Annals of Occupational Hygiene Advance Access originally published online on April 21, 2005
Annals of Occupational Hygiene 2005 49(7):587-602; doi:10.1093/annhyg/mei014
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© 2005 British Occupational Hygiene Society Published by Oxford University Press
Original Article |
Investigation of Determinants of Past and Current Exposures to Formaldehyde in the Reconstituted Wood Panel Industry in Quebec
1 Groupe de recherche interdisciplinaire en santé (GRIS), Département de santé environnementale et santé au travail, Faculté de médecine, Université de Montréal, P.O. Box 6128, Main Station, Montreal (QC) H3C 3J7, Canada; 2 Quebec Research Institute for Occupational Health and Safety (IRSST), 505, De Maisonneuve blvd. West, Montréal (QC) H3A 3C2, Canada
* Author to whom correspondence should be addressed. Tel: +1 514 343 6134; fax: +1 514 343 2200; e-mail: michel.gerin@umontreal.ca
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
OBJECTIVES: Past and present formaldehyde measurements made in facilities manufacturing reconstituted wood panels in Quebec have been collected in order to assess formaldehyde exposure and its determinants in this industry.
METHODS: All 12 plants manufacturing Oriented-strand board (OSB), Medium density fibreboard (MDF) and Particle board (PB) in Quebec were visited by a research team which took area and personal measurements. Past measurements taken by governmental occupational health teams in these plants were also collected. Log-transformed formaldehyde concentrations were analysed with extended linear mixed-effects models.
RESULTS: During 2001–2002, 275 measurements were taken by the research team, while 590 measurements dating back to 1984 were collected from governmental files. The area measurements had a global geometric mean (GM) of 0.28 p.p.m. [geometric standard deviation (GSD): 3.1]. The GM of the personal measurements was 0.17 p.p.m. (GSD: 2.3). The fixed-effects of the models for personal and area measurements explained 61 and 57% of the variance, respectively. Job (working area for area concentrations), process (PB, MDF, OSB), season of sampling, origin of the data (research, governmental) and year of sampling were significant determinants of exposure. Proximity to the press, winter conditions, PB and MDF processes and governmental data resulted in the highest exposures. Significant within-sampling campaign correlation was found for both personal and area models. The final models include different residual variances by process for personal measurements and by working area for area measurements.
CONCLUSIONS: Several determinants of exposure to formaldehyde in the reconstituted wood panel industry were successfully identified. Higher levels found in governmental data as compared to research data may be explained by a ‘worst-case’ strategy bias. The observed intra-sampling campaign correlation supports existing results suggesting that measurements taken in a small time frame tend to be correlated. Exposures in this sector are low compared to most 8 h-TWA occupational exposure limits (e.g. 1 p.p.m.) but close to the most demanding ones (e.g. 0.3 p.p.m.).
Keywords: formaldehyde • determinants of exposure • mixed-effects models • particle board • oriented-strand board • medium density fibre board
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Context
The manufacture of reconstituted wood panels has long been associated with exposure to formaldehyde, which is an irritant gas considered by the International Agency for Research on Cancer (IARC) as carcinogenic to humans (group 1) (International Agency for Research on Cancer, 2005). Most of the processes in this industry involve the use of formaldehyde-based resins as the binding agent. The recent change in the IARC classification of formaldehyde from group 2A (probably carcinogenic) to group 1, based on sufficient evidence that it causes nasopharyngeal cancer in humans, constitutes an incentive for improved exposure surveillance in workplaces where formaldehyde is present. Furthermore, information on levels and determinants of exposure to this substance are needed to help improve exposure assessment in epidemiological studies on the carcinogenicity of formaldehyde at other sites (such as the nasal cavity or the blood-forming system). Within the framework of a more global project aimed at estimating the economic and health impacts of lowering the occupational exposure limit (OEL) for formaldehyde in Quebec, an exposure assessment was conducted in the reconstituted wood panels industry (Goyer et al., 2004
). Scenarios being studied for full shift exposure include an 8 h-TWA OEL of 1, 0.75 and 0.3 p.p.m. With nearly 2000 workers in 2001–2002, this industry constitutes the largest industrial sector with formaldehyde exposure in Quebec.
Description of the industry
The reconstituted wood panel industry includes several processes and can be classified as either producing plywood products (in which panels are formed by the assembly of thin layers of wood) or composition boards (in which wood particles are bonded together to form the panels) (Zimowski, 1986; US Census Bureau, 2002). The processes included in this study are limited to particle board (PB), medium density fibre board (MDF) and oriented-strand board (OSB), which all belong to the composition board category. These three processes involve the mixing of wood chips with a binding agent and compressing the mixture under high temperature (USEPA, 1998). For PB, the main raw materials comprise wood chips, saw dust and planer shavings. In the MDF process the chips are formed into fibres prior to mixing with the resin. Liquid urea-formaldehyde (UF) and melamine-urea-formaldehyde (MUF) resins are generally used in the manufacture of MDF and PB. OSB panels are made from wood wafers produced on site from logs and mixed with liquid or powder phenol-formaldehyde (PF) resins.
Exposure to formaldehyde
The main sources of occupational exposure to formaldehyde in this sector include emissions from resins before pressing, mostly during the formation of the mat and along the system conveying it to the press, emissions from the press and emissions from the newly formed panels in the finishing, storage and shipping areas. Formaldehyde emissions from resins are caused in part by their free formaldehyde content, but most emissions at the press and from the hot panels arise from the hydrolysis of the cured resin and condensation reactions between wood compounds (Tohmura et al., 2001). Tohmura et al. have observed a decrease in formaldehyde emissions when the melamine content in MUF resins is increased, caused by an increased resistance of the cured resins to hydrolysis (Tohmura et al., 2001). Wolcott et al. have investigated variables affecting formaldehyde emissions from the press during laboratory production of UF-bonded PB (Wolcott et al., 1996). They found that emissions increased when the following parameters increased: press time, press temperature, mat humidity, mat content in resin and formaldehyde/urea molar ratio of the resin. Formaldehyde emissions were inversely associated with the thickness of the panels.
The OSHA Health Response Team sampled formaldehyde in four facilities in 1986, three manufacturing PB and one producing MDF (Zimowski, 1986). They report geometric means (GMs) of personal measurements for different jobs ranging from 0.10 to 0.32 p.p.m. and from 0.18 to 1.8 p.p.m. in PB and MDF, respectively. All measurement durations were >250 min. Kauppinen and Niemelä, in a review of formaldehyde exposure levels in facilities manufacturing PB in Finland prior to 1985, report GMs of area measurements between 0.4 and 2.3 p.p.m. depending on the localization in the plant. All measurement durations were below 120 min (Kauppinen and Niemelä, 1985). Niemelä et al. also report an arithmetic mean (AM) of 1.15 p.p.m. for 220 measurements made between 1977 and 1979 (Niemelä and Vainio, 1981). More recently, Niemelä et al. presenting historical trends in this industry based on measurements taken in eight facilities between 1980 and 1994, report successive measurement medians of 0.91 (n = 21), 0.26 (n = 31) and 0.46 p.p.m. (n = 9) for the periods 1980–1985, 1986–1990 and 1991–1994 (Niemelä et al., 1997). In their review of occupational exposure to formaldehyde, the authors of the 1995 IARC monograph, citing the results of a Swedish study, report an AM of 0.2 p.p.m. (obtained from 19 values) measured between 1980 and 1989 in facilities manufacturing MDF (International Agency for Research on Cancer, 1995). A value of 0.3 p.p.m. for facilities manufacturing PB is also reported. Edling et al. studied the effects of formaldehyde on the physiology of nasal mucosa in workers of three plants manufacturing PB. They report that measurements taken by in-house hygienists between 1975 and 1983 were in the range 0.08–0.9 p.p.m. with peaks reaching 4.1 p.p.m. (Edling et al., 1988). In a study of the effects of formaldehyde on the mucous membranes and lungs of workers in a PB facility, Horvath et al., report that area and personal measurements taken in the plant during the study (supposedly in 1987 or 1988) had a median of 0.62 p.p.m. (Horvath et al., 1988). Herbert et al. studied the effects of formaldehyde on the respiratory system of workers in the OSB industry in Alberta. The authors took 21-h long measurements at different fixed stations in a facility using a PF resin. All 10 reported values were below 0.2 p.p.m. (Herbert et al., 1995). Imbus et al. report that the results of 15 measurements taken in an OSB facility using PF resin were all below 0.05 p.p.m. (Imbus and Tochilin, 1988).
The available literature provides little information about determinants of occupational exposure to formaldehyde. A few influential factors seem to modify formaldehyde emissions but their quantitative influence on occupational exposure is not documented. Moreover, there is a lack of recent data on exposure levels, most of the few reported exposure levels being prior to 1990. The present study aimed at documenting current formaldehyde exposure levels in Quebec in this industry and identifying their determinants.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Industrial hygiene surveys
The 12 plants manufacturing PB, MDF or OSB in Quebec were visited during the period from June 2001 to March 2002. The visits were conducted by a team of 2–4 industrial hygienists and technicians and lasted 1 to 2 days.
A standardized form was created to facilitate and systematize the information gathering process. One member of the research team completed the form either from observation of the workplace or from interview with employees. In order to describe industry-wide exposure trends, lists of standardized jobs and work zones in the plants were created (Table 1).
|
Time-weighted type monitoring was performed in each plant at fixed stations (area sampling) and in the respiratory zone of employees (personal sampling) by using adsorbent tubes coupled with sampling pumps. Employees and locations monitored were not chosen randomly but were selected with the aim to cover the greatest number of exposure circumstances in a facility given the available sampling resources. Although all jobs/areas in the facilities were sampled, more sampling resources were allocated to those thought to be associated with the highest exposures. Sampling was not task oriented but rather aimed at representing full-shift exposure.
Governmental data
In Quebec, all companies in certain regulated industries are visited by governmental occupational health teams that identify health hazards, evaluate health risks, define required medical surveillance and devise corrective measures with the employers and employees. The visits may be conducted several times in the same plant if periodical reassessment is deemed necessary. All but one of the plants identified as manufacturing PB, MDF or OSB panels in Quebec were visited by these teams, as early as 1984. During each visit by the research team, all governmental exposure data were collected from the corresponding local health center along with the available ancillary information.
Quebec legislation requires sampling strategies to be designed according to the recommendations of the Quebec Research Institute for Occupational Health and Safety (IRSST) (IRSST, 2000). While mentioning several types of strategies (e.g. ‘worst-case’, random sampling), the IRSST guide does not provide precise guidelines to be used by governmental health teams. Moreover, the purpose of sampling may differ from situation to situation, e.g. hazard identification, follow up, task monitoring, compliance monitoring or exposure profiling. Therefore, since the sampling strategies associated with the collected exposure data were not explicitly stated in the paper records, they are mostly unknown and may vary considerably across the exposure data.
Analytical methods
During the visits by the research team, formaldehyde concentrations were evaluated with sampling pumps coupled with solid sorbent tubes (type XAD-2) impregnated with 2-(hydroxymethyl) piperidine. Analysis was conducted by gas chromatography with Nitrogen Phosphorus detection (GC-NPD) (IRSST, 1995). Governmental teams used three different methods sequentially from 1984 to 2002. Before 1985, sampling was mostly performed with an impinger filled with a collecting solution containing dinitrophenylhydrazine and perchloric acid and the analysis was conducted with liquid chromatography (IRSST, 1985). From 1985 to 1995, the sampling device used was a solid sorbent tube (type ORBO 22) impregnated with n-benzylethanolamine while the analysis was conducted by GC-NPD (IRSST, 1988). Since 1995, the sampling and analytical methods have been those used by the research team.
Data formatting
Area and personal measurements were analysed separately, which was motivated by the fact that area measurements rarely represent personal exposures adequately and may be influenced by different determinants. Inside the Area and Personal datasets, data originating from the research team and governmental records were merged for analysis. This allowed for increased statistical power during the analysis. A variable identifying the source of data was included in the statistical modelling to explore potential systematic differences.
A limit of detection (LOD) was calculated according to the sampling volume using the lowest reported formaldehyde mass of each analytical method. Values reported under a LOD were treated according to the recommendations of Hornung and Reed, separately for the area and personal datasets (Hornung and Reed, 1990). This led to assigning the value of the LOD divided by 2 to the non detected area measurements and the value of the LOD divided by the square root of 2 to the non detected personal measurements.
Statistical analysis
Based on a graphical assessment of the frequency distributions of area and personal concentrations, the response variable selected for analysis was the natural logarithm of formaldehyde concentrations. The data were analysed with extended linear mixed-effects models. Linear mixed-effects models have recently been used successfully to analyse occupational exposure data (Teschke et al., 1994; Lagorio et al., 1997; Leena et al., 1999; Rappaport et al., 1999, 2003; Burstyn et al., 2000; Symanski et al., 2001; van Tongeren and Gardiner, 2001; Weaver et al., 2001; Egeghy et al., 2002; Peretz et al., 2002; Raaschou-Nielsen et al., 2002). Extended linear mixed-effects models, in addition to the possibility of modelling correlation patterns, allow exploring changes in the variability of the response as a function of other variables (Pinheiro and Bates, 2000). The model framework used in this study is described by the following equation:
 | (1) |
ln Cijk is the logarithm of the kth observation in the jth group of variable B in the ith group of variable A. The model assumptions are: (Random.effectA) and (Random.effectB) are distributed normally with mean 0; (Random.effectA), (Random.effectB) and (Error) are statistically independent; and (Error) follows a multinormal distribution with mean 0 and different possible variance-covariance structures.
All fixed effects tested for inclusion in the model are presented in Table 2. The year of sampling was tested as either a continuous variable or a nominal variable representing different time periods. The facility variable was tested as a random effect for the first level of grouping, with the sampling campaign as a random effect for the second level of grouping. The facility variable was tested as a random effect to determine the extent of similarity of exposures among different plants manufacturing the same product with the same process. The sampling campaign was tested as a random effect to explore potential correlation between measurements made in a small time window after controlling for the fixed effects. The correlation between measurements taken during the same campaign can be determined from the estimated inter-group and intra-group variances by the calculation of the intra-class correlation coefficient (ICC, equation 2).
 | (2) |
is the inter-group variance and 
is the intra-group or residual variance.
|
The variance structures tested for the error term were: a different residual variance
for each level of one of the available nominal variables (equation 3) and a residual standard deviation varying linearly (equation 4) or exponentially (equations 5 and 6) with one of the continuous variables. Because of the unbalanced nature of our datasets and the limited number of measurements, interactions were not tested in the models.
Different models are tested for the residual standard deviation (
):
 | (3) |
 | (4) |
 | (5) |
 | (6) |
The S-plus software provides two ways of modelling random effects, which yield a slightly different interpretation of the parameters when the residual error is not modelled as homoscedastic: the lme function provides an estimate of the inter-group variability (
in equation 2), therefore yielding a variable intra-group correlation (ICC) if the residual (or intra-group) variability (
in equation 2) has a heteroscedastic structure. The gls function provides an estimate of the intra-group correlation, therefore yielding a variable inter-group variability if the residual (or intra-group) variability has a heteroscedastic structure. The lme function was preferred in our study because it allows modelling of several levels of nested random effects.
REML optimization was used to choose the random effects and residual variance structures and estimate the final model parameters. ML optimization was used to compare models with different fixed effects structures (Pinheiro and Bates, 2000). Model building was performed by means of the following procedure: The best fixed effects model was first constructed with a forward stepwise routine using the Bayesian information criterion (BIC) as a discrimination criterion. Then the best random effect structure was added by comparing the BIC of the four possible models (no random effect, random effect A, random effect B, random effects A and B). Finally the variance structure for the residual error was assessed in a similar way. The next step consisted in retesting the fixed effects for removal or addition of variables. The random effects and variance structure were adjusted again if the fixed effect model had changed.
In order to illustrate the quantitative influence on exposures of the fixed effects coded as nominal variables, relative indices of exposure (RIE, equation 7) were calculated. Hence, for each variable, the category corresponding to the highest number of observations (the reference category) was assigned the value 100%. The RIE of each of the other categories of the variable was then calculated by finding the exponent of the difference between the estimated coefficient for that category and the one for the reference category.
 | (7) |
Internal validation was primarily conducted by graphical assessment of residuals and estimates of random effects regarding the assumptions underlying the estimation. There was no external validation of the final models.
All analyses were conducted with the statistical software S-Plus 6.1 professional edition for Windows Release 1 (Insightful Corp., Seattle, WA).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Information on the plants
Among the 12 plants visited, 6 manufactured OSB panels and had a median workforce of 120. Three plants manufactured PB panels, with a median workforce of 140. Three plants produced MDF panels, with a median workforce of 71. All plants except one OSB facility had exhaust ventilation systems extracting air above the press. The remaining plant had only a roof opening above the press. In all the plants most of the production jobs were on a 12-hour shift structure, with 3 different teams working alternately. No operation requiring the use of respiratory protection was monitored during the visits performed by the research team.
Data collection
The visits performed by the research team yielded 275 measurements while 590 were available from the governmental files. Fifty nine measurements were removed from this dataset prior to further analysis: 10 corresponding to a job performed in only one facility and for only a few months, 7 corresponding to task sampling or infrequent events, 6 samples considered ‘dubious’ by the technician, 1 sample taken directly above the press beside the exhaust system, 22 because they were labelled ‘other’ based on the job or work zone classification and 13 because the year of sampling was missing. In addition, 47 measurements were not included in the modelling datasets because they corresponded to jobs existing in only one or two of the three processes in the study. Inclusion of these data in the statistical modelling would have required nesting the variable job inside the variable process, which could not be achieved due to the unbalanced nature of our data. These 47 measurements were nevertheless included in the descriptive analysis. The variables documented for both sources of data are presented in Table 2.
Area measurements
Summary statistics along with the empirical distributions of area measurements stratified by their origin (research or governmental) are presented in Figure 1. The percentages of measurements below the LOD were 12 and 8 for the research measurements and the governmental data, respectively. The research and governmental measurements were taken, respectively, during 11 and 63 sampling campaigns. Concentrations reported by the research team covered a median duration of 4.9 h (interquartile interval: [2.4–5.9]). The corresponding values were 2.1 h [2.0–5.6] for the governmental data. Table 3 presents the GMs, geometric standard deviations (GSDs) and corresponding number of measurements stratified by work zone, process and origin of the data.
|
|
The final ‘whole dataset’ model includes the following fixed effects, which explains 57% of the total variance: Process, Work zone, Season of sampling, Origin of data and Year. The coefficients of all fixed effects of the model are presented in Table 4, along with their standard error. The estimated REIs associated with each work zone are presented in Figure 2 along with an
95% confidence interval (95% CI). The presented CIs correspond to the comparison of the levels of each variable with the reference category. Their exact interpretation is that if they exclude the 100% value, the exposure level associated with the category is significantly higher (or lower) than the exposure level associated with the reference category. The REIs corresponding to the other categorical variables are presented in Figure 3.
|
|
|
The best fit for the sampling year was obtained with a linear spline structure containing one knot, the line-break year being 1991. The other tested structures were a nominal variable with three levels (1984–1989, 1990–1994 and 1995–2002), a linear, and a quadratic trend (one 1st order term and one 2nd order term). The line-break year was determined by graphical assessment and refined by using the BIC. The estimated coefficient showed a yearly decrease of 19.7% until 1991 followed by a yearly increase of 9.4%.
A one-level random effect structure with the variable Sampling campaign as the random effect yielded the best fit compared with no random effect, Facility alone or Sampling campaign within Facility. The best residual variability structure in terms of BIC was one with different residual standard deviation for each work zone. With the Finishing zone as the reference (i.e. value 100%), the relative residual standard deviations estimated by the model for the other work zones are as follows: Main production (153%), Resin production-storage (136%), Storage-shipping (131%), Raw materials (206%), Other department (226%), Operator booths (both interior and exterior) (119%). The estimated within-sampling campaign residual standard deviation is 0.57 for the reference category Finishing zone while the between-sampling campaign standard deviation is 0.36, giving ICC values varying between 0.07 and 0.29, with an average of 0.17.
Graphical assessment of the distribution of the normalized residuals yielded satisfactory fit to the normal distribution. Likewise, the estimates of the random effects indicated graphical conformity to the normal distribution. No systematic trend was found during examination of the variations of residuals and estimated random effects stratified by the different levels of the other available variables, indicating satisfactory conformity to the independence hypothesis.
Personal measurements
Summary statistics of the personal measurements are presented in Figure 4. The percentages of measurements below the LOD were 23 and 9 for the research measurements and the governmental data, respectively. The research and governmental personal measurement were taken during 12 and 67 sampling campaigns, respectively. Concentrations reported by the research team covered a median duration of 4.8 h (interquartile interval: [3.3–6.0]). The corresponding value was 5.9 [4.4–8.0] for the governmental data. GMs, GSDs and corresponding number of measurements stratified by job, process and origin of the data can be found in Table 3.
|
The initial ‘whole dataset’ model included the following fixed effects, which explained 62% of the total variance: Process, Job, Season of sampling, Origin of data, Year and Measurement duration. Further examination of the coefficients for the different jobs led to the testing of the grouping of jobs into four exposure groups: group 1 includes Administration and Foreman, group 2 includes Laboratory technician, Maintenance worker and Cleaner, group 3 includes Press operator, Assistant press operator, Finisher, Shipper, and group 4 includes Floater and Press-miscellaneous tasks. The resulting model yielded a better fit than the original one in terms of BIC. The percentage of total variance explained by the fixed effects was marginally reduced (to 61%). The RIEs associated with each job group are presented in Figure 5 while the RIEs corresponding to the other nominal variables in the model can be found in Figure 3. The coefficients for the fixed effects of the reduced model can be found in Table 4.
|
The best fit for the sampling year was obtained with a linear spline structure containing one knot, the line-break year being 1995. The estimated trend corresponds to a yearly decrease of 7% until 1995 followed by a yearly increase of 12.2%. The estimated influence of the measurement duration on exposure levels corresponds to a decrease of 5.2% when the duration is increased by 60 min (95% CI: 2.1–8.2).
A one-level random effect structure with the variable Sampling campaign as the random effect yielded the best fit compared with Facility alone or Sampling campaign within Facility. The best residual variance structure in terms of BIC was one with different residual standard deviations for each process. With the process PB as the reference (i.e. value 100%), the relative residual standard deviations estimated by the model for the other processes are as follows: MDF (68%), OSB (63%). The estimated within-sampling campaign residual standard deviation (of the log-transformed concentrations) is 0.53 for the reference category while the between-sampling campaign standard deviation is 0.46, giving ICC values varying between 0.42 and 0.66, with an average of 0.56.
As for the area measurements, graphical assessment of the distribution of the normalized residuals and of the estimates of the random effects yielded satisfactory fit to the normal distribution and satisfactory conformity to the independence hypothesis.
Table 5 presents, side by side for Research and Government data, yearly GMs of area and personal levels stratified by work zone/exposure group and estimated for the year 2002 by the statistical models. For the personal measurements, the estimates were calculated for a duration equal to the median duration of the measurements (5.6 h).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Determinants of exposure to formaldehyde in the reconstituted wood panel industry
The percentages of total variability explained by the fixed effects of both personal and area models (61 and 57%, respectively) compare favourably with similar studies in the field of occupational exposure assessment (ranging between 20 and 70%) (Burstyn and Teschke, 1999
; van Tongeren and Gardiner, 2001; Bakke et al., 2002). Thus, strong determinants of exposure to formaldehyde in this industry have been identified. The estimated coefficients for the variable Process confirm, in both area and personal models, the lower potential for formaldehyde emissions when PF resins are used (all OSB facilities in our study used PF resins). However, while personal exposures tend to be similar for MDF and PB, MDF ambient levels seem higher than PB levels. This difference might be partly explained by the fact that measures to prevent worker exposure, such as reducing time spent in proximity to the press, might have been implemented to a greater extent in MDF. However, confounding by variables not accounted for in the modelling process cannot be ruled out as an explanation of this observation.
For both area and personal models, the origin of data was also a strong predictor of formaldehyde levels, research data being lower than governmental data by a factor of 3 (Figure 3). Several hypotheses may explain this observation also apparent in the empirical cumulative distributions shown in Figures 1 and 4. First, the research visits having been planned long in advance with the management of the plants and after several meetings with representatives of this industry, it is possible that ‘favourable’ production conditions were implemented on the days of the visits. This would cause a negative bias compared to real exposure conditions. While Olsen et al., among others, have mentioned this source of bias as a possibility, it has, to our knowledge, not been further discussed in the published literature (Olsen et al., 1991). A second explanation is confounding by the temporal variable in the models. All research measurements were taken in 2001–2002 while some governmental data date back to 1984. This points to a potential collinearity issue between the two variables. This possibility was further explored in two different ways. First, the area and personal models were fitted to the data restricted to years common to both research and governemental datasets (2001 for the area measurements and 2001 and 2002 for the personal measurements). For the area analysis, with 109 observations (of which 26 were governmental), the estimated relative exposure index for the research data (the governmental data constitutes the reference category) was 47% compared to 34% for the initial model. For the personal analysis, with 138 observations (of which 24 were governmental), the estimated relative exposure index for the research data was 45% compared to 35% for the initial model. These results support the presence, albeit limited, of some confounding by the temporal trends. Furthermore, the full models were fitted without the ‘origin of data’ variable to datasets restricted to governmental data in order to assess the robustness of the observed temporal trend. For the area analysis, with 341 observations, the coefficients for (year-1984) and [max(year-1984,11)] (see Table 4) were, respectively, –0.17 (–0.22 for the initial model) and 0.25 (0.31 for the initial model). For the personal analysis, with 171 observations, the coefficients for [year-1984] and [max(year-1984,7)] were, respectively, –0.10 (–0.07 for the initial model) and 0.23 (0.29 for the initial model). These results tend to show that a temporal trend actually exists independently of the origin of the data, and that the estimated trend is close to that observed in the global models. Thus, although a degree of overestimation of the difference between research and governmental data may exist because of confounding, its extent should be small. We believe that, because of the limited resources that can be devoted to sampling during industrial visits, the hygienists tend to monitor worst-case scenarios in order to optimize the interpretation of their results. Since, by devoting more sampling resources to monitor supposedly ‘high’ exposure jobs or area, the strategy used by the researchers might be regarded as biased if jobs or work-zones are not controlled for, it must be concluded that within the jobs and area classification used in this study, governmental teams tended to sample tasks or sub-areas specifically associated with high formaldehyde concentrations.
The ‘worst-case’ bias in data from governmental sources, often associated with compliance monitoring, has already been mentioned in the literature (Stewart and Rice, 1990; Vinzents et al., 1995). Olsen et al., reported an actual comparison of measurements taken with a random sampling strategy with measurements existing in a governmental database. The authors found that the governmental data were higher than the ‘random’ data by a factor ranging from 5 to 10, which is compatible with our interpretation. On the other hand, a variable identifying the sampling strategy as either ‘research’ or ‘compliance’ did not improve the fit of a multiple regression model applied to softwood dust levels in British Columbia lumber mills (Hall et al., 2002).
It might be argued that research and governmental data should not be analysed together because they may not be influenced by the same determinants. In order to explore that possibility, both area and personal models were fitted to a restricted dataset to allow the testing of 1st order interactions between Origin of data and the other fixed effects, and of a possible difference of residual variance structure for the research and governmental data. With respectively 392 and 236 data available for analysis for the area and personal measurements, none of the tested additional models improved the fit compared to the original ones. In addition to the increased statistical power and the fact that the research and governmental measurements were taken in the same workplaces, we believe that the merged analysis of both types of data was justified in our case.
A significant time trend, best modelled by a one knot linear spline, existed both for the area and personal measurements. The use of linear splines to model irregular temporal trends in occupational exposures has already been reported (Raaschou-Nielsen et al., 2002; Friesen et al., 2003). Although the spline knot year was different for the personal and area measurements (1995 and 1991), both estimated trends show high exposures at the beginning of the study period, decreasing until 1991–1995 and then increasing, although moderately, until the end of the study period (2002). Both the nominal and quadratic coding of the year of sampling yielded the same temporal pattern. This pattern differs from the reported generic occupational exposure decrease over time reported by Symanski et al. (Symansky et al., 1998, 2001). A possible explanation might be the documented important increase in production in that industry in North America in the middle of the 90s (Spelter, 1997).
The other influent variable common to both the area and personal models is the season of sampling. Thus, winter, and, to a lesser extent, autumn, are associated with higher exposures than summer and spring. From direct observation and interviews with the employees of the visited plants, it appears that during the cold season (in Quebec temperatures are commonly below –15°C during this period) the ratio of recirculated air to fresh outside air is increased because of associated heating costs. Thus, the facilities tend to be in a negative pressure relative to the outside, which in turn causes the exhaust systems to lose efficiency. Furthermore, most doors and windows are open during the hot season, improving the air replacement rates. van Tongeren and Gardiner have reported lower exposure levels during summer for some job titles in the carbon black manufacturing industry (van Tongeren and Gardiner, 2001).
For the area model, the work zone had a significant influence on exposure levels. The main production area is associated with the highest exposures, with levels decreasing as one gets further from the press to the finishing, and then the shipping areas. As expected, departments separated from the main production area are associated with low formaldehyde levels. The results shown in Figure 2 also demonstrate the importance of supplying operator booths with outside air. Thus, the exposure levels inside operator booths ventilated with air from the plant are comparable to those found outside the booths (most operator booths with such ventilation were in the finishing zone).
Regarding the personal measurements, the modelling of the variable ‘job’ allowed the identification of four similar exposure groups, presented hereafter from the least to the most exposed: group 1 includes employees spending most of their time outside the production zone (e.g. foreman), group 2 includes employees spending part of their shift in the production zone (e.g. mechanics), group 3 includes workers spending their whole shift in the production zone but with a variable proportion of time spent in operator booths (e.g. press operator) and group 4 includes workers spending their whole shift in the production zone unprotected by operator booths (e.g. floater).
Higher personal exposure levels were also associated with shorter measurement duration. This trend was not modified when fitting the model separately to research and governmental data. It may be partly explained by the fact that longer measurements may include unexposed periods, such as breaks (Raaschou-Nielsen et al., 2002). Since there was no significant interaction between duration and the source of data for the personal measurements, together with the absence of any influence in the case of area measurements, we conclude that the upward bias found in governmental data is not due to merely shorter measurement durations associated with a task-based strategy.
The significance of the variables identifying work-zones and jobs in both models underlines the usefulness of the standardized lists created during this study. They appeared as potentially strong predictors of exposure to formaldehyde, based on technical literature and field observation: this was validated and refined by the statistical modelling. Consequently, their use should be explored in other studies of formaldehyde exposure in the same industry.
The determinants identified in our study will be useful from the industrial hygiene standpoint for devising sampling strategies in this industrial sector, for example to help identify a priori potential overexposure situations. Within the framework of epidemiology, the identification of time trends is important for retrospective exposure assessment, while determinants, such as jobs or process can help in the elaboration of specialized questionnaires in population-based case-control studies.
Structures of variance-covariance of exposure data
Random effects models have recently been used to explore patterns of correlation among occupational exposure measurements. However, the studies have mostly focused on modelling intra- and inter-worker variability (Kromhout et al., 1993; Rappaport et al., 1999, 2003; Burstyn et al., 2000; van Tongeren and Gardiner, 2001). In our study, information on the identity of workers was not available in the majority of governmental files. Moreover, the sampling campaigns performed by the research team, lasting 2 days at the most, did not allow resampling the same workers because of the rotating teams work structure. Failing to model intra- and inter-worker variability constitutes a limitation of the present study in that this hampers rigorous conclusions about the long-term risks posed to workers in this industrial sector. In particular we can not estimate probabilities of overexposure of a random worker as described by Rappaport et al. (Rappaport et al., 1995; Lyles et al., 1997). However the estimated yearly GMs that we present in Table 5 allow drawing a general picture of exposure levels in this industrial sector, together with our results on the determinants of exposure.
Although the variable Facility was tested as a random effect, the results of both area and personal models show no significant intra-facility correlation, indicating similarity of exposure levels across facilities when all other variables are taken into account. This result is plausible since during the industrial hygiene visits by the research team, it was observed that most facilities were quite similar in terms of architecture (the press being the main emission source inside the plant, contaminating other areas), machinery used and exposure control measures.
The variable sampling campaign improved the fit significantly when tested as a random effect. This corresponds to the fact that after controlling for all fixed effects in the models, there are systematic differences between formaldehyde levels measured during different campaigns, and is equivalent to the existence of correlation among measurements taken during the same campaign. The average intra-sampling campaign correlation coefficients estimated in our study were 0.17 and 0.56, respectively for the area and personal measurements. Teschke et al. report a coefficient of correlation of 0.31 between wood dust levels measured during the same inspection of data taken from OSHA's occupational exposure database (Teschke et al., 1999). While some authors have reported evidence of correlation between shift-long measurements taken on consecutive days (Buringh and Lanting, 1991; Symanski and Rappaport, 1994; Deadman et al., 1996), some others observed no evidence of such correlation (Francis et al., 1989; George et al., 1995). Our results confirm the presence of correlation between shift-long exposures measured in short time periods in this industrial sector, implying a potential for underestimation of the day-to-day exposure variability when an assessment is based only on one sampling campaign conducted over a few consecutive days.
Significant structures of heteroscedasticity of the error term were found in both area and personal models. Hence, personal exposure appeared more variable for the PB process than for MDF and OSB. Moreover, ambient formaldehyde levels from the work zones Main production, Department other than production, and Raw material reception were much more variable than those measured in other locations in the facilities. The variability of exposure levels determines the number of samples necessary to assess an exposure situation with adequate precision. Furthermore, with regard to statistical modelling, estimates of other parameters in the model depend on the variance-covariance structures in the case of unbalanced data, which is present in most datasets in this field of research. Therefore it appears important to take into account and explore such structures of variability when modelling occupational exposures (Pinheiro and Bates, 2000).
Formaldehyde exposure levels in the reconstituted wood panel industry in Quebec
The observed (Table 3) and estimated (Table 5) levels of exposure to formaldehyde in this industry are consistent with the few results reported in the most recent literature (posterior to 1985). In particular, the exposure levels reported by Zimowski show very close agreement with those measured in Quebec. Exposures in the OSB manufacturing industry are mostly below 0.1 p.p.m. with some work zones/jobs associated with somewhat higher levels according to the governmental data. Exposure levels in the MDF and PB manufacturing industries are similar, with observed and estimated GMs of personal and area levels between 0.1 and 0.4 p.p.m. depending on the source of data. The highest estimated ambient GM is 0.43 p.p.m. in the main production area of MDF manufacturing facilities (this estimate is increased to 1.28 p.p.m. when governmental data are used). The highest personal estimated GM is 0.22 p.p.m. for workers in close proximity to the press in the PB process (this estimate is increased to 0.62 p.p.m. if the governmental data are considered). Levels reported prior to 1985 are consistently higher than those in our database. This maybe explained by the generalized implementation, around 1985, of low formaldehyde emission resins (OSHA, 1987).
Area measurements are consistently higher than personal measurements. This is consistent with the fact that the most exposed workers are those who spend the most time in the production area unprotected by ventilated booths. While none of the personal measurements in this study was greater than 2 p.p.m. (the current Quebec ceiling OEL), 4 and 6% of the research and governmental ambient concentrations, respectively, were greater than this value.
Globally, our results point to generally low personal full-shift exposure to formaldehyde in this industrial sector. However, the potential for short-term high exposures associated with specific and occasional tasks cannot be ruled out on the basis of our study. Several OELs exist for formaldehyde, varying both in type and level. In the US, OSHA enforces an 8 h TWA limit of 0.75 p.p.m. with a short-term exposure limit (STEL) of 2 p.p.m. The ACGIH recommends a ceiling limit of 0.3 p.p.m. (ACGIH, 2003).
Validity of the statistical models
The internal validity of our models appears satisfactory considering the results of the graphical assessments of residuals and estimates of random effects described earlier. In addition, bootstrapping the model estimates 1000 times (results not shown) resulted in the observation that among 29 coefficient estimates for the two models the absolute relative difference between the model and the bootstrap estimate was >5% only in six cases, with a maximum of 17%. The sign of the difference between the two estimates was negative almost the same number of times as it was positive. The standard errors of the bootstrap estimates were on average 30% smaller than the corresponding model estimates, indicating moderate overestimation of the widths of the confidence intervals presented in Figures 2, 3 and 5. These results support reliance on the asymptotic assumptions linked to the estimation by ML or REML of confidence intervals for the model parameters (Harrel, 2001).
Although there was no formal external validation of the models, several elements suggest that our results may be applicable outside the restricted scope of our dataset. Thus, as seen in Figure 3, the estimates of the effects of variables common to the area and personal measurements are similar. Moreover, our estimates of the influence of several potential determinants of exposure are consistent with published observations. Finally, the exposure levels estimated by the models are similar to those reported in the most recent literature. These observations, while not constituting an external validation per se, provide some insight into the potential for generalization of our results. However, the presence in our dataset of a bias not accounted for can not entirely be excluded since the data were not generated through a randomized sampling process.
|
CONCLUSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Through statistical modelling of area and personal exposure measurements performed in the reconstituted wood panel industry in Quebec, several determinants of exposure to formaldehyde in this industrial sector were identified. The MDF and PB processes were associated with high exposure levels compared to OSB. Higher exposures also occurred during winter conditions compared to other seasons. While plant was not a strong predictor of exposure levels, work zones and jobs were strongly associated with area and personal measured concentrations, respectively. The highest levels were measured in areas close to the press and on workers spending most of their time in the press area. Moderate historical variations in exposure levels were also identified, best modelled by a one-knot linear spline. Governmental measurements were consistently found higher than those measured by the research team, pointing to the probable existence of a ‘worst-case’ bias in governmental data. The use of extended linear mixed-effects models allowed the identification of a moderate correlation between measurements taken during the same sampling campaign. Significant heteroscedasticity structures of the error term were also identified during modelling, stressing the need to take them into account in similar studies to reduce bias and error in the estimation of other model parameters. The measured and estimated time-weighted average levels of exposure to formaldehyde in this sector can be considered low compared to the 8 h-TWA OELs of most jurisdictions (e.g. 0.75 p.p.m.) but close to the most demanding ones (e.g. 0.3 p.p.m.). The possibility of higher short-term exposures can not be ruled out. The successful identification of several determinants of exposure to formaldehyde in the reconstituted wood panel industry will allow for better sampling strategies in this industrial sector. Furthermore, these determinants may be used in future epidemiological studies to improve prospective and retrospective exposure assessments.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
The authors would like to thank Denis Bégin for his help in the interpretation and identification of the published literature relevant to the reconstituted wood panel industrial sector. We would also like to thank Jan-Erik Deadman for his useful and much appreciated comments on the present manuscript. This research project was funded by the Quebec Research Institute for Occupational Health and Safety (IRSST, grant number 099-011 and scholarship for J.L.).
Received October 7, 2004; in final form February 23, 2005
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
ACGIH. (2003) TLVs and BEIs threshold limit values for chemical substances and physical agents/biological exposure indices. Cincinnati, OH: American Conference of Governmental Industrial Hygienists.
Bakke B, Stewart P, Eduard W. (2002). Determinants of dust exposure in tunnel construction work. Appl Occup Environ Hyg; 17: 783–96.[CrossRef][Medline]
Buringh E, Lanting R. (1991). Exposure variability in the workplace: its implications for the assessment of compliance. Am Ind Hyg Assoc J; 52: 6–13.[Web of Science][Medline]
Burstyn I, Teschke K. (1999) Studying the determinants of exposure: a review of methods. Am Ind Hyg Assoc J; 60: 57–72.[Web of Science][Medline]
Burstyn I, Kromhout H, Kauppinen TP 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.
Deadman JE, Armstrong BG, Thériault GP. (1996) Exposure to 60-Hz magnetic and electric fields at a Canadian electric utility. Scand J Work Environ Health; 22: 415–24.[Web of Science][Medline]
Edling C, Hellquist H, Oedkvist L. (1988) Occupational exposure to formaldehyde and histopathological changes in the nasal mucosa. Br J Ind Med; 45: 761–5.[Web of Science][Medline]
Egeghy P, Nylander-French L, Gwin KK et al. (2002) Self-collected breath sampling for monitoring low-level benzene exposures among automobile mechanics. Ann Occup Hyg; 46: 489–500.
Francis M, Selvin S, Spear R et al. (1989) The effect of autocorrelation on the estimation of workers' daily exposures. Am Ind Hyg Assoc J; 50: 37–43.[Web of Science][Medline]
Friesen MC, Demers P, Spinelli J et al. (2003) From expert-based to quantitatve exposure assessment: Updating a job exposure matrix at a Söderberberg aluminum smelter. In: CARWH symposium; 26 October 2003, Montreal: Canadian association for research on work and health. p. 45.
George DK, Flynn MR, harris RL. (1995) Autocorrelation of interday exposures at an automobile assembly plant. Am Ind Hyg Assoc J; 56: 1187–94.
Goyer N, Perrault G, Beaudry C et al. (2004) Impact d'un abaissement de la valeur d'exposition admissible au formaldéhyde. Montréal: Institut de recherche Robert-Sauvé en santé et en sécurité du travail (R-386).
Hall A, Teschke K, Davies H et al. (2002) Exposure levels and determinants of softwood dust exposures in BC Lumber Mills, 1981–1997. Am Ind Hyg Assoc J; 63: 709–14.
Harrel FEJ. (2001) Regression modeling strategies—with applications to linear models, logistic regression, and survival analysis. New York: Springer.
Herbert FA, Hessel PA, Melenka LS et al. (1995) Pulmonary effects of simultaneous exposures to MDI, formaldehyde and wood dust on workers in an oriented strand board plant. J Occup Env Med; 37: 461–5.[Web of Science][Medline]
Hornung R, Reed LD. (1990) Estimation of average concentration in the presence of nondetectable values. Appl Occup Environ Hyg; 5: 46–51.
Horvath EP, Anderson H, Pierce WE. (1988) Effects of formaldehyde on the mucous membranes and lungs. A study of an industral population. J Am Med Assoc; 259: 701–7.
Imbus HR, Tochilin SJ. (1988) Acute effect upon pulmonary function of low level exposure to phenol-formaldehyde-resin-coated wood. Am Ind Hyg Assoc J; 49: 434–7.[Web of Science][Medline]
International Agency for Research on Cancer. (1995) IARC monographs on the evaluation of carcinogenic risks to humans vol. 62: wood dust and formaldehyde. Lyon: World Health Organization.
International Agency for Research on Cancer. (2005) IARC Monograph on the evaluation of carcinogenic risks to humans vol. 88: formaldehyde, 2-butoxyethanol and 1-tert-butoxy-2-propanol. Lyon: World Health Organization. (in press).
IRSST. (1985) Guide d'échantillonnage des contaminants de l'air en milieu de travail—Aldehyde formique. Montréal, QC: Institut de recherche en santé et en sécurité du travail.
IRSST. (1988) Guide d'échantillonnage des contaminants de l'air en milieu de travail—Methode 216-1—Aldehyde formique. Montréal, QC: Institut de recherche en santé et en sécurité du travail.
IRSST. (1995) Analyse du formaldehyde dans l'air—Méthode 295-1. Montréal, QC: Institut de recherche Robert-Sauvé en santé et en sécurité du travail.
IRSST. (2000) Sampling guide for air contaminants in the workplace—7th edition revised and updated. Montreal, QC: Research Istitute for Occupational Health and Safety (IRSST).
Kauppinen TP, Niemelä RI. (1985) Occupational exposure to chemical agents in the particleboard industry. Scand J Work Environ Health; 11: 357–63.[Web of Science][Medline]
Kromhout H, Symansky 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.
Lagorio S, Iavarone I, Iacovella N et al. (1997) Variability of benzene exposure amonf filling station attendants. Occup Hyg; 4:15–30.
Leena A, Nylander-French L, Kupper LL et al. (1999) An investigation of factors contributiong to styrene and styren-7,8–oxide exposures in the reinforced-plastics industry. Ann Occup Hyg; 43: 99–109.
Lyles RH, Kupper LL, Rappaport SM. (1997) A lognormal distribution-based exposure assessment method for unbalanced data. Ann Occup Hyg; 41: 63–76.
Niemelä RI, Priha E, Heikkila P. (1997) Trends of formaldehyde exposure in industries. Occup Hyg; 4: 31–46.
Niemelä RI, Vainio H. (1981) Formaldehyde exposure in work and the general environment. Scand J Work Environ Health; 7: 95–100.[Web of Science][Medline]
Olsen E, Laursen B, Vinzents PS. (1991) Bias and random errors in historical data of exposure to organic solvents. Am Ind Hyg Assoc J; 52: 204–11.[Web of Science][Medline]
OSHA. (1987) Regulatory impact and regulatory flexibility analysis of the formaldehyde standard. Washington, DC: United States Departement of Labor, Occupational Safety and Health Administration (Docket No. 225B. Exhibit No. 206).
Peretz C, Goren A, Smid T et al. (2002) Application of mixed-effects models for exposure assessment. Ann Occup Hyg; 46: 69–77.
Pinheiro JC, Bates DM. (2000) Mixed-effects models in S and S-plus. New York: Springer-Verlag.
Raaschou-Nielsen O, Hansen J et al., (2002) Exposure of Danish workers to trichloroethylene, 1947–1989. Appl Occup Environ Hyg; 17: 693–703.[CrossRef][Medline]
Rappaport SM, Goldberg M, Herrick RF. (2003) Excessive exposure to silica in the US construction industry. Ann Occup Hyg; 47: 111–22.
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 MA, 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.
Spelter HN. (1997) Capacity, production and manufacture of wood-based panels in North America. Madison, WI: United States Department of Agriculture, Forest Service, Forest Products Laboratory.
Stewart PA, Rice C. (1990) A source of exposure data for occupational epidemiology studies. Appl Occup Environ Hyg; 5: 359–63.
Symanski E, Chan W, Chang C-C. (2001) Mixed-effects models for the evaluation of long-term trends in exposure levels with an example from the nickel industry. Ann Occup Hyg; 45: 71–81.
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: 300–09.
Symanski E, Rappaport SM. (1994) An investigation of the dependence of exposure variability on the interval between measurements. Ann Occup Hyg; 38: 361–72.
Teschke K, Hertzman C, Morrison B. (1994) Level and distribution of employee exposures to total and respirable wood dust in two Canadian sawmills. Am Ind Hyg Assoc J; 55: 245–50.[Web of Science][Medline]
Teschke K, Marion SA, Vaughan TL et al. (1999) Exposure to wood dust in US industries and occupation, 1979 to 1997. Am J Ind Med; 35: 581–9.[CrossRef][Web of Science][Medline]
Tohmura SI, Inoue A, Sahari SH. (2001) Influence of the melamine content in melamine-urea-formaldehyde resins on formaldehyde emissions and cured resin structure. J Wood Sci; 47: 451–7.[CrossRef]
US Census Bureau. (2002) North American Industry Classification System (NAICS). Washington, DC: United States Census Bureau.
USEPA. (1998) Compilation of Air Pollutant Emission Factors AP-42 Volume 1, Fifth edition. Research Triangle Park, NC: US Environmental Protection Agency, Office of Air Quality Planning and Standards.
van Tongeren M, Gardiner KG. (2001) Determinants of inhalable dust exposure in the European carbon black manufacturing industry. Appl Occup Environ Hyg; 16: 237–45.[CrossRef][Medline]
Vinzents PS, Carton B, Fjeldstad P et al. (1995) Comparison of exposure mesurements stored in European databases on occupational air pollutants and definitions of core information. Appl Occup Environ Hyg; 10: 351–4.
Weaver MA, Kupper LL, Taylor D et al. (2001) Simultaneous assessment of occupational exposures from multiple worker groups. Ann Occup Hyg; 45: 525–42.
Wolcott JJ, Motter WK, Daisy NK et al. (1996) Investigation of variables affecting hot-press formaldehyde and methanol emissions during laboratory production of urea-formaldehyde-bonded particleboard. Forest Products J; 46: 62–8.
Zimowski EF. (1986) Final report of the OSHA health response team on the wood product industry. Washington, DC: United States Department of Labor, Occupational Safety and Health Administration (Docket No. 225B Exhibit No. 206).
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  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]
|
|
|
|
|
|
J. LAVOUE, R. VINCENT, and M. GERIN Statistical Modelling of Formaldehyde Occupational Exposure Levels in French Industries, 1986-2003 Ann. Hyg., April 1, 2006; 50(3): 305 - 321. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||