Annals of Occupational Hygiene Advance Access published online on July 10, 2008
Annals of Occupational Hygiene, doi:10.1093/annhyg/men033
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Stoffenmanager Exposure Model: Development of a Quantitative Algorithm
1 Business Unit Food & Chemical Risk Analysis, TNO Quality of Life, PO Box 360, 3700 AJ Zeist, The Netherlands
2 Arbo Unie, Expert Centre for Chemical Risk Management, Nijmegen, The Netherlands
* Author to whom correspondence should be addressed. Tel: +31-30-694-4990; fax: +31-30-694-4070; e-mail: Erik.Tielemans{at}tno.nl
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
In The Netherlands, the web-based tool called ‘Stoffenmanager’ was initially developed to assist small- and medium-sized enterprises to prioritize and control risks of handling chemical products in their workplaces. The aim of the present study was to explore the accuracy of the Stoffenmanager exposure algorithm. This was done by comparing its semi-quantitative exposure rankings for specific substances with exposure measurements collected from several occupational settings to derive a quantitative exposure algorithm. Exposure data were collected using two strategies. First, we conducted seven surveys specifically for validation of the Stoffenmanager. Second, existing occupational exposure data sets were collected from various sources. This resulted in 378 and 320 measurements for solid and liquid scenarios, respectively. The Spearman correlation coefficients between Stoffenmanager scores and exposure measurements appeared to be good for handling solids (rs = 0.80, N = 378, P < 0.0001) and liquid scenarios (rs = 0.83, N = 320, P < 0.0001). However, the correlation for liquid scenarios appeared to be lower when calculated separately for sets of volatile substances with a vapour pressure >10 Pa (rs = 0.56, N = 104, P < 0.0001) and non-volatile substances with a vapour pressure
10 Pa (rs = 0.53, N = 216, P < 0.0001). The mixed-effect regression models with natural log-transformed Stoffenmanager scores as independent parameter explained a substantial part of the total exposure variability (52% for solid scenarios and 76% for liquid scenarios). Notwithstanding the good correlation, the data show substantial variability in exposure measurements given a certain Stoffenmanager score. The overall performance increases our confidence in the use of the Stoffenmanager as a generic tool for risk assessment. The mixed-effect regression models presented in this paper may be used for assessment of so-called reasonable worst case exposures. This evaluation is considered as an ongoing process and when more good quality data become available, the analyses described in this paper will be expanded. Based on these analyses, the algorithm will be refined in the near future.
exposure assessment methodology • risk assessment
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
Important drivers of the development of generic and user-friendly approaches for assessment of workplace health risks are the introduction of the Chemical Agents Directive (European Commission, 1998) and, more recently, the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH) legislation in Europe (European Commission, 2006). As exposure is a complex process and varies enormously between workers and over time (Kromhout et al., 1993), the assessment of chemical risks requires a logical strategy or tool to focus resources on those situations with the greatest potential for adverse health effects (Mulhausen and Damiano, 1998). Currently, a vast range of screening tools exists that are intended to systematically address workplace chemical risks (Money, 2003). The COSHH Essentials system (Maidment, 1998; Russell et al., 1998; Garrod and Rajan-Sithamparanadarajah, 2003) and the ECETOC TRA (ECETOC, 2004) are among the most prominent and accepted examples for chemical exposure. Comparable tools are also available for pharmaceutical active ingredients (Naumann et al., 1996). Some of the tools (e.g. COSHH Essentials) have been primarily developed for providing assistance to small- and medium-sized enterprises (SME) with respect to workplace risk assessment and control, whereas others (e.g. ECETOC TRA) are specifically developed for the regulatory risk assessment process. Available screening models for chemical exposure have recently been reviewed in the context of guidance setting for REACH (http://ecb.jrc.it/home.php?contenu=/document/reach/rip-find-reports/rip-3.2-1-CSA-CSR).
The core requirements of any screening tool should be that it is simple, readily understood and with an appropriate level of conservatism (Tielemans et al., 2007). In general, one of the main weaknesses of the available screening tools is that only few have been properly validated. This prohibits a comprehensive evaluation and weighing of the available tools. Tools that are, at least to some extent, validated are COSHH Essentials (Tischer et al., 2003; Jones and Nicas, 2006a,b; Money et al., 2006), ECETOC TRA (ECETOC, 2004) and EASE (Bredendiek-Kämper, 2001; Cherrie and Hughson, 2005; Creely et al., 2005; Hughson and Cherrie, 2005; Johnston et al., 2005). In the near future, insight into accuracy of models should substantially grow in order to make transparent decisions concerning the selection of sound screening tools. This may also result in the selection of several complementary tools, each with a distinct validity domain.
In The Netherlands, the web-based tool called ‘Stoffenmanager’ was initially developed to assist SMEs to prioritize and control risks of handling chemical products in their workplaces. The background and underlying assumptions of the Stoffenmanager are described by Marquart et al. (2008). The rationale of the underlying exposure algorithm is based on works of Cherrie et al. (1996) and Cherrie and Schneider (1999) but is adapted in several ways. The model uses process information, physicochemical characteristics and mass balances to give a relative ranking of exposure situations. To guarantee a sound risk assessment process and further acceptance of the Stoffenmanager, a comprehensive evaluation of its underlying exposure algorithm is highly desirable.
The aim of the present study was to explore the accuracy of the Stoffenmanager exposure algorithm. This was done by comparing its semi-quantitative exposure rankings for specific substances with exposure measurements collected from several occupational settings to derive a quantitative exposure algorithm. Mixed-effect models were used to evaluate the predictive value of Stoffenmanager scores and to quantify the level of uncertainty in the algorithm.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
Outline of Stoffenmanager exposure algorithm
The Stoffenmanager exposure algorithm has been described elsewhere by Marquart et al. (2008). For description of specific parameters and classes within parameters, we refer to that paper. The exposure algorithm is based on a source–receptor approach developed by Cherrie and Schneider (1999) and incorporates modifying factors related to source emission and dispersion of contaminants. Most parameters are divided into classes with scores on a logarithmic scale, i.e. ranging from 0 through 0.03, 0.1, 0.3, 1, 3 to 10. These weighing factors can be derived from tables as described by Marquart et al. (2008). The volatility is the only parameter that is assumed to be linearly related to exposure and is expressed on a continuous scale. The total personal exposure score (Ct) is the sum of exposure levels due to near field (NF) sources (Cnf), far field (FF) sources (Cff) and diffusive sources (Cds), adjusted for possible use of control measures at the worker such as a control room (
imm):
 | (1) |
Exposure due to NF sources (Cnf) is a multiplicative function of type of handling of the product (H), intrinsic emission of the product (E), local control measures (lc) and general ventilation in combination with room size (gv_nf). A source is considered to be in the NF if it is located within 1 m of the head of the worker; the FF comprises the remainder of the room. The scores for handling are related to a number of characteristics such as energy transfer by a process that causes a product to become airborne and the scale of use. Intrinsic emission is a parameter that relates to vapour pressure of liquids and dustiness of powders. The Stoffenmanager incorporates various local control measures such as containment of the source, local exhaust ventilation (LEV) and reduction of dust exposure due to wetting. Mixing and dilution of contaminants in workroom air is taken into account by general ventilation in conjunction with room size (Cherrie, 1999). Exposure due to NF sources is expressed as follows:
 | (2) |
Exposure due to FF sources (Cff) is described according to a similar multiplicative function:
 | (3) |
Note that for the FF source, if present, the same intrinsic emission, handling and local control measures are assumed as for the NF source. The impact of general ventilation in combination with room size is different for NF and FF sources.
The diffusive source (Cds) representing background concentration is expressed as follows:
 | (4) |
The intrinsic emission for liquids is the only continuous parameter in the Stoffenmanager and is expressed as follows:
 | (5) |
10 Pa are assigned the same minimum score for Pi (i.e. 10), whereas substances with a vapour pressure 
30 000 Pa are assigned the same maximum score for Pi (i.e. 30 000). In order to predict exposure to a group of substances (e.g. n volatile organic compounds or n isocyanates), we used the following intrinsic emission equation:
 | (6) |
Collation of exposure data
Exposure data were collected using two strategies. First, we conducted seven surveys specifically for validation of the Stoffenmanager. Sector and companies were selected from a network of industry participating in the VASt program. The VASt program is established by the Dutch Ministry of Social Affairs and Employment to assist SMEs in reinforcing the working condition policy on hazardous substances (http://vast.szw.nl). In total, 63 companies in seven different sectors were recruited. All participating workers were experienced professionals who preformed their work as normal. For scenarios describing the handling of solids, we used inhalable dust measurements for comparison with Stoffenmanager scores. Respirable dust measurements were considered to be outside the scope of the presented validation study and should be dealt with in a later stage. Inhalable dust measurements were conducted in the animal feed industry, construction industry, textile industry and bakeries and flour handling industry. Personal air measurements were obtained from a random sample of potentially exposed workers in the companies. The dust samples were collected using a portable pump with a flow rate of 2 l min–1 and a Teflon filter mounted in a PAS6 sampling head. Sampling was performed in the breathing zone of the worker for 
4 h. Dust levels were determined by weighing the filter in a climate-controlled weighing room where the filters were conditioned for 24 h prior to weighing. The limit of detection (LOD) was assessed as the average weight difference of the blank filters plus three times the standard deviation.
For scenarios describing the handling of liquids, task-based measurements to solvents were conducted in auto body repair shops, printing industry and metal industry. Inhalation exposure to solvents during a specific task was assessed by personal air sampling using an air sampling pump (flow rate of 250 ml min–1) and charcoal adsorption tubes. Samples were transported to an external laboratory. After extraction with CS2, the samples were analysed on a broad range of organic solvents (250), using gas chromatography-flame ionization detection.
Occupational hygienists conducted all surveys using a checklist to collect information in a structured way. Workers were followed throughout their measurement period and information on tasks performed was registered. This checklist allowed the hygienist to record frequency and duration of tasks conducted and the relevant Stoffenmanager parameters for each task. Information on substances and their concentrations in a mixture were retrieved from safety data sheets (SDSs) available at the workplace. In those cases that the concentration was given in ranges, the midpoint of the range was used in the analyses.
Secondly, occupational exposure data sets were collected from archives of TNO. These data originated from research projects funded by the Dutch Government in the past years. For details on the methodology, we refer to the individual publications (Vreede de et al., 1994; Vreede de and Amelsfort van, 1997a,b; Jong de et al., 1998; Marquart et al., 1999; Preller and Schipper, 1999; de Cock and Drooge, 2002; Links et al., 2002; Brouwer et al., 2006; Pronk et al., 2006; Links et al., 2007). In addition, a network of industry and occupational health services participating in the Dutch ‘VASt program’ was used to collect more exposure data. In the context of this program, a large number of research and consultancy projects has been conducted and funded (partly) by the Dutch Government. We used the momentum of this project to collate exposure data. The data collection process was facilitated by a request for data on the ‘VASt’ website. In addition, a specific newsletter concerning the evaluation study was sent to contact persons of various sectors and companies. Both task-based and shift-based exposure measurements were collected. All data reflect personal exposure measurements. Table 1 shows the number of measurements available from each data source and separate for handling of solids and liquids.
|
Evaluation of data quality and assignment of Stoffenmanager scores
Based on the contextual information, Stoffenmanager scores were assigned by one occupational hygienist. Subsequently, these scores were reviewed by another occupational hygienist. Both occupational hygienists were involved in the development of the Stoffenmanager exposure algorithm. In case of inconsistencies between the two occupational hygienists, the assessment was discussed until consensus was reached. Subsequently, a larger expert panel of four persons (including the initial two occupational hygienists) met to verify and discuss all potential inconsistencies with respect to assigning Stoffenmanager scores. This consensus meeting only occasionally resulted in modification of a Stoffenmanager score for a particular data point due to misinterpretation of contextual information during the initial assessment. The whole consensus procedure was conducted blind to the measurement results.
When multiple tasks were conducted during a measurement, Stoffenmanager scores (Ct) were calculated for each task and then combined together as a time-weighted summation for the tasks making up the measurement period. Multiple tasks were considered when identifiable differences existed in type of handling, product, controls or room during a particular measurement period.
Guidelines for data quality were applied to rank data into one of three categories: good, moderate or poor. Only good quality data were eventually used in the analyses. All exposure reports were reviewed to evaluate whether the work was undertaken competently and valid sampling and analytical techniques were used. In addition, exposure data were only labelled to be of good quality if required core information was documented (Rajan et al., 1997; Tielemans et al., 2002), if all Stoffenmanager parameters could be retrieved and if time registration was accurate. These criteria were considered stringent and we rejected any data sets not meeting these criteria. Often an occupational hygienist had to make further enquiries with the original researchers to retrieve additional details with respect to Stoffenmanager parameters and help clarify any ambiguities.
Data processing and statistical analyses
Both the measured exposure data and contextual information to derive Stoffenmanager scores were collected in a relational database in Microsoft Access 2003. To safeguard confidentiality, data were entered anonymously into the database. The data were analysed using SAS Statistical Software (version 9.1.3; SAS Institute, Cary, NC, USA). Visual inspection of the measured concentrations for solid and liquid scenarios showed a log-normal rather than a normal distribution, so descriptive statistics are presented both as arithmetic and geometric mean (GM) levels with geometric standard deviation (GSD) and range. In situations where measured values were below the LOD, 0.5 times the LOD was substituted for measured values (Hornung and Reed, 1990).
Spearman correlation coefficients were calculated to study the relation between Stoffenmanager scores and measured exposure concentrations. Mixed-effect regression models were used to further explore this relation by using the natural log of exposure data as dependent variable and the natural log of Stoffenmanger scores as independent variable, with random between- and within-company components of variance. Alternatively, a model with untransformed Stoffenmanager scores as independent variable was also tested but showed a poorer fit (statistically significant using likelihood ratio test). A compound symmetric covariance structure was used to model the data. The mixed model is given in equation (7), where Yij is the exposure level for the ith company and the jth worker; Xij is the log-tranformed exposure level; β0 is the intercept; β1 represents the fixed effect of the log of Stoffenmanager scores; 
i represents the random effect of the ith company and 
ij represents the random effect of the jth worker in the ith company. It is assumed that i and ij values are normally distributed with mean equal to 0 and variance of 
and 
, respectively, representing the between- and within-company variability component.
 | (7) |
The mixed-effect regression models can be used to predict a GM exposure level (
) for a given Stoffenmanager score Ct:
 | (8) |
The variation around the prediction is given by the components of variance. Hence, the random components of variance in conjunction with relevant z values of the standard normal distribution can be used to predict any cut point for a given Stoffenmanager score. For instance, to arrive at a conservative 90 percentile, the prediction of the GM should be multiplied using the following factor M:
 | (9) |
Graphical analyses of residuals were performed to evaluate assumptions of homoscedasticity. Statistical analyses were conducted separately for scenarios covering the handling of solids and liquids. For liquids, more detailed analyses were performed for volatile and non-volatile substances. For solid scenarios, a stratified analysis was conducted for handling powders and granules (e.g. mixing and weighing) and comminution of solid materials (e.g. sawing and grinding). In addition, stratified analyses were conducted based on the type of data source, i.e. (A) data collected in this study, (B) data from previous TNO research projects funded by the Dutch Ministry of Social Affairs and Employment and (C) data collected in the context of the VASt program.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
The results presented in Table 2 show the wide range of 14 different industries with in total 378 measured exposure data ranging from 0.0004 to 420 mg m–3 for inhalable dust scenarios. The data represent both short-term and long-term (shift) measurements. A relatively large number of inhalable dust measurements was available for handling powders (pyridoxine as a marker substance) in pharmacy shops (N = 78), flour dust among bakery workers (N = 56), dust among various construction sites (N = 74), wood dust in woodworking shops (N = 23), organic dust in the animal feed industry (N = 40) and pigment powders in textile (N = 28) and paint industry (N = 20). Small data sets were available for the fertilizer industry (N = 6), dairy industry (N = 3), metal industry (N = 4), transhipment industry (N = 5), rubber industry (N = 4) and a publishing company (N = 1). One simulated workplace study was included focusing on the impact of dustiness of products on exposure levels using standard scenarios like scooping, weighing and adding (N = 36).
|
The highest GM dust exposure levels were found among measurements in the paint industry (GM = 31.9 mg m–3, GSD = 4.26), the construction industry (GM = 14.0 mg m–3, GSD = 3.02) the rubber/plastic industry (GM = 12.2 mg m–3, GSD = 3.05) and the simulated workplace study (GM = 36.2 mg m–3, GSD = 4.01). As expected, very low exposure levels were found among pharmacy workers (GM = 0.05 mg m–3, GSD = 5.55).
Similar results are presented in Table 3 for the liquid scenarios (measuring solvents, pesticides/biocides or isocyanates) with in total 320 measured exposure data in different industries ranging from 0.0002 to 1762 mg m–3. The range in median sampling times across studies is large (9–510 min). Data in the agricultural setting represent application of pesticides in tree nurseries (bitertanol, N = 19) and horticulture (methomyl, N = 17). Data on biocide exposure were available for application of antifouling paint in boatyards (dichlofluanid and copper, N = 31) and pest control/disinfection operations (cyfluthrin and deltamethrin, N = 16; chlorpyrifos, N = 29 and quaternary ammonium compounds, N = 14). Data on volatile organic compound exposure levels were collected for handling paint and degreasing activities in the car body repair industry (N = 15) and metal industry (N = 56), handling of printing inks (N = 7) and gluing in orthopaedic shoe manufacturing (N = 26). Isocyanate exposure (HDI oligomers) was measured among car body repair workers involved in mixing and spraying of paint and gun cleaning (N = 90).
|
The highest GM solvent exposure levels (total volatile organic compounds) were found in the orthopaedic shoe manufacturing (GM = 128 mg m–3, GSD = 3.50) and the metal industry (GM = 56.7 mg m–3, GSD = 5.90). Activities with non-volatile substances (pesticides, biocides and isocyanates) resulted in much lower exposure levels.
Tables 4 and 5 describe the occurrence of key parameters for calculating the Stoffenmanager score for both solid (Table 4) and liquid scenarios (Table 5). For solid scenarios, tasks with handling scores >0 covered 81% of the total sampling time (=66 386 min), and for liquid scenarios, this appeared to be 66% of the total sampling time (=29 264 min) across all measurements. This implies that 81 and 66% of the time activities were conducted with at least some potential for exposure. In the remainder of the time, activities were conducted in the FF and NF which were not related to relevant exposure (i.e. handling score equal to zero). In these time periods, a diffusive source may still be present.
|
|
Stoffenmanager parameters of the NF and FF component were only reported for the sampling time with handling scores >0. For solid scenarios, Table 4 shows that some parameter classes were not or only to a very limited extent covered by the exposure data, i.e. outside work, solids with very low intrinsic emission scores (i.e. firm granules or flakes), enclosure, LEV in combination with enclosure and wetting. Other parameter classes were reasonably covered. For the handling parameter, the exposure data were distributed as follows: 21.6% in category 0.1, 25.7% in category 0.3, 3.6% in category 1, 20.4% in category 3 and 28.7% in category 10.
For liquid scenarios, Table 5 shows that the following parameters are not or only to a limited extent covered by the data, i.e. handling score 0.1, enclosure and LEV in combination with enclosure. For the handling parameter, the exposure data were distributed as follows: 0% in category 0.1, 6.9% in category 0.3, 24.5% in category 1, 20.7% in category 3 and 47.9% in category 10.
The Spearman correlation coefficients between Stoffenmanager scores and measurements appeared to be good for handling solids (rs = 0.80, N = 378, P < 0.0001) and liquid scenarios (rs = 0.83, N = 320, P < 0.0001) (Table 6). However, the correlation for liquid scenarios appeared to be lower when calculated separately for sets of volatile substances with a vapour pressure >10 Pa (rs = 0.56, N = 104, P < 0.0001) and non-volatile substances with a vapour pressure 10 Pa (rs = 0.53, N = 216, P < 0.0001) (Table 6). Whether volatile substances were reported in milligrams per cubic metre or in parts per million did not influence the correlation with the Stoffenmanager score. The dust scenarios could be subdivided into handling resulting in comminuting of bound products (e.g. sawing and grinding, N = 52) and handling of powders and granules (N = 326). The latter type of handling resulted in a correlation coefficient of 0.81, whereas activities leading to comminuting of bound products showed a lower correlation coefficient (0.41). Stratifying the data according to source did not show substantial differences in correlation coefficients, i.e. (A) this study (solids: rs = 0.58, N = 154, P < 0.0001 and liquids rs = 0.58, N = 78, P < 0.0001), (B) previous research projects (solids: rs = 0.64, N = 100, P < 0.0001 and liquids: rs = 0.53, N = 216, P < 0.0001) and (C) VASt program (solids: rs = 0.75, N = 124, P < 0.0001 and liquids: rs = 0.44, N = 26, P = 0.02).
|
The further exploration of the data by using mixed-effects models with a random company effect resulted in the models presented in Table 7. The relationship is graphically illustrated for handling of solids (Fig. 1) and liquids (Fig. 2). Both models had a statistically significant intercept [dust: β0 = 1.55, standard error (SE) = 0.17 and liquids: β0 = 6.17, SE = 0.36]. The slope of the regression line appeared to show a positive linear relation between the natural log of Stoffenmanager scores and natural log of measurement results for scenarios describing handling of solids (β1 = 0.69, SE = 0.05) and liquids (β1 = 0.87, SE = 0.04). These two regression equations enable the prediction of GM exposures for a given Stoffenmanager score (Ct):
 |
|
|
|
Total variance appeared to be higher for the liquid scenarios (
= 4.43) compared with the solid scenarios ( = 2.88). Based on these variance components, the difference between the predictions of the GM and the reasonable worst case (RWC) (90th percentile) was estimated to be a factor 8.8 (
) for solid scenarios and a factor 14.8 (
) for liquid scenarios. |
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
Although the concept of validation has been recognized as an indispensable part of model development (Armstrong et al., 1992; Schneider and Holst, 1996), only few validations of exposure models for risk assessment are described in the open literature (Bredendiek-Kämper, 2001; Tischer et al., 2003; ECETOC, 2004; Cherrie and Hughson, 2005; Hughson and Cherrie, 2005; Johnston et al., 2005; Money et al., 2006; Jones and Nicas, 2006a,b). The present study indicated that there is good agreement between Stoffenmanager scores and exposure measurements for both solid and liquid scenarios. The mixed-effects regression models with natural logged Stoffenmanager score as independent parameter explained a substantial part of the total exposure variability (52% for solid scenarios and 76% for liquid scenarios). This proportion of explained variance is well in accordance with other, more specific exposure studies focusing on a particular industrial setting (Burstyn and Teschke, 1999). Hence, this performance increases our confidence in the use of the Stoffenmanager as a generic tool for risk assessment. Yet, a cross-validation has to be conducted in order to evaluate the accuracy of the mixed-effect models (Hornung, 1991). This cross-validation using a small set of good quality exposure data will be conducted in a subsequent step (J. Schinkel, W. Fransman, D. Noy, H. Heussen, E. Tielemans, in preparation).
Notwithstanding the good correlation and parameter estimates from the mixed-effects model, the data show substantial variability in exposure measurements given a certain Stoffenmanager score. It is likely that various sources of uncertainty are responsible for this observed variability. First, there is uncertainty in the information describing the input parameters. Some parameters were most likely estimated with substantial error. For instance, the fraction of a substance in a mixture is often indicated in very broad ranges (e.g. 25–50%) in the available SDS. Likewise, other factors at the workplace may be assessed with varying degrees of error. Although we applied rigid quality control criteria for inclusion of data, this source of uncertainty undoubtedly resulted in discrepancies between model estimates and measurements.
Secondly, there is the usual degree of error inherent in the measurement data (Tielemans et al., 2002). Hence, measurements do not reflect true exposure and are itself an ‘alloyed gold standard’ (Wacholder et al., 1993). Uncertainty in the measurement data may be introduced by analytical error varying across laboratories or for instance differences in aerosol sampling instruments (Kenny et al., 1997). We consider this to be a relatively unimportant source of uncertainty, as measurement error is generally believed to be minor as compared to true exposure variability (Nicas et al., 1991). Stratified analyses did not reveal substantial differences in results between data sources, suggesting that uncertainty issues are not overrepresented in a particular data source.
A third, more fundamental reason for discrepancies between Stoffenmanager estimates and exposure measurements is model uncertainty (Morgan and Henrion, 1990). The Stoffenmanager exposure algorithm is to a large extent based on a well-described conceptual model of Cherrie et al. (1996) and Cherrie and Schneider (1999) with some modifications, i.e. the definition of intrinsic emission for liquid scenarios, assumptions with respect to the strength of FF sources (similar tasks and local controls are assumed as for NF sources) and the definition of background exposure due to diffusive sources. It is generally felt that the underlying concepts incorporate the critical determinants of exposure (Creely et al., 2005). Yet, as exposure is influenced by so many aspects, only the most dominant processes can be accounted for. For instance, one important aspect, personal behaviour, was explicitly not taken into account as this parameter is very difficult to characterize and quantify. Hence, there is scope for improvement by a further description of how workers become exposed. Furthermore, the scaling of the individual parameters is an important area for additional research. A logarithmic scale is used for most variables. The scaling is based on expert judgement and should be evaluated again when more exposure data become available. Determinant analyses using a large exposure database may provide important new insights into proper parameterization of individual variables of the Stoffenmanager.
Priority areas for further development of the Stoffenmanager are refinement of handling classes, inclusion of more options for control measures and the provision of high-quality guidance information for reliable classification. A larger number of classes in handling and local control parameters are envisaged to achieve a more precise algorithm in the near future. For instance, at the very low end of the exposure range, such as well-controlled activities in laboratories or pharmacies, one may consider the introduction of smaller handling scores than currently exist in Stoffenmanager. Similarly, there is scope for more differentiation in control efficacy values. These refinements require an expert elicitation procedure using a panel of multiple experts to reliably capture the current state of knowledge with respect to these parameters (Walker et al., 2001).
An important issue related to model uncertainty is that the Stoffenmanager inherently assumes that exposure is linearly dependent on the fraction of a substance in a mixture. However, the evaporation of a substance is also dependent on the specific composition of the mixture and on the activity coefficient of each component reflecting molecular interactions (Nielsen and Olsen, 1995; Fehrenbacher and Hummel, 1996). In addition, one should ideally use the mole fraction of a substance in a mixture instead of a mass fraction to predict partial vapour pressure. In practice, however, there is only limited information on characteristics of the mixture available so that we had to rely on less adequate, but accessible information.
The model of Cherrie et al. has previously also been evaluated (Cherrie and Schneider, 1999; Semple et al., 2001). The correlation coefficients found in their evaluation study are in accordance with or somewhat higher than the results presented in this paper. We found more scattering of exposure levels within a given Stoffenmanager score (i.e. ‘noise’). This discrepancy may well be explained by the fact that the methodology of Cherrie et al. was tailored to the specific assessment situation. Specific guidance material also included range-finding exposure data to calibrate the assessor. This is likely to help improve the accuracy and reliability of estimates (Hawkins and Evans, 1989; Post et al., 1991). In contrast, a flexible approach with additional guidance for each specific situation is impossible for the generic version of the Stoffenmanager. However, branch-specific versions of the Stoffenmanager may be more accurate and reliable than the generic version.
The mixed-effects regression models presented in this paper may be used for assessment of typical and so-called RWC exposures. The assessment can be based upon an appropriate percentile from the log-normal distribution as determined by the random components of variance. The technical guidance document (TGD) for risk assessment of new and existing substances currently recommends the 50th percentile for typical exposure and the 90th percentile for RWC exposure (ECB, 2003). However, these recommendations are not necessarily relevant for use of the Stoffenmanager. The TGD recommendations relate to measured data sets in rather broad exposure scenarios (e.g. ‘spray painting with solvent-based paints’). Stoffenmanager scenarios can be defined much more specifically. Therefore, if conservative Stoffenmanager inputs are used to describe a scenario, we recommend using the 75th percentile as the estimator of the RWC exposure level. If more average Stoffenmanager inputs are used for parameters that vary within a broad scenario, such as room size and local controls, the 90th percentile would be preferred as estimator of the RWC.
As the scatter of exposure measurements for a given Stoffenmanager score is rather large, the differences between 50th and 90th percentile is a factor up to 8.8 for solid and 14.8 for liquid scenarios, respectively. For the 75th percentile, a factor up to 3.1 (solids) and 4.1 (liquids) should be applied. These factors can be considered ‘safety factors’ to incorporate model uncertainty and inherent exposure variability in the risk assessment process. However, more exposure data are needed in the future to properly investigate the stability of variance across the whole range of Stoffenmanager scores. Several authors have highlighted the relevance of heterogeneity of variance across fixed effects (Friesen et al., 2006; Van Tongeren et al., 2006).
The unexplained variability (i.e. uncertainty) might be reduced by further optimization of the conceptual model and refinement of parameters in the future (see Discussion above). Yet, an additional strategy will be to combine the model estimates with available measurements relevant for the particular assessment scenario. Such an alternative strategy using Bayesian techniques to update model results with exposure data is proposed by Creely et al. (2005) and elaborated on by Tielemans et al. (2007). A few applications of a Bayesian approach to exposure assessment have already been described (Ramachandran and Vincent, 1999; Ramachandran et al., 2003; Hewett et al., 2006). As random between-company exposure variability in the mixed-effect models is large, there is potential for substantial improvement of Stoffenmanager estimates using site-specific data, even if only few measurements are available. We are currently exploring the possibilities of Bayesian techniques to update Stoffenmanager predictions.
The Stoffenmanager scores were derived by one assessor and these results were reviewed by a larger group using a consensus procedure. The consensus process is recommended by others as it helps to control for and resolve differences among experts as they gain knowledge from each other (Seel et al., 2007). There was good concordance among the experts in the consensus procedure. Yet, in our consensus procedure, we did not look explicitly at the reliability of the algorithm. Some comparable methods have been evaluated and show good inter-rater agreement (Semple et al., 2001; Van Wendel de Joode et al., 2005). Most parameters in the Stoffenmanager algorithm are not prone to subjective interpretation and simply require an objective description of the situation (e.g. LEV is present or not; a subjective assessment of efficacy of LEV is not required). However, two parameters provide the opportunity for subjective judgement, i.e. handling parameter and intrinsic emission for solids (dustiness). This potential for inconsistent interpretations was reduced as much as possible by providing transparent descriptions and by giving various examples for each parameter class. Nevertheless, a reliability study focusing on these aspects should be conducted in the near future.
Although the collated exposure data cover a wide range of situations, not all Stoffenmanager parameter combinations are included in the validation data set. For solid scenarios, not all intrinsic emission scores are well represented, e.g. substances with very low dustiness potential are not covered by the data. In addition, a very limited number of measurements in the data set were conducted outside. Likewise, completely contained and controlled process conditions (e.g. glove boxes) as well as wet suppression techniques were not included in the data. Hence, the performance of the model for these situations is not properly described in this study. Occupational activities such as processing of melted or burning materials (e.g. hot moulding and calendaring) or hot work techniques (e.g. welding and soldering) are lacking in the data set. In addition, only inhalable dust measurements are used in the validation study and thus the predictive value of the algorithm for respirable dust could not be assessed. Hence, these types of activities and exposures are currently outside the validity domain of the Stoffenmanager algorithm and should be dealt with in a later stage.
In general, we believe it is important to regularly update validation and calibration of exposure models as workplace scenarios, exposure levels and relations between determinants and exposure will change over time (Kromhout and Vermeulen, 2000; Creely et al., 2007). Currently, a web-based exposure database containing relevant contextual information is under development in The Netherlands (STEAMbase: SToffenmanager Exposure And Modelling database). The analyses described in this paper will be expanded when more good quality data become available in STEAMbase. In addition, new data will be used to re-examine the scaling of individual Stoffenmanager parameters. Such a cycle of regular model refinement and subsequent validation guarantees a method tailored to current work environments and process conditions. A larger number of measurements may also facilitate development of separate mixed-effects models for scenarios with different exposure mechanisms, e.g. handling of volatile substances (vapour exposure) and non-volatile substances (aerosol exposure).
In conclusion, Stoffenmanager appears to be a promising generic tool for exposure assessment. The Stoffenmanager is increasingly used as a tool to support SME in The Netherlands. This study shows that Stoffenmanager may also be used as a quantitative model. The mixed-effect models provide an explicit treatment of uncertainty and definition of so-called uncertainty factors. Several refinements in model parameters are planned for the near future. The link between Stoffenmanager and STEAMbase will hopefully result in a gradual increase of data available for calibration of the model.
|
FUNDING |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
Dutch Ministry of Social Affairs and Employment.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
We further acknowledge gratefully the industries for their contribution in the collection of data, Christiaan van Daalen for assisting in the data collation process and Jochem Liemburg and Marc van de Kerkhof for their contribution to the evaluation study. Hans Marquart is acknowledged for his critical comments on an earlier version of the manuscript.
Received December 24, 2007; in final form April 20, 2008
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONFUNDINGACKNOWLEDGEMENTSREFERENCES |
|---|
Armstrong BK, White E, Saracci R. Principles of exposure measurement in epidemiology (1992) Oxford: Oxford University Press.
Bredendiek-Kämper S. Do EASE scenarios fit workplace reality? A validation study of the EASE model. Appl Occup Environ Hyg (2001) 16:182–7.[CrossRef][Medline]
Brouwer DH, Links IHM, de Vreede JAF, et al. Size selective dustiness and exposure; simulated workplace comparisons. Ann Occup Hyg (2006) 50:445–52.
Burstyn I, Teschke K. Studying the determinants of exposure: a review of methods. Am Ind Hyg Assoc J (1999) 60:57–72.[Web of Science][Medline]
Cherrie JW. The effect of room size and general ventilation on the relationship between near and far-field concentrations. Appl Environ Occup Hyg (1999) 14:539–46.[CrossRef]
Cherrie JW, Hughson GW. The validity of the EASE expert system for inhalation exposure. Ann Occup Hyg (2005) 49:125–34.
Cherrie JW, Schneider T. Validation of a new method for structured subjective assessment of past concentrations. Ann Occup Hyg (1999) 43:235–45.
Cherrie JW, Schneider T, Spankie S, et al. A new method for structured, subjective assessment of past concentrations. Occup Hyg (1996) 3:75–83.
Creely KS, Tickner J, Soutar AJ, et al. Evaluation and further development of EASE model 2.0. Ann Occup Hyg (2005) 49:135–46.
Creely KS, Cowie H, van Tongeren M, et al. Trends in inhalation exposure: a review of the data in the published scientific literature. Ann Occup Hyg (2007) 51:665–78.
de Cock JS, Drooge H. Field study on occupational exposure during spraying of biocidal products by pest control operators using deltamethrin and (β)-cyfluthrin (2002) TNO-report V3806. Zeist, The Netherlands: TNO.
ECB. Technical guidance document on risk assessment in support of Commission Directive 93/67/EEC on risk assessment for new notified substances; Commission Regulation (EC) no 1488/94 on risk assessment for existing substances; Directive 98/8/EC (2003) Ispra, Italy: European Commission, Joint Research Centre.
ECETOC. Targeted risk assessment, technical report no. 93 (2004) Brussels, Belgium: European Centre for Ecotoxicology and Toxicology of Chemicals.
European Commission. (1998) Council Directive 98/24/EC on the protection of the health and safety of workers from the risks related to chemical agents at work. Off J Eur Commun L131. 11–23.
European Commission. (2006) EC Regulation (EC) No 1907/2006 of the European Parliament and of the Council of 18 December 2006 concerning the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH), establishing a European Chemicals Agency, amending Directive 1999/45/EC and repealing Council Regulation (EEC) No 793/93 and Commission Regulation (EC) No 1488/94 as well as Council Directive 76/769/EEC and Commission Directives 91/155/EEC, 93/67/EEC, 93/105/EC and 2000/21/EC. Official Journal L396. 1.
Fehrenbacher MC, Hummel AA. Evaluation of the mass balance model used by the environmental protection agency for estimating inhalation exposure to new chemical substances. Am Ind Hyg Assoc J (1996) 57:526–36.[Web of Science][Medline]
Friesen MC, Macnab YC, Marion SA, et al. Mixed models and empirical Bayes estimations for retrospective exposure assessment of dust exposures in Canadian sawmills. Ann Occup Hyg (2006) 50:281–8.
Garrod ANI, Rajan-Sithamparanadarajah R. Developing COSHH Essentials: dermal exposure, personal protective equipment and first aid. Ann Occup Hyg (2003) 47:577–88.
Hawkins N, Evans J. Subjective estimate of toluene exposures: a calibration study of industrial hygienists. Appl Ind Hyg (1989) 4:61–8.
Hewett P, Logan P, Mulhausen J, et al. Rating exposure control using Bayesian decision analysis. J Occup Environ Hyg (2006) 3:568–81.[CrossRef][Web of Science][Medline]
Hornung RW. Statistical evaluation of exposure assessment strategies. Appl Occup Environ Hyg (1991) 6:516–20.
Hornung RW, Reed LD. Estimation of average concentrations in the presence of nondetectable values. Appl Occup Environ Hyg (1990) 5:46–51.
Hughson GW, Cherrie JW. Comparison of measured dermal dust exposures with predicted exposures given by the EASE expert system. Ann Occup Hyg (2005) 49:111–23.
Jones RM, Nicas M. Evaluation of COSHH essentials for vapor degreasing and bag filling operations. Ann Occup Hyg (2006a) 50:137–47.
Jones RM, Nicas M. Margins of safety provided by COSHH essentials and the ILO Chemical Control Toolkit. Ann Occup Hyg (2006b) 50:149–56.
Jong de P, Dijk van W, Brouwer DH. Onderzoek naar de stofbelasting van stukadoors (1998) TNO-CSD-report 98 CSD 343 PJ/VL. Zeist, The Netherlands: TNO.
Johnston KL, Phillips ML, Esmen N, et al. Evaluation of an artificial intelligence program for estimating occupational exposures. Ann Occup Hyg (2005) 49:147–53.
Kenny LC, Aitken R, Chalmers C, et al. A collaborative European study of personal inhalable aerosol sampler performance. Ann Occup Hyg (1997) 41:135–53.
Kromhout H, Vermeulen R. Long-term trends in occupational exposure: are they real? What causes them? What shall we do with them? Ann Occup Hyg (2000) 44:325–7.
Kromhout H, Symanski E, Rappaport SM. A comprehensive evaluation of within- and between-worker components of occupational exposure to chemical agents. Ann Occup Hyg (1993) 37:253–70.
Links IHM, Jagt van der KE, Tielemans E, et al. Pest control operators’ inhalation and dermal exposure to chlorpyrifos during spraying of Aadifec® (2002) TNO-report V4534. Zeist, The Netherlands: TNO.
Links I, Jagt van der KE, Christopher Y, et al. Occupational exposure during application and removal of antifouling paints. Ann Occup Hyg (2007) 51:207–18.
Maidment SC. Occupational hygiene considerations in the development of a structured approach to select chemical control strategies. Ann Occup Hyg (1998) 42:393–400.
Marquart H, Lansink C, Engel R, et al. Effectiveness of local exhaust ventilation during dumping of powders from bags (1999) TNO-report V99.267. Zeist, The Netherlands: TNO.
Marquart H, Heussen H, le Feber M, et al. Stoffenmanager: a web-based control banding tool using an exposure process model (2008) Chem Occup Hyg; in press.
Money CD. European experiences in the development of approaches for the successful control of workplace health risks. Ann Occup Hyg (2003) 47:533–40.
Money C, Bailey S, Smith M, et al. Evaluation of the utility and reliability of COSHH Essentials. Ann Occup Hyg (2006) 50:642–4.
Morgan MG, Henrion M. Uncertainty: a guide to dealing with uncertainty in quantitative risk and policy analysis (1990) Cambridge, UK: Cambridge University Press.
Mulhausen JR, Damiano J. A strategy for assessing and managing occupational exposures (1998) 2nd edn. Fairfax, VA: AIHA Press.
Naumann BD, Sargent EV, Starkman BS, et al. Performance-based exposure control limits for pharmaceutical active ingredients. Am Ind Hyg Assoc J (1996) 57:33–42.[Web of Science][Medline]
Nicas M, Simmons P, Spear RC. Environmental versus analytical variability in exposure measurements. Am Ind Hyg Assoc J (1991) 52:553–7.[Web of Science][Medline]
Nielsen F, Olsen E. On the prediction of evaporation rates—with special emphasis on aqueous solutions. Ann Occup Hyg (1995) 39:513–22.
Post W, Kromhout H, Heederik D, et al. Semiquantitative estimates of exposure to methylene chloride and styrene: the influence of quantitative exposure data. Appl Occup Environ Hyg (1991) 6:197–204.
Preller EA, Schipper HJ. Respiratory and dermal exposure to desinfectants: a study in slaughterhouses and the meat processing industry (1999) TNO-report V98.1306. Zeist, The Netherlands: TNO.
Pronk A, Tielemans E, Skarping G, et al. Inhalation exposure to isocyanates of car body repair shop workers and industrial spray painters. Ann Occup Hyg (2006) 50:1–14.
Rajan B, Alesbury R, Carton B, et al. European proposal for core information for storage and exchange of workplace exposure measurements on chemical agents. Appl Occup Environ Hyg (1997) 12:31–9.
Ramachandran G, Vincent JH. A Bayesian approach to retrospective exposure assessment. Appl Occup Environ Hyg (1999) 14:547–57.[CrossRef][Medline]
Ramachandran G, Banerjee S, Vincent JH. Expert judgement and occupational hygiene: application to aerosol speciation in the nickel primary production industry. Ann Occup Hyg (2003) 47:461–75.
Russell RM, Maidment SC, Brooke I, et al. An introduction to a UK scheme to help small firms control health risk from chemicals. Ann Occup Hyg (1998) 42:367–76.
Schneider T, Holst E. Validation of exposure assessment in occupational epidemiology. Occup Hyg (1996) 3:59–71.
Seel EA, Zaebst DD, Hein MJ, et al. Inter-rater agreement for a retrospective exposure assessment of asbestos, chromium, nickel and welding fumes in a study of lung cancer and ionizing radiation. Ann Occup Hyg (2007) 51:601–10.
Semple SE, Proud LA, Tannahill SN, et al. A training exercise in subjectively estimating inhalation exposures. Scand J Work Environ Health (2001) 27:395–401.[Web of Science][Medline]
Tielemans E, Marquart H, de Cock J, et al. A proposal for evaluation of exposure data. Ann Occup Hyg (2002) 46:287–97.
Tielemans E, Warren N, Schneider T, et al. Tools for regulatory assessment of occupational exposure: development and challenges. J Exp Anal Environ Epidemiol (2007) 17:S72–80.[CrossRef]
Tischer M, Bredendiek-Kämper S, Poppek U. Evaluation of the HSE COSHH Essentials exposure predictive model on the basis of BAuA field studies and existing substances exposure data. Ann Occup Hyg (2003) 47:557–69.
Van Tongeren M, Burstyn I, Kromhout H, et al. Are variance components of exposure heterogeneous between time periods and factories in the European carbon black industry? Ann Occup Hyg (2006) 50:55–64.
Van Wendel de Joode B, van Hemmen JJ, Meijster T, et al. Reliability of a semi-quantitative method for dermal exposure assessment (DREAM). J Expo Anal Environ Epidemiol (2005) 15:111–20.[CrossRef][Web of Science][Medline]
Vreede de JAF, Amelsfort van M. Exposure to pesticides during application in tree nurseries using an airblast technique (1997a) TNO-report V97.119. Zeist, The Netherlands: TNO.
Vreede de JAF, Amelsfort van M. Exposure to pesticides in a tree nursery using the spray boom and spray lance (1997b) TNO-report V97.111. Zeist, The Netherlands: TNO.
Vreede de JAF, Haan de M, Brouwer DH, et al. Exposure to pesticides during high-volume application in greenhouses (1994) TNO-report V94.300. Zeist, The Netherlands: TNO.
Wacholder S, Armstrong B, Hartge P. Validation studies using an alloyed gold standard. Am J Epidemiol (1993) 137:1251–8.
Walker KD, Evans JS, Macintosh D. Use of expert judgement in exposure assessment part 1. Characterization of personal exposure to benzene. J Expo Anal Environ Epidemiol (2001) 11:308–22.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  T. Ogden Data Sharing, Federal Rule of Evidence 702, and the Lions in the Undergrowth Ann. Hyg., October 1, 2009; 53(7): 651 - 655. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Tielemans, T. Schneider, H. Goede, M. Tischer, N. Warren, H. Kromhout, M. Van Tongeren, J. Van Hemmen, and J. W Cherrie Conceptual Model for Assessment of Inhalation Exposure: Defining Modifying Factors Ann. Hyg., October 1, 2008; 52(7): 577 - 586. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||