Annals of Occupational Hygiene Advance Access originally published online on March 20, 2006
Annals of Occupational Hygiene 2006 50(5):491-503; doi:10.1093/annhyg/mel014
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Task-based Dermal Exposure Models for Regulatory Risk Assessment
1 Health and Safety Laboratory, Harpur Hill Buxton, Derbyshire, UK
2 TNO Chemistry, Zeist The Netherlands
3 TNO Chemistry, Zeist, The Netherlands, presently Institute of Occupational Medicine Edinburgh, UK
4 Kuopio Regional Institute of Occupational Health, Kuopio Finland
*Author to whom correspondence should be addressed. Tel: +44-1298-218442; fax: +44-1298-218471; e-mail: nick.warren{at}hsl.gov.uk
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
The regulatory risk assessment of chemicals requires the estimation of occupational dermal exposure. Until recently, the models used were either based on limited data or were specific to a particular class of chemical or application. The EU project RISKOFDERM has gathered a considerable number of new measurements of dermal exposure together with detailed contextual information. This article describes the development of a set of generic task-based models capable of predicting potential dermal exposure to both solids and liquids in a wide range of situations. To facilitate modelling of the wide variety of dermal exposure situations six separate models were made for groupings of exposure scenarios called Dermal Exposure Operation units (DEO units). These task-based groupings cluster exposure scenarios with regard to the expected routes of dermal exposure and the expected influence of exposure determinants. Within these groupings linear mixed effect models were used to estimate the influence of various exposure determinants and to estimate components of variance. The models predict median potential dermal exposure rates for the hands and the rest of the body from the values of relevant exposure determinants. These rates are expressed as mg or µl product per minute. Using these median potential dermal exposure rates and an accompanying geometric standard deviation allows a range of exposure percentiles to be calculated.
Keywords: dermal exposure • exposure modelling • risk assessment
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
The RISKOFDERM project (EU Fifth Framework Program, project QLK4-CT-1999-01107) had two main goals (van Hemmen, 2004): (i) the creation of dermal exposure model(s) for use in regulatory risk assessment of substances and (ii) the creation of a toolkit for risk assessment and risk management of dermal exposure in small and medium-sized enterprises (SMEs). Both these aims were supported by a programme of field studies that gathered quantitative and qualitative information on dermal exposure in the workplaces of five EU member states. Whilst the development of dermal exposure models is the subject of this publication, a companion paper (Marquart et al., 2006) describes how the same exposure data may be used to define default exposures for a number of standard scenarios. The development of the toolkit has previously been described (Goede et al., 2003; Oppl et al., 2003; Schuhmacher et al., 2003; van Hemmen et al., 2003; Warren et al., 2003).
There are manifold exposure situations that may lead to dermal exposure. To facilitate the modelling of this multitude of exposure scenarios, each with potentially different relationships between dermal exposure and its determinants, the exposure scenarios have been grouped into clusters. The RISKOFDERM project categorized exposure scenarios using a task-based approach into so-called dermal exposure operation units or DEO-units. A DEO unit is a cluster of more or less similar exposure processes and exposure routes for which similar relationships between potential dermal exposure and exposure determinants are expected. All exposure scenarios studied within the RISKOFDERM project have been assigned to one of the following six DEO units:
- DEO unit 1: Handling of (contaminated) objects, includes transferral of products from one object (or container) into another (e.g. mixing, filling) as well as any task where the primary source of exposure is contact with contaminated objects and surfaces. Exposure is mostly due to contact with contaminated surfaces but deposition of aerosols and some direct contact or immersion may also occur.
- DEO unit 2: Manual dispersion of products, relates to the dispersion of a product onto or over a surface by hand. It includes dispersion by a rag or sponge or other ‘tool’ without handle and often includes partial immersion (of the hands) into a container with the product. Exposure is mainly due to direct contact (immersion) and due to contact with contaminated surfaces.
- DEO unit 3: Dispersion of products with a hand-held tool, relates to dispersion of a product onto or over a surface by means of a ‘tool’ with a handle that is handled manually. Exposure is mostly due to contact with contaminated surfaces, but some direct contact (splashing, dripping) may also occur.
- DEO unit 4: Spray dispersion of a product, relates to dispersion of a product onto or over a surface by means of (pressurized) spraying equipment; exposure is due to deposition of aerosols and due to contact with contaminated surfaces.
- DEO unit 5: Immersing of objects into a product, in the scope of the measurements done, this largely relates to mechanical immersing of objects into baths (using hoists, etc.), but may alsoinclude manual immersion of, for example, a basket containing objects or parts of objects held by hand. Exposure is due to direct contact (immersion) and contact with contaminated surfaces.
- DEO unit 6: Mechanical treatment of solid objects, relates to the treatment of solid objects (such as grinding, sawing), where fractionated components of the object become airborne, together with any material that is on the surface of the object; it includes the exposure to metal working fluids used in the process. Exposure is due to deposition of aerosols and due to contact with contaminated surfaces.
DEO unit 1 is clearly more broadly defined than the other five categories and represents a wide range of exposure scenarios. Due to this breadth of definition, and the predominance of available exposure data within a smaller subset of exposure scenarios, an exposure model is only presented for mixing, loading and filling tasks.
A variety of different metrics exist for measuring dermal exposure including: mass, mass per unit of time, mass per skin surface area and mass per unit of time per skin surface area. In order to develop models with the widest domain of application we have elected to model dermal exposure in µl min–1 formulation (liquids) or mg min–1 formulation (solids). The following two assumptions are implicit in adopting this choice of exposure metric.
- Exposure to a specific chemical agent is proportional to its concentration in the formulation.
- Average exposure increase linearly with duration of work.
This method of assessing dermal exposure, namely to predict dermal exposure to a formulation and then calculate exposure to a specific chemical as a fraction of this deposited amount, is only appropriate for non-volatile substances. For volatile chemicals consideration must be given to the subsequent evaporation of the chemical from the skin and the consequential loss of chemical available for uptake. Exposure assessments frequently assume a linear relationship with duration but the evidence for this is conflicting (Marquart et al., 2003). It is clear, however, that the risk assessor is frequently forced to make extrapolations from the durations of measured data to a variety of other time periods and this requires a method of scaling.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Datasets
The datasets used in the development of the models are summarized in Table 1. The majority of data were collected specifically for the development of exposure models by work part 2 of RISKOFDERM. Although some additional data were obtained from the literature, such data could frequently not be used either because the relevant exposure determinants had not all been studied or because only summary statistics had been published and not the individual data. Finally, after the initial modelling work had been completed, work part 3 of RISKOFDERM provided a relatively small number of new data to benchmark the initial models against. In this article, we present the final analysis of all the assembled data.
|
We do not provide complete descriptive statistics for every dataset as these can already be found not only in the individual study publications but also in the excellent overviews of the data provided by Rajan-Sithamparanadarajah et al. (2004) and Kromhout et al. (2004). Summary statistics for exposure to the whole body (excluding hands) and hands for each DEO unit are given in Tables 2 and 3. These tables summarize exposure data for both liquid and solid formulations expressed in mg min–1 where liquid formulations have an assumed density of 1 mg µl–1. Tables 2 and 3 show the enormous range of dermal exposure that can occur even within groupings of similar tasks. For example, potential dermal exposures to the whole body as a result of spray applications range by 6 orders of magnitude. No measurements of potential hand exposure were obtained for scenarios within ‘mechanical treatment’. For safety reasons it was not possible to attach sampling media to the outside of workers’ protective gloves whilst they were working with machinery. Although measurements of potential whole body exposures were obtained for scenarios within DEO 1 no model of body exposure is presented for this DEO unit. No satisfactory explanation of exposure in terms of exposure determinants could be found with the best relationship still considerably poorer than simply presenting each dataset as a distinct exposure scenario.
|
|
Statistical analysis
Linear mixed effects models were fitted using restricted maximum likelihood estimation available within S-Plus (version 6.0) (2001)). Repeated measurements on an individual can include concurrent measurements of hand and whole body exposures, measurements repeated within the same day and measurements repeated on subsequent days. The models can be represented in the following generic form:
 | (1) |
i,0 represents the mean (log) potential dermal exposure for the DEO, 1 represents the first fixed effect for that DEO. The presence or absence of each determinant is represented by indicator variables. For example, I1,i,j takes value 1 if the first determinant was present for the jth measurement on the ith worker and 0 otherwise. More complex fixed-effects, for example exposure determinants acting in combination or differential effects for determinants acting upon different body locations, can be modelled through the inclusion of interaction terms.
For models containing a continuous covariate X1 (such as the logarithm of the application rate) the following formulation would apply:
 | (2) |
In both instances ßi represents the random effect for the ith worker and 
i,j is the random error associated with the jth measurement on the ith individual. We assume that ßi and i,j are independent and normally distributed with means of 0 and standard deviations of 
BW and WW, respectively. Furthermore, we allow the possibility that the within-worker variability differs between hand and body exposures. In formal notation this is represented by
 | (3) |
Treatment of non-detected exposures
A common approach for the treatment of exposures below the limit of detection (LOD) in occupational studies is to substitute with a single common value. Occasionally a value of 0 may be substituted (or a value very close to 0 to allow logarithms to be taken), but more commonly a value of one half or 2–0.5 times the LOD is chosen. Whilst such approaches are appealing on grounds of their simplicity any resulting analysis with appreciable numbers of data below LOD is highly sensitive to the value chosen. Instead we have chosen to address exposures below the LOD through multiple imputation. In this approach a random sample of exposures is created to substitute for non-detected values. Whilst no specific assigned exposure will be correct, the distribution of the assigned exposures ought to be similar to the distribution of true unobserved exposures that were less than the LOD. The data is then analysed as though these assigned values were really observed. This is a well-established statistical technique for the treatment of truncated data especially where complex subsequent analysis is to be performed. The merits of multiple imputation are discussed in full by Rubin (1987).
Our imputed values are drawn from a log-normal distribution conditional upon being less than their LOD. Log-normal distributions were fitted to each dataset using a maximum likelihood algorithm for left censored data (implemented in MATLAB, 2003). This technique has been proposed for use in occupational hygiene by, among others, Finkelstein and Verma (2001). This method gives estimates of the GM and geometric standard deviation (GSD) that reflect the imprecise knowledge of exposures below LOD.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
In total, 21 statistically significant exposure determinants are included in the models representing many of the factors commonly perceived to influence exposure (Marquart et al., 2003; Van Wendel-de-Joode et al., 2003). An explanation of each determinant is presented in Table 4. All of the retained determinants are dichotomous factors with the exception of the natural logarithm of the application rate, which is continuous and expressed in l min–1 (liquid formulations) or kg min–1 (solid formulations).
|
For each model the estimated fixed effects along with their precision and significance levels are given in Tables 5–10. These tables also contain the anti-log of the coefficient for each fixed effect [and accompanying 95% confidence interval (CI)]. These values represent the multiplicative change in average exposures that occur with each determinant. The models for the six DEO units vary considerably in complexity from the model for mechanical treatment (DEO 6) with 6 parameters (including variance components) to the model for spraying (DEO 4) with 13 parameters.
|
|
Filling, mixing and loading
The model for filling, mixing and loading (Table 5) is applicable only to hand exposures. For manual processes the coefficient for the logarithm of the application rate is 0.927 (95% CI 0.745–1.109) implying that exposure rates are approximately proportional to the rate of application. For semi-automated filling operations a much smaller coefficient for (log) application rate is obtained implying less dependence between rate of exposure and the rate at which products are handled (exposure rates are proportional to the rate of application raised to the power of 0.19, 95% CI 0–0.38). Whilst significant effects were observed for infrequent and light contact these had a modest influence on exposure levels (both reducing exposure by approximately a third). In contrast, the physical properties of the product had a marked effect with fine dusty solid particles producing exposures >7 times higher than less dusty solids. Exposures to liquids were 
30 times higher than those typical for solids.
Wiping
The wiping model presented in Table 6 is fairly simple containing just seven estimated parameters (intercept, three fixed effects and three variance parameters). The model predicts body exposures that are much lower than hand exposures except where there is extensive contact between the body and the freshly wiped surfaces. The model has separate variance components for the hands and body with significantly greater variation predicted for the body (total GSD = 5·8) than the hands (total GSD = 3·5, P < 0·001).
|
Hand-tool dispersion
Estimated fixed effects for the hand-tool dispersion are given in Table 7. No data were available concerning dispersion of a solid substance with a hand held tool and so the model is only applicable to liquids. A number of significant exposure determinants are incorporated including work orientation, substance viscosity and the amount of substance handled (expressed as a rate). The length of tool handle (proximity) only had a significant effect for the hands where exposures were 3 times higher when a short-handled tool was used. This is consistent with exposure to the hands being predominately due to contact with contaminated tools whereas body exposure being primarily splashes or accidental contact with treated surfaces. The within-worker variance component is substantially larger for the hands (WW = 1.77) than the body (WW = 0.67, P = 0.004). The total variance represents a GSD for hand exposures of 11.2. Although this represents a considerable reduction on the level of variation exhibited in the dataset (GSD = 23.7, see Table 3) it is still higher than would be expected for many individual exposure scenarios (Garrod et al., 2000; Eriksson and Wiklund, 2004).
|
Spray dispersion
The spray dispersion model presented in Table 8 is based upon the largest of the available datasets (475 data covering 10 scenarios). Not surprisingly, the model has the greatest number of exposure determinants with several having quite modest effects that are discernable by virtue of the extensive dataset. The model relates potential dermal exposure to application rate, direction of application, environment, direction of airflow, segregation, distance of the worker to the source and volatility of the product. Surprisingly, spray pressure (which was highly correlated with nozzle diameter) was not a useful predictor of exposure once the rate of application was included. Although exposure assessments frequently distinguish between low and high-pressure scenarios often the distinction is between low volume, low pressure spraying and high volume, higher pressure spraying. The highest spray pressures are often associated with the application of viscous liquids such as antifoulant paints. In the RISKOFDERM project these products also contained a high proportion of volatile ingredients and so there is also an association between the highest spray pressures and volatility. Rates of application were found to have a notable effect on exposures with body exposures being discernibly more sensitive to this determinant than the hands.
|
Whilst the fixed effects only explain 31% of the total variance, the residual errors (corresponding to within worker variability) were quite modest (WW = 1.22). The combined variance components equate to a GSD for the log-normally distributed exposure rates of 6.0. This value is not inconsistent with levels of variability reported elsewhere for dermal exposure during spraying (HSE, 1999a; ECB, 2002).
Immersion
The immersion model (presented in Table 9) is the least satisfactory of the developed models only incorporating proximity and ventilation as exposure determinants and having large residual errors for both the hands and the body. The model is based upon a limited dataset and includes just 13 measurements of potential exposure to the hands (all relating to degreasing of machine parts using N-methyl pyrrolidone). Statistically significant effects were found for local exhaust ventilation (LEV) (leading to 75% reduction exposures) and proximity to the substance (<30 cm giving exposures 5 times higher, >1 m an 80% reduction in exposures).
|
Mechanical treatment
The model for mechanical treatment processes (see Table 10) is only applicable to body exposures and is based upon a fairly limited set of 97 data covering four scenarios. Significantly lower exposures were observed for solid particles (grinding of acid-proof steel and carpentry) than for liquids (machining and metal working fluids) with a reduction factor of 0.074. Effects were also noted for proximity to source of contamination and frequency of contact with contaminated surfaces. The use of LEV was not observed to lead to a reduction in dermal exposure.
Variance components
Estimates of between and within-worker variability for each DEO unit are presented in Table 11. In general between-worker variability is the smaller component with BW ranging from 0.90 to 1.65. Four models are combined models for both hand and body exposures, and of these three (wiping, hand-tool dispersion and immersion) have separate within-worker variance components for the hands and body. The combined GSD (i.e. the variability of the untransformed exposures for a random worker on a random day but with a specified set of exposure determinants) varies greatly from 3.5 to 34.2 (see also Table 11). The combined GSD reflects not only the intrinsic level of exposure variability but also the success of the fixed effects in explaining this variability. Consequently, although the fixed effects for hand-tool dispersion explain 75% of the variation, the enormous range of exposures in the modelling dataset (see Table 2 and Table 3) result in combined GSDs that are quite high (body 5.9, hands 11.2, Table 11).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
The models presented in this article are based upon datasets of quite varying sizes with some datasets that are relatively large, e.g. 224 data for hand-tool dispersion, 475 data for spray dispersion. Nevertheless, even with the larger datasets high correlations between determinants, or inadequate variation in these determinants, precluded a complete evaluation of all their potential influences. Therefore, to varying extents, the models are lacking some exposure determinants that are known or believed to be influential.
For wiping there was little real opportunity to study the effects of viscosity as 110 out of 136 data were exposures to water-based solutions and only 8 data (4 body, 4 hands) were for viscous liquids. Ventilation was not found to have a significant effect on dermal exposure which is unsurprising given that wiping tasks frequently involved the immersion of acloth (and hand) into a cleaning solution and therefore exposure is predominantly through direct contact.
The orientation of work has been identified as an important determinant for both hand-tool and spray dispersion. Although both models contain fixed effects for downward applications, overhead work was only found to have a statistically significant effect on spray applications. We suspect that the small number of data (12) for overhead hand-tool applications accounts for this lack of significant effect rather than there genuinely being no increase in exposure with this pattern of work.
For immersion two separate studies of electroplating provide the majority of the measurements for the body (55 out of 68) but report widely differing exposures. Roff et al. (2004a) found median exposure rates of 7.70 µl min–1 whereas Mäkinen and Linnainmaa (2004b) report a value of 0.056 µl min–1. Two possible explanations are proposed for such a difference between two apparently similar scenarios. One explanation is that Roff et al. (2004a) measured only the specific immersing tasks (with sample times between 4 and 60 min), whilst Mäkinen and Linnainmaa (2004b) sampled for much longer periods (80–483 min). Thus, the sampling of Mäkinen and Linnainmaa (2004b) may have included substantial periods of other tasks that did not lead to dermal exposure and therefore lower overall rates of dermal exposure. Differences in the methodologies for dermal sampling and chemical analysis may also have contributed to the difference in between the studies. Unfortunately, ‘level of automation’ was not always recorded in the work part 2 questionnaires so, in order to have a larger dataset without having to assign levels for automation, the model has been based upon proximity. However, if a model is fitted utilizing automation rather than proximity (including both gives non-significant effects as there was a strong correlation between the two determinants) then automated processes have an effect comparable with ‘proximity >1 m’, whilst manual dipping has a magnitude of effect comparable to ‘proximity <30 cm’.
Surprisingly, LEV was not found to have an effect for mechanical treatment processes. This lack of effect for LEV taken in conjunction with significant effects for proximity to source and frequency of contact suggests deposition may not be the primary route of exposure for mechanical treatment and that contact with contaminated surfaces and objects may play a major role. Other than physical state, the model contains no determinants relating to physico-chemical properties such as particle size or viscosity. Whilst a lack of variation in the determinant prevented a meaningful evaluation of the effect of particle size, metal working fluids with different viscosities were studied with no significant effect for viscosity being found. As all the available data were for machining tasks using powered tools care must be exercised if the model were applied to ‘low energy processes’ (such as carpentry using hand tools). It should not necessarily be assumed that exposures will be lower in such instances—if surface contact is a major contributor to dermal exposure then standards of cleanliness (which can be unrelated to the type of tool being used) will play a major role in determining levels of exposure.
There are four models that predict both body and hand exposures (DEOs 2–5) and they share no common relationship between body and hand exposures. For wiping and immersion the fixed effects for body-part predict body exposures that are approximately one-tenth of the corresponding hand exposures. For dispersion using a hand tool and spray dispersion the fixed effect for body-part suggests median body exposures that are 10 and 5 times the corresponding hand exposures, respectively (in terms of total mass per minute). However, these models also include an interaction between body-part and the effect of application rate, so the ratio of body to hand exposure varies with the rate of application. For dispersion using a hand-held tool, hand exposures are proportional to the application rate raised to the power of 1.18 (i.e. approximately a linear relationship) whereas body exposures are proportional to the application rate raised to the power of 1.53 (i.e. exponentially dependent upon the rate of application). For spray dispersion both hands and body exposures have a sublinear relationship with application rate with a coefficient of 0.36 for the hands and 0.64 for the body. These differential effects for application rate result in the ratio of body to hand exposures increasing with higher application rates (with the quoted values of x10 and x5 corresponding to an application rate of 1 l min–1).
For manual mixing, loading and filling exposures have an approximately linear relationship with the rate of application (coefficient 0.93, 95% CI 0.74–1.11) which when coupled with the assumed linear relationship between total exposure and duration implies exposures are approximately proportional to the amount of product handled. Such a relationship is frequently assumed in the regulation of plant protection products where exposures are typically expressed in mg per kg handled, e.g. EUROPOEM (1996).
Although most of the fixed effects show considerable statistical significance the precision of their estimated effects is quite varied. For example, a 95% CI for the effect of liquid formulations in DEO 1 is 5.5–155, a 28-fold range, whereas for the effect of downward spraying in DEO 4 is 0.29–0.74, a 2.5-fold range. Although it is clear that tighter CIs for the effects of exposure determinants imply more accurate predictions it is difficult to use the standard errors to make general statements about the accuracy of the models. Each instance of a model's use will use a specific combination of exposure determinants and so CIs must be calculated on a case-by-case basis. These calculations are not trivial as the parameter estimates for each determinant are not independent but instead have their own correlation structure. Calculating CIs for estimates of exposure percentiles other than the 50th (median) also necessitates consideration of the uncertainty in the variance parameter estimates. A rough indication of model accuracy may be obtained by considering CIs for the anti-log of each of the intercept terms. These give CIs for median hand exposures (body exposures for mechanical treatment) under certain default conditions—namely the absence of any of the DEO specific exposure determinants. On this basis the models for wiping (95% CI 453–1464 µl min–1), spray dispersion (95% CI 14.7–39.3 µl min–1) and mechanical treatment (95% CI 6.4–34.4 µl min–1) do not appear unreasonable whilst the models for mixing, filling and loading (95% CI 0.06–1.73 mg min–1) and immersion (95% CI 0.8–76.7 µl min–1) would be expected to give very imprecise estimates of exposure.
The P-values quoted in the results section are for two-sided significance tests. In reality, often we have fairly strong a priori beliefs about the effects of determinants, at least to the extent of whether a determinant increases or decreases exposure. Such an argument can provide justification for using one-sided significance tests with a corresponding reduction in P-values (of one half). Only three determinants have P-values in excess of 0.05 and for the effects of high volatility and overhead application on exposures resulting from spray dispersion it would not be unreasonable to apply a one-sided test, for we would not expect either of these to lead to a reduction in exposures. The third instance is for the effect of body part on dermal exposures resulting from immersion tasks. This is a test for whether potential exposure levels are different for the hands than for the rest of the body. It is important to recognize that a parameter value of 0 only corresponds to no difference in the chosen exposure metric (µl min–1) and that the null hypothesis represents a 23-fold difference in exposures per unit area (µl cm–2 min–1). In this situation the standard choice of null hypothesis for determining the statistical significance of a fixed effect is quite arbitrary, and so despite a P-value of 0.081, we have chosen to adopt a model with a parameter distinguishing between hand and body exposures.
A single within-worker variance component was only adequate in the spray dispersion model. In the other three (out of four) models that predict both body and hand exposures there are significant differences in the residual errors between the body and hands although there does not appear to be any systematic difference across all DEOs. In our models the estimates of within-worker variance are determined from the model residuals and so are heavily determined by the success of the fixed effects in accounting for differences in measured exposures. Also, for each DEO there are different numbers of hand and body data and these data may be distributed differently across the studied exposure scenarios. Therefore, the estimates of combined variability should not be interpreted as indicative of exposure variability more generally outside of the modelling context. For example, it should not be inferred from the models that for all hand-tool dispersion scenarios there is greater variability in dermal exposure to the hands than to the body.
Although our findings appear to confirm those of Kromhout et al. (2004), who showed that within-worker variability outweighed the between-worker component, in general no direct comparison should be made. Whereas our analysis includes an extensive study of exposure determinants (represented as fixed effects) their analysis has a more limited treatment of determinants. As a result variance that in our models is explained by exposure determinants will in their analysis be apportioned to the within and between-worker variance components. A further difference is that whilst they have performed separate analyses for the hands and body, where data permitted we have developed combined body/hands models. Although our primary motivation for combined analyses is the increase in power and precision when studying the effects of exposure determinants, the use of a common between-worker variance component is also intuitively appealing. If the between-worker variance component represents worker characteristics such as experience, training, care, awareness and concern about hazard then these ought to exert a similar influence on both hand and body exposures. Furthermore, it might be supposed that these characteristics should be reasonably consistent across different DEO units. In our analysis between-worker variability exhibits greater consistency across DEO units than within-worker with BW ranging from 0.899 to 1.651 whilst for the body WW is between 0.670 and 2.056 (hands 0.863–3.415).
Clearly, more complex covariance structures could have been adopted. Specifically, we have not considered possible autocorrelation of repeated measurements. However, the main emphases of our analysis were the possible relationships between dermal exposure and workplace exposure determinants and we do not believe that these would change substantially with a different covariance structure. Indeed, earlier statistical analyses using a single within-worker variance component, or using multiple linear regression instead of mixed effects models, have yielded essentially the same estimates of the fixed effects.
Using the models in quantitative risk assessment
The models presented in this article were developed primarily for higher tier risk assessments, so the discussion that follows focuses on using the models to predict exposures for specific, well described, workplace exposure scenarios. The models might also be used for screening level risk assessments but only by first describing a number of default, reasonable worse case scenarios for which possible uses of a chemical could be rapidly screened against.
Exposure estimates for specific workplace conditions can be obtained by multiplying together the effects (the anti-logs of the fixed effects) of relevant exposure determinants along with the intercept term to obtain a median exposure rate for the specified exposure scenario. More generally, the distribution of dermal exposure for a scenario can be modelled through a log-normal distribution with the calculated geometric mean (median) and a GSD taken from Table 11. The full range of exposure percentiles can be calculated from this estimated distribution.
The appropriate choice of percentile is a subject of much debate and the percentiles adopted vary between European member states and different classifications of chemicals. RISKOFDERM is primarily aimed at industrial chemicals and the Technical Guidance Document (TGD) for risk assessment of new and existing substances currently recommends the 50th percentile for typical exposures and the 90th percentile for reasonable worse case exposures (ECB, 2003). The User Guidance for the Biocides Technical Notes for Guidance (ECB, 2004) presents an approach where higher exposure percentiles are recommended where datasets are small or highly heterogeneous. Specifically, the User Guidance suggests using the 75th percentile when a 90% CI for this percentile is not more than a factor of two times higher or lower and the 95th percentile when the uncertainty is greater than this. We have already noted that calculating CIs for exposure percentiles derived from linear mixed effect models is non-trivial and must be done on a case-by-case basis. However, an examination of the CIs for the intercept terms suggests that the User Guidance criteria for adopting the 75th percentile are only likely to be met for wiping, spray dispersion and mechanical treatment.
Although risk assessments could be based upon an appropriate percentile from the log-normal distributions described by the models, in most real exposure scenarios it is likely there will be some variation in at least one determinant. For example, orientation and application rate are unlikely to be exactly the same for every worker/work session except in special instances. In these situations a conservative approach would be to select the combination of exposure determinants that give rise to the highest exposures and base the risk assessment on the resulting (median) predicted exposure rate. Subsequently, a worst-case estimate of exposure for this scenario might be represented by the 90th percentile of the log-normal distribution predicted for this combination of determinants. If this procedure is followed for both hand and body exposures the compounded conservatism can easily become unacceptably large. An alternative is to use the models inside of a Monte Carlo simulation to predict the complete distribution of exposures that arise under typical variation in workplace conditions. Probabilistic techniques might also be used to obtain realistic estimates of whole shift exposures rather than simple aggregation of the predicted percentiles from more than one model.
The use of the models in risk assessment should take account of the limitations of the models. Extrapolation in time beyond the duration of measurement encountered in the measurement sets should be done very cautiously, as calculations can indicate that prolonged durations lead to exposure loadings higher than those that are found when the skin is fully immersed in viscous liquids. Maximum dermal retentions in excess of 10 mg cm–2 are considered extremely unlikely for both solids and liquids (SAIC, 1996). It should however be remembered that the models are for potential dermal exposures and so maximum skin retentions may not be directly relevant depending upon the use of gloves and the nature of work clothing.
In principle models may be developed that are better at explaining variation in dermal exposure, without being very useful in a regulatory scope. Regulatory assessments in the scope of the new and existing substances regulations in Europe rely on information of limited specificity. If a model needed highly site or product-specific information, it would not be very useful for its intended regulatory purpose. Although this consideration was already taken into account in the data gathering of work parts 1 and 2 (which was aimed at generally available and assessable information) it remained a consideration during the model development. Also, because of the high correlation between exposure determinants and scenarios, often several candidate models could be proposed. We have taken a holistic approach to the final choice of model focussing on their intended end purpose as tools to aid risk assessments. Where possible, we have chosen models where it is reasonable to suppose an exposure assessor will know (or be able to obtain information about) the values of all exposure determinants included in the model.
It was the intention of the project to create models for all possible dermal exposure situations. This is a rather ambitious goal and it is not surprising that the final results cannot be taken as the definitive models for all dermal exposure situations. For some exposure scenarios there are insufficient representative data to allow models to be developed. For example, in DEO 6 no potential hand exposure measurements were available for modelling. For DEO 5, most of the situations involved mechanical immersion and hardly any manual immersion was included. In DEO 1 the variation in tasks and in the specific manner in which they are done is very large and we have only presented a model for the mixing, loading and filling tasks. For hand tool dispersion (DEO 2) all exposures were for liquid substances. For all DEO units high correlations were observed between determinants and between the effects of certain determinants and the scenarios (or measurement sets). These correlations precluded a full independent evaluation of all influences and prevented the inclusion of some combinations of determinants.
These correlations result from a measurement strategy based upon real, occupational exposure scenarios where for both financial reasons and to minimize disruption to industry, several measurements are typically taken in each workplace on the same day. Such measurements are likely to have very similar associated determinants of exposure and results in an overall dataset (for each DEO unit) that contains a number of ‘clusters’ of quite similar measurements. Furthermore, the need to measure a specific substance (for analytical reasons) limits the variation in situations in which measurements can be done. The measured substance might only be present in one kind of product designed for only a few specific purposes, thereby reducing the possible variation in determinants that relate to the product and the task.
A further complication is the use of different sampling methods to collect the data—some of the data were obtained using cotton gauze patches, others by activated charcoal or by skin washes. As each of these methods may give a different measure of dermal exposure, observed differences between exposure scenarios—that are attributed to differences in exposure determinants—could in principle result from a difference in the sampling methods. For modelling, the ideal would be to eliminate sampling differences altogether, but with a multicentre study covering a wide range of tasks and substances, some variation in sampling and analytical methods is inevitable. Instead, the statistical analysis could incorporate fixed effects for each distinct sampling methodology but for RISKOFDERM this approach was compromised by overlaps between the measurement methodology, physical characteristics of the substance and the exposure scenario.
Owing to all these limitations, the models are by no means ‘final’ from a scientific point of view. However, they are a substantial improvement in the modelling of dermal exposure to industrial chemicals over the earlier existing methods, being the TGD approach and the dermal part of EASE (ECB, 2003).
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Work part 3 of RISKOFDERM has resulted in useful models for estimating potential dermal exposure rates in the scope of risk assessment. However, several situations were not sufficiently covered in the measured datasets and several potential determinants could not be studied due to lack of variation within the sets, strong correlations between determinants or lack of practical methods to quantify them. Therefore, while the models provide a very important step forward, further studies in the future would lead to substantial improvements in parts of the model set.
Note. A software version of the final models is available. At the time of publication, the software version is in the format of a Microsoft EXCEL® spreadsheet available on CD-ROM from Dr J.J.H. Although this provides an easy means of using the models it should be remembered however that proper use of this dermal exposure model set requires expertise in occupational hygiene or exposure assessment and preferably some experience in dermal exposure measurement and assessment.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
The RISKOFDERM project was funded by the European Commission (project QLK4-CT-1999-01107). Additional funding came from the UK Health and Safety Executive and the Netherlands Ministry of Social Affairs and Employment.
Received January 24, 2005; in final form January 20, 2006
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
de Cock JS and van Drooge H. (2002) Field study on occupational exposure during spraying of biocidal products by pest control operators using deltamethrin and (ß)-cyfluthrin. TNO-report V3806. (TNO, Zeist, The Netherlands).
Delgado P, Porcel J, Abril I, et al. (2004) Potential dermal exposure during the painting process in car body repair shops. Ann Occup Hyg 48:229–36.
ECB. (2002) Technical notes for guidance, human exposure to biocidal products, guidance on exposure estimation. (European Commission. Joint Research Centre, Ispra, Italy).
ECB. (2003) 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. (European Commission. Joint Research Centre, ISPRA, Italy).
ECB. (2004) User guidance to human exposure to biocidal products (TnsG 2002). ISPRA. (European Commission. Joint Research Centre, Italy).
Eriksson K and Wiklund L. (2004) Dermal exposure to styrene in the fibreglass reinforced plastics industry. Ann Occup Hyg 48:203–8.
Eriksson K, Wiklund L, Larsson C. (2004) Dermal exposure to terpenic resin acids in swedish carpentry workshops and sawmills. Ann Occup Hyg 48:267–76.
EUROPOEM. (1996) The development, maintenance and dissemination of a European Predictive Operator Exposure Model (EUROPOEM) database. Draft final report. (TNO Bibra, Carshalton, UK).
Finkelstein M and Verma D. (2001) Exposure estimation in the presence of non-detectable values: another look. Am Ind Hyg Assoc J 62:195–8.
Fransman W, Vermeulen R, Kromhout H. (2004) Occupation dermal exposure to cyclophosphamide in dutch hospitals: a pilot study. Ann Occup Hyg 48:237–44.
Garrod ANI, Rimmer DA, Robershaw L, et al. (1998) Occupational exposure through spraying remedial pesticides. Ann Occup Hyg 42:159–65.
Garrod ANI, Guiver R, Rimmer DA. (2000) Potential exposure of amateurs (consumers) through painting wood preservative and antifoulant preparations. Ann Occup Hyg 44:421–6.
Goede HA, Tijssen SCHA, Schipper HJ, et al. (2003) Classification of dermal exposure modifiers and assignment of values for a risk assessment toolkit. Ann Occup Hyg 47:609–18.
Gijsbers JHJ, Tielemans E, Brouwer DH, et al. (2004) Dermal exposure during filling, loading and brushing with products containing 2-(2-butoxyethoxy)ethanol. Ann Occup Hyg 48:219–28.
HSE. (1999a) Exposure to chlorpyrifos in orchard spraying—technical development survey. (HSE, Liverpool, United Kingdom).
HSE. (1999b) Dermal exposure to non-agricultural pesticides—exposure assessment document EH74/3. (HSE, Liverpool, United Kingdom).
Hughson GW and Aitken RJ. (2004) Determination of dermal exposures during mixing, spraying and wiping activities. Ann Occup Hyg 48:245–56.
Johnson PD, Rimmer DA, Garrod ANI, et al. (2005) Operator exposure when applying amenity herbicides by all-terrain vehicles and controlled droplet applicators. Ann Occup Hyg 49:25–32.
Kromhout H, Fransman W, Vermeulen R, et al. (2004) Variability of task-based dermal exposure measurements from a variety of workplaces. Ann Occup Hyg 48:187–96.
Lansink CJM, Beelen MSC, Marquart J, et al. (1996) Skin exposure to calcium carbonate in the paint industry, preliminary modelling of skin exposure levels to powders based on field data. TNO-report V96.064. (TNO, Zeist, The Netherlands).
Llewellyn DM, Brazier A, Cocker J, et al. (1996) Occupational exposure to permethrin during its use as a public hygiene insecticide. Ann Occup Hyg 40:499–509.
Mäkinen M and Linnainmaa M. (2004a) Dermal exposure to chromium in the grinding of stainless and acid-proof steel. Ann Occup Hyg 48:197–202.
Mäkinen M and Linnainmaa M. (2004b) Dermal exposure to chromium in electroplating. Ann Occup Hyg 48:277–84.
Marquart J, Brouwer DH, Gijsbers JHJ, et al. (2003) Determinants of dermal exposure relevant for exposure modelling in regulatory risk assessment. Ann Occup Hyg 47:599–607.
Marquart J, Warren ND, Van Hemmen JJ. (2006) Default values for assessment of potential dermal exposure of the hands to industrial chemicals in the scope of regulatory risk assessments. Ann Occup Hyg 50:469–89.
MATLAB (Version 6.5). (2003) The Mathworks, Inc., Natick, MA, USA. 2003.
Oppl R, Kalberlah F, Evans PG, et al. (2003) A toolkit for dermal risk assessment and management: an overview. Ann Occup Hyg 47:629–40.
Rajan-Sithamparanadarajah R, Roff M, Delgardo P, et al. (2004) Patterns of dermal exposure to hazardous substances in European Union workplaces. Ann Occup Hyg 48:285–97.
RISKOFDERM. (2002a) Deliverable 30: Main Study Report of Partner 2. (TNO, Zeist, The Netherlands).
RISKOFDERM. (2002b) Deliverable 32: Main Study Report of Partner 4. TNO, Zeist, The Netherlands.
RISKOFDERM. (2003a) Deliverable 40: Benchmark Study Report of Partner 1. TNO, Zeist, The Netherlands.
RISKOFDERM. (2003b) Deliverable 41: Benchmark Study Report of Partner 2. TNO, Zeist, The Netherlands.
RISKOFDERM. (2003c) Deliverable 42: Benchmark Study Report of Partner 3. TNO, Zeist, The Netherlands.
Roff M, Bagon D, Chambers H, et al. (2004a) Dermal exposure to electroplating fluids and metal working fluids in the UK. Ann Occup Hyg 48:209–17.
Roff M, Bagon D, Chambers H, et al. (2004b) Dermal exposure to dry powder spray paints using PXRF and the method of Dirichlet tessellations. Ann Occup Hyg 48:257–65.
Rubin D. (1987) Multiple imputation for non-response in surveys. (Wiley, New York).
SAIC, Science Applications International Corporation. (1996) Occupational dermal exposure assessment, a review of methodologies and field data. (US EPA, Office of Toxic Substances, Washington, DC, USA).
Schuhmacher-Wolz U, Kalberlah F, Oppl R, et al. (2003) A toolkit for dermal risk assessment: toxicological approach and hazard characterisation. Ann Occup Hyg 47:641–52.
S-Plus (Version 6.0). (2001) Insightful Corporation, Seattle, USA.
van Hemmen JJ. (2004) Dermal exposure to chemicals. Ann Occup Hyg 48:183–5.
van Hemmen JJ, Auffarth J, Evans PG, et al. (2003) RISKOFDERM: risk assessment of occupational dermal exposure to chemicals. An introduction to a series of papers on the development of a toolkit. Ann Occup Hyg 47:595–8.
Van Wendel-de-Joode B, Brouwer DH, Vermeulen R, et al. (2003) DREAM; a method for semi-quantitative dermal exposure assessment. Ann Occup Hyg 47:71–87.
Warren N, Goede HA, Tijssen SCHA, et al. (2003) Deriving default dermal exposure values for use in a risk assessment toolkit for small and medium-sized enterprises. Ann Occup Hyg 47:619–28.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  Y. Liu, M. H. Stowe, D. Bello, J. Sparer, R. J. Gore, M. R. Cullen, C. A. Redlich, and S. R. Woskie Skin Exposure to Aliphatic Polyisocyanates in the Auto Body Repair and Refinishing Industry: III. A Personal Exposure Algorithm Ann. Hyg., January 1, 2009; 53(1): 33 - 40. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. Boogaard Biomonitoring as a tool in the human health risk characterization of dermal exposure Human and Experimental Toxicology, April 1, 2008; 27(4): 297 - 305. [Abstract] [PDF]
|
|
|
|
|
|
T. OGDEN Annals of Occupational Hygiene at Volume 50: Many Achievements, a Few Mistakes, and an Interesting Future Ann. Hyg., November 1, 2006; 50(8): 751 - 764. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. MARQUART, N. D. WARREN, J. LAITINEN, and J. J. VAN HEMMEN Default Values for Assessment of Potential Dermal Exposure of the Hands to Industrial Chemicals in the Scope of Regulatory Risk Assessments Ann. Hyg., July 1, 2006; 50(5): 469 - 489. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||