995a) Determinants of dust and endotoxin exposure of pig farmers: development of a control strategy using empirical modelling. Ann Occup Hyg; 39: 545–57.
Preller L, Kromhout H, Heederik D, Tielen M. (1995b) Modelling long-term average exposure in occupational exposure–response analysis. Scand J Work Environ Health; 21: 504–12.[Web of Science][Medline]
Rappaport SM. (1991) Assessment of long-term exposure to toxic substances in air. Ann Occup Hyg; 35: 61–121.
Rappaport SM, Smith TH. (1991) Exposure assessment for epidemiology and hazard control. Michigan: Lewis Publishers Inc.
Rappaport SM, Kromhout H, Symanski E. (1993) Variation of exposure between workers in homogenous exposure groups. Am Ind Hyg Assoc J; 54: 654–62.[Web of Science][Medline]
Rappaport SM, Lyles RH, 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, Kupper L, Susi P. (1999) Application of mixed models to assess exposures monitored by construction workers during hot processes. Ann Occup Hyg; 43: 457–69.
SAS. (1996) SAS/STAT software changes and enhancements. Cary, NC: SAS Institute.
Searle SR. (1988) Mixed models and unbalanced data: wherefrom, whereat and whereto. Commun Statist A Theory Methods; 17: 935–68.
Seixas NS, Sheppard L. (1996) Maximizing accuracy and precision using individual and grouped exposure assessments. Scand J Work Environ Health; 22: 94–101.[Web of Science][Medline]
S-Plus. (1997) Data analysis products division. Seattle, WA: Math-Soft.
Symanski E, Kupper LL, Kromhout H, Rappaport SM. (1996) An investigation of systematic changes in occupational exposure. Am Ind Hyg Assoc J; 57: 724–35.[Web of Science][Medline]
Woskie SA, Shen P, Eisen EA, Finkel MH, Smith TJ, Smith R, Wegman DH. (1994) The real time dust exposures of sodium borate workers: examination of exposure variability. Am Ind Hyg Assoc J; 55: 207–17.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  K. S. Galea, M. Van Tongeren, A. J. Sleeuwenhoek, D. While, M. Graham, A. Bolton, H. Kromhout, and J. W. Cherrie Trends in Wood Dust Inhalation Exposure in the UK, 1985-2005 Ann. Hyg., October 1, 2009; 53(7): 657 - 667. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J M Dement, D Myers, D Loomis, D Richardson, and S Wolf Estimates of historical exposures by phase contrast and transmission electron microscopy in North Carolina USA asbestos textile plants Occup. Environ. Med., September 1, 2009; 66(9): 574 - 583. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T Mee, P Whatmough, L Broad, C Dunn, M Maslanyj, S Allen, K Muir, P A McKinney, and M van Tongeren Occupational exposure of UK adults to extremely low frequency magnetic fields Occup. Environ. Med., September 1, 2009; 66(9): 619 - 627. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. Sbihi, K. Teschke, Y. C. Macnab, and H. W. Davies Determinants of Use of Hearing Protection Devices in Canadian Lumber Mill Workers Ann. Hyg., July 1, 2009; (2009) mep043v1. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. Hagstrom, C. Lundholm, K. Eriksson, and I. Liljelind Variability and Determinants of Wood Dust and Resin Acid Exposure during Wood Pellet Production: Measurement Strategies and Bias in Assessing Exposure-Response Relationships Ann. Hyg., November 1, 2008; 52(8): 685 - 694. [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]
|
|
|
|
 |
|
 S. Spaan, G. Doekes, D. Heederik, P. S. Thorne, and I. M. Wouters Effect of Extraction and Assay Media on Analysis of Airborne Endotoxin Appl. Envir. Microbiol., June 15, 2008; 74(12): 3804 - 3811. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Spaan, D. J. J. Heederik, P. S. Thorne, and I. M. Wouters Optimization of Airborne Endotoxin Exposure Assessment: Effects of Filter Type, Transport Conditions, Extraction Solutions, and Storage of Samples and Extracts Appl. Envir. Microbiol., October 1, 2007; 73(19): 6134 - 6143. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. OGDEN Annals of Occupational Hygiene at Volume 50: Many Achievements, a Few Mistakes, and an Interesting Future Ann. Hyg., November 1, 2006; 50(8): 751 - 764. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. SYMANSKI, S. MABERTI, and W. CHAN A Meta-Analytic Approach for Characterizing the Within-Worker and Between-Worker Sources of Variation in Occupational Exposure Ann. Hyg., June 1, 2006; 50(4): 343 - 357. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. C. FRIESEN, Y. C. MACNAB, S. A. MARION, P. A. DEMERS, H. W. DAVIES, and K. TESCHKE Mixed Models and Empirical Bayes Estimation for Retrospective Exposure Assessment of Dust Exposures in Canadian Sawmills Ann. Hyg., April 1, 2006; 50(3): 281 - 288. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L A M Smit, I M Wouters, M M Hobo, W Eduard, G Doekes, and D Heederik Agricultural seed dust as a potential cause of organic dust toxic syndrome Occup. Environ. Med., January 1, 2006; 63(1): 59 - 67. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. VAN TONGEREN, I. BURSTYN, H. KROMHOUT, and K. GARDINER Are Variance Components of Exposure Heterogeneous Between Time Periods and Factories in the European Carbon Black Industry? Ann. Hyg., January 1, 2006; 50(1): 55 - 64. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I. M. WOUTERS, S. SPAAN, J. DOUWES, G. DOEKES, and D. HEEDERIK Overview of Personal Occupational Exposure Levels to Inhalable Dust, Endotoxin, {beta}(1->3)-Glucan and Fungal Extracellular Polysaccharides in the Waste Management Chain Ann. Hyg., January 1, 2006; 50(1): 39 - 53. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R Vermeulen and H Kromhout Historical limitations of determinant based exposure groupings in the rubber manufacturing industry Occup. Environ. Med., November 1, 2005; 62(11): 793 - 799. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. LAVOUE, C. BEAUDRY, N. GOYER, G. PERRAULT, and M. GERIN Investigation of Determinants of Past and Current Exposures to Formaldehyde in the Reconstituted Wood Panel Industry in Quebec Ann. Hyg., October 1, 2005; 49(7): 587 - 602. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A Burdorf Identification of determinants of exposure: consequences for measurement and control strategies Occup. Environ. Med., May 1, 2005; 62(5): 344 - 350. [Full Text] [PDF]
|
|
|
|
|
|
I. BURSTYN Principal Component Analysis is a Powerful Instrument in Occupational Hygiene Inquiries Ann. Hyg., November 1, 2004; 48(8): 655 - 661. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B Bakke, B Ulvestad, P Stewart, and W Eduard Cumulative exposure to dust and gases as determinants of lung function decline in tunnel construction workers Occup. Environ. Med., March 1, 2004; 61(3): 262 - 269. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I Liljelind, S Rappaport, K Eriksson, J Andersson, I A Bergdahl, A-L Sunesson, and B Jarvholm Exposure assessment of monoterpenes and styrene: a comparison of air sampling and biomonitoring Occup. Environ. Med., August 1, 2003; 60(8): 599 - 603. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. BURDORF and M. V. TONGEREN Commentary: Variability in Workplace Exposures and the Design of Efficient Measurement and Control Strategies Ann. Hyg., March 1, 2003; 47(2): 95 - 99. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. M. RAPPAPORT, M. GOLDBERG, P. SUSI, and R. F. HERRICK Excessive Exposure to Silica in the US Construction Industry Ann. Hyg., March 1, 2003; 47(2): 111 - 122. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||