Ann. Occup. Hyg., Vol. 54, No. 4, pp. 443–452, 2010 Ó The Author 2010. Published by Oxford University Press on behalf of the British Occupational Hygiene Society doi:10.1093/annhyg/meq014
Development of a Task-Exposure Matrix (TEM) for Pesticide Use (TEMPEST) F. D. DICK1,*, S. E. SEMPLE1, M. VAN TONGEREN2, B. G. MILLER2, P. RITCHIE2, D. SHERRIFF1 and J. W. CHERRIE2 1
Environmental and Occupational Medicine, Population Health Section, Division of Applied Health Sciences, School of Medicine and Dentistry, Polwarth Building, University of Aberdeen, Foresterhill, Aberdeen, AB25 2ZD, UK; 2Institute of Occupational Medicine, Research Avenue North, Edinburgh, EH14 4AP, UK Received 1 May 2009; in final form 13 February 2010; published online 25 March 2010 Introduction: Pesticides have been associated with increased risks for a range of conditions including Parkinson’s disease, but identifying the agents responsible has proven challenging. Improved pesticide exposure estimates would increase the power of epidemiological studies to detect such an association if one exists. Methods: Categories of pesticide use were identified from the tasks reported in a previous community-based case–control study in Scotland. Typical pesticides used in each task in each decade were identified from published scientific and grey literature and from expert interviews, with the number of potential agents collapsed into 10 groups of pesticides. A pesticide usage database was then created, using the task list and the typical pesticide groups employed in those tasks across seven decades spanning the period 1945–2005. Information about the method of application and concentration of pesticides used in these tasks was then incorporated into the database. Results: A list was generated of 81 tasks involving pesticide exposure in Scotland covering seven decades producing a total of 846 task per pesticide per decade combinations. A Task-Exposure Matrix for PESTicides (TEMPEST) was produced by two occupational hygienists who quantified the likely probability and intensity of inhalation and dermal exposures for each pesticide group for a given use during each decade. Conclusions: TEMPEST provides a basis for assessing exposures to specific pesticide groups in Scotland covering the period 1945–2005. The methods used to develop TEMPEST could be used in a retrospective assessment of occupational exposure to pesticides for Scottish epidemiological studies or adapted for use in other countries. Keywords: exposure estimation; pesticides; task-exposure matrix
control study of genetic, environmental, and occupational risk factors for parkinsonism and Parkinson’s disease, showed a positive association between exposure to pesticides and risk of Parkinson’s disease and parkinsonism. However, exposure was assessed for all pesticides and no attempt was made to assess exposure to specific pesticide groups (Semple et al., 2004). While an association between all pesticide exposure and diseases such as Parkinson’s disease is suggestive of a causal link, it is far from conclusive. In addition, it is impractical for regulators to take action on the strength of such evidence, particularly
INTRODUCTION
Pesticides have been associated with a wide variety of ill-health outcomes ranging from neurological conditions such as Parkinson’s disease (Brown et al., 2006) through to developmental abnormalities of the male external genitalia (van Tongeren et al., 2002) or cancer (Samanic et al., 2008). For example, the Geoparkinson study (Dick et al., 2007), a multi-centre case– *Author to whom correspondence should be addressed. Tel: þ44(0)-1224-558191; fax: þ44(0)-1224-550925; e-mail:
[email protected] 443
444
F. D. Dick et al.
given the non-specific nature of the association. For such exposures to be considered causal, further corroboration is needed from toxicology and, ideally, from a more specific exposure assignment in the epidemiological analysis. Recognizing that better evidence of associations between pesticide exposure and ill health is needed, efforts to improve pesticide exposure estimates for occupational epidemiology have been made over a number of years. Approaches to exposure assessment include collection of work histories (job titles), job-exposure matrices (JEMs), expert assessment of work histories, and self-reports of exposure. The validity and reliability of these methods, when employed in case–control studies, have been extensively reviewed (Teschke et al., 2002). One approach to assessing exposure has been the use of job titles as exposure surrogates. Self-reported work histories can be an efficient means of assessing exposure where job titles reflect tasks with similar exposures but this method has important limitations, when applied to farming, because job titles such as ‘farmer’ encompass a range of tasks with widely varying pesticide exposures. Recent work has shown that farm-related job titles are a poor surrogate for pesticide exposure, with over three-quarters of farm jobs being assessed as having no likelihood of pesticide exposure when considered by an occupational hygienist (MacFarlane et al., 2009). JEMs have been developed for epidemiological studies of pesticide effects but as they rely on job titles, these can lead to exposure misclassification. JEMs have at least two axes, one covering a range of jobs and the other axis being the agents of interest. Some matrices have a third axis, time, to allow for changes in work practices or agents over the study period. The cells of the matrix are then populated with exposure estimates that may indicate exposure (exposed/ unexposed), exposure ranking (low/medium/high), or the probability of exposure. Some JEMs are based on quantitative exposure data but many are based on the judgement of the researchers involved. In general, study-specific JEMs have been found to have higher specificity and sensitivity than generic JEMs (Teschke et al., 2002). Tielemans and his colleagues have argued for job-specific data, especially in situations of low exposure (Tielemans et al., 1999). An extension of the JEM is the crop- or taskspecific exposure matrix, which takes account of exposures during specific tasks within jobs and such matrices have been created for pesticide studies in Europe (Daures et al., 1993; Miligi et al., 1993), South Africa (London and Myers, 1998), and California (Young et al., 2004).
Some studies have relied on self-reported pesticide exposures and have shown reasonable agreement (60% for use of herbicides and insecticides) with suppliers’ reports of pesticide use (Blair and Zahm, 1993) and correlation with biomonitoring results (Hertzman et al., 1988). Self-reports of pesticide exposure are more reliable where individuals have been involved in the purchase or application of pesticides and so farmers would be expected to have better recall of agents applied than farm labourers (Young et al., 2004). Expert assessment of pesticide exposures is generally seen as the best approach to exposure estimation where reliable biomonitoring data are not available. Several studies have employed expert assessment, either to generate pesticide exposure estimates (de Cock et al., 1996; Garcı´a et al., 2000) or to validate a JEM or crop-exposure matrix (Kauppinen et al., 1992; Miligi et al., 1993). While most exposure assessments generate subjective rankings or semi-quantitative estimates of exposure, a recent development has seen efforts to quantify exposures. The Agricultural Health Study used an algorithm, drawing on individual questionnaire responses on exposure determinants together with published pesticide exposure data to estimate exposures (Dosemeci et al., 2002). In contrast, Canadian researchers developed an agricultural JEM covering physical, chemical, and biological exposures that employed crop-specific data (Wood et al., 2002). They used growers’ production guides for British Columbia, published over many decades by the Ministry of Agriculture and Food, to identify pesticides recommended for use on specific crops over time. This informed the development of a questionnaire used to interview experienced farmers regarding pesticide usage in a range of crops and animals. These interviews were used to describe 36 agricultural tasks in British Columbia. That JEM had three axes: work type (a combination of agricultural region, crop, job title, and task), exposure agent [for pesticides, these were categorized by primary function and then by pesticide category e.g. insecticide, organochlorine (OC)], and time with an exposure value generated by experts provided within each cell of the matrix. Exposures were expressed quantitatively, where data allowed, or qualitatively where data were lacking (exposed/unexposed). For some pesticides, the skin is the main route of exposure and so it is important to assess both inhalational and dermal exposure. A Dutch study estimated long-term pesticide exposure by developing an exposure index that sought to capture both dermal and inhalational exposures. The authors combined database
Task-exposure matrix for pesticides
records of pesticide usage with data from field surveys to produce an exposure index (Brouwer et al., 1994). Improvements in exposure assessment for specific pesticide groups could result in identification of pesticides that are associated with an increased risk of conditions such as Parkinson’s disease. This would also enable regulators to take any necessary specific actions to protect public health. This study sought to develop a task-exposure matrix (TEM) for specific pesticides or groups of pesticides in Scotland. This provides a framework that could be utilized for retrospective exposure assessment in other countries and for epidemiological studies examining the relationship between pesticide exposure and ill-health outcomes. METHODS
Overview of the approach The development of the TEM involved a number of steps, as shown in the flow chart in Fig. 1. First, broad categories of pesticide use, e.g. insecticides, fungicides etc, were identified from the information reported in a recent community-based epidemiological study (Scottish Geoparkinson study; Dick et al., 2007) and supplemented by a literature review. A list was then generated of tasks involving pesticide exposure in Scotland from 1945 to 2005. Tasks were defined according to typical activities where pesticides were used. In general, tasks were defined by the crop or livestock being treated, the pest or plant that was to be eradicated and, in some cases, the time of year. For example task AF8 was ‘Winter/Spring Oats weed control’ and task AL1 applied to ‘Fly control for cattle’. Tasks are listed in full on the project website at http://www.iom-world.org/research/tempest/. Tasks were subsequently grouped into eight categories, e.g. amenity uses, arable farming etc.
445
The next step was to determine the typical pesticides used in each task in each decade in which they were used. This involved a search of published and grey literature and expert interviews. A pesticide usage database was constructed incorporating all of the available information about pesticide usage, work tasks, and the concentration of pesticide active ingredients used along with the pesticide application methods. Finally, exposure assessment was performed independently by two occupational hygienists for dermal and inhalational exposures separately. Categories of pesticide use and production of task list Review of 290 Scottish Geoparkinson occupational history questionnaires (Dick et al., 2007) identified pesticide exposure tasks involving hobby activities (n 5 237), occupational agricultural tasks (n 5 62), and occupational amenity tasks (n 5 18). The amenity category included activities such as timber preservation, parks, sports ground and golf course maintenance, vermin control, and weed control on roads, paths, and around buildings. This information was used as the basis of the eight task groups for the TEM, which was expanded to include further activities, not reported in the Geoparkinson study, during which exposure to pesticides was likely. Two tasks (rat/mouse control and rabbit control using warfarin or gas) were removed from the task groups as the agents’ used in these processes had mechanisms of action that were quite different from other pesticides. The final list of 81 tasks in eight task groups is summarized in Table 1. Identifying and classifying crops and pesticide application Next, the typical pesticides used for a given agricultural task in a given decade in Scotland were
Fig. 1. Flow chart showing development of the TEM.
446
F. D. Dick et al.
determined. Selecting a year to represent each decade avoided the dataset becoming unmanageably large. To establish which crops were likely to have been treated, pesticide usage data for 1950, 1960, 1970, and 1980, representing their decades, were obtained from the Analytical Services Statistics Branch, Scottish Executive’s Environment and Rural Affairs Department (D. Rowley, personal communication). For 1990 and 2000, Central Science Laboratory pesticide usage reports (http://www.fera.defra.gov.uk/plants/ pesticideUsage/) were used. These reports give information on the active ingredients employed, the crops treated, and the area (in hectares) covered. Grasslands were excluded as, in Scotland, pesticide applications to them were, and continue to be, rare. Pesticide application methods were generally specific to the task being described and included open spraying from tractors, spraying using low-pressure back-pack systems, handling and laying of powders, painting, dipping, and fogging systems. Table 1. Eight task groups—codings and descriptions Task group code
Task group description
Number of tasks in group
AF
Arable farming
41
AL
Agricultural livestock
AM
Amenity
AQ DI
Aquaculture Domestic indoor
FY
Forestry
7
GA
Gardening
5
OC
Other occupational not elsewhere classified
9
4 10 1 4
Identifying and classifying pesticides To augment the information described above, a comprehensive literature search was conducted to identify pesticides used in Scotland over the decades. Where literature gaps were identified, knowledge was supplemented by interviewing pesticide experts using semi-structured questions designed for the purpose. Insecticides were classed into OC, organophosphates (OP), carbamates, and ‘other’ (consisted primarily of botanicals and pyrethroids). Subdivision to the level of specific active ingredients proved impractical because of the limited information available. Herbicides were classified into 24 groups by their mode of action, using the HRAC system (HRAC, 2005). To avoid an unmanageably large number of herbicide categories in subsequent analyses, these were combined into four classes. We selected three commonly used HRAC herbicide groups: C1 which includes atrazine; D which includes diquat and paraquat; and O which includes 2,4-D and mecoprop. The remaining 21 groups were collapsed into a fourth ‘Other’ herbicides class. Fungicides were classed into carbamates (which includes dithiocarbamates—the largest group of fungicides by quantity sold) and ‘other’ fungicides. This resulted in the coverage of pesticide types being used in TEMPEST shown in Table 2. The pesticides listed by pesticide class in Table 3 were chosen to be typical examples used for the range of application tasks over the seven decades. These typical pesticides within each pesticide class were then combined with tasks to generate the pesticide usage database.
Table 2. Pesticide coverage in TEMPEST Pesticide use
Pesticide class
Examples in this class
Insecticides
OC
Aldrin, DDT, dieldrin, and chlordane
Insecticides
OP
Malathion, parathion, diazinon, and methylparathion
Insecticides
Carbamates
Carbaryl, carbofuran, and methylcarbamate
Insecticides
Other insecticides
Herbicides
C1
Permethrin, lambda-cyhalothrin, rotenone, and nicotine Simazine, bromacil, atrazine, lenacil, and terbacil
Herbicides
D
Diquat and paraquat
Herbicides
O
2,4-D, dichlorprop, MCPA, mecoprop, fluroxypyr, and clopyralid
Herbicides
Other herbicides
Glyphosate, linuron, diuron, and dinoseb.
Fungicides Fungicides
Carbamate fungicides Other fungicides
Mancozeb and maneb Copper sulphate and benzimidazole
Avicides
Assorted pesticides used
Rabbit control
Assorted pesticides used
Rodenticides
Assorted pesticides used
DDT, dichlorodiphenyltrichloroethane; MCPA, 2-methyl-4-chlorophenoxyacetic acid.
Task-exposure matrix for pesticides
Combining pesticides with tasks The task list and pesticide use groups were combined to produce a pesticide usage database for each task in a given decade. The concentration of pesticides used in applications was determined where possible: information concerning hobby and amenity pesticide concentrations was not available. The Pesticide Usage Reports for Scotland permitted the compilation of detailed data on agriculture from the 1960s onwards in the pesticide usage database. The pesticide usage database was then used to populate the task-pesticide matrix. For each decade, the probability of a particular pesticide class being used for the given task was assigned. The total probability for all pesticide classes used for a given task within a decade always summed to 100%. This gave a total of 846 task groups per pesticide class per decade combinations where there was a probability of pesticide use that was .0%. Each of these was then scored for both dermal and inhalation exposure, resulting in a total of 1692 exposure assessments being made by each assessor. Expert exposure estimation For both inhalation and dermal exposure routes, for all task groups across the seven decades, two occupational hygienists (S.S. and M.v.T.) independently assessed and assigned a score for the intensity of exposure implied by the task and pesticide usage combinations, based on information about active ingredients, amount used per hectare, likely application technique, and the probable use of personal protective equipment. Scores were assigned on a five-level scale (no exposure, very low, low, medium, and high), according to the guidance shown in Table 4.
447
Table 3. Pesticides and application groups used in TEMPEST Typical pesticide name
WELa for typical pesticide in this class (8-h TWA)
Pesticide use
Pesticide class
Avicide Avicide
Carbamates Propoxur OC Endosulfan
0.5 mg m3 0.1 mg m3
Fungicide
Carbamates Propoxur
0.5 mg m3
Fungicide
OC
Endosulfan
0.1 mg m3
Fungicide
Other
Captan
5 mg m3
Herbicide
C1
Bromacil
11 mg m3
Herbicide
D
Diquat dibromide
0.5 mg m3
Herbicide
O
2,4-D
10 mg m3
Herbicide
Other
Diuron
10 mg m3
Insecticide
Carbamates Propoxur
0.5 mg m3
Insecticide
OC
Endosulfan
0.1 mg m3
Insecticide
OP
Chlorpyrifos 0.2 mg m3
Insecticide
Other
Pyrethrins
1 mg m3
Rabbit control OC Rabbit control Other
Endosulfan Pyrethrum
0.1 mg m3 1 mg m3
Rodenticide
OC
Endosulfan
0.1 mg m3
Rodenticide
Other
Pyrethrum
1 mg m3
a WEL—workplace exposure limits published by HSE in EH40/2005 (ISBN 0717629775).
Table 4. Guidance for assigning inhalation and dermal exposure scores for pesticides Inhalation exposure score
Description
Corresponding concentrations
0
No exposure
1
Very low
2
Low
6–25% WEL
3
Medium
26–100% WEL
4
High
.100% WEL
Dermal exposure score
Description
Skin loading (based on Cattani et al., 2001)
0–5% WELa
0
No exposure
(Or no Sk notationb)
RESULTS
1
Very low
0–1 mg h1
The two hygienists’ independent assignments agreed completely on 51% (n 5 862) of assessments, with a further 41% (n 5 698) differing by just one level on the five-level scale. Eight percent (n 5 132) of cases differed by two or three levels, and no assessments were four levels apart. Agreement was higher for dermal exposure scores (62% total agreement) than for inhalation assessments (42%). Inhaled exposures were also more likely to differ by more than one level—11% compared to 5% for dermal exposure ratings. There was no clear pattern to the tasks that produced disagreements between the two hygienists, with examples from weed and fly control in agricultural settings to timber pres-
2
Low
1–10 mg h1
3 4
Medium High
11–100 mg h1 .100 mg h1
a WEL—workplace exposure limits published by HSE in EH40/2005 (ISBN 0717629775) using the WEL of the typical pesticide in each group as shown in Table 3. b Sk notation—‘. . . some substances have the ability to penetrate intact skin and become absorbed into the body, thus contributing to systemic toxicity; these substances are marked with a ‘‘Sk’’ notation’. Workplace exposure limits EH40/2005, page 43, HSE (2005).
ervation. In general, there were more disagreements over tasks from the earliest period (1940–1960), when information on the activity was limited.
448
F. D. Dick et al.
Where there was a difference of more than one level (n 5 132), the two hygienists discussed this task in detail and modified one or both of their assessments so they either agreed (n 5 18) or differed by no more than one level (n 5 114). This produced a TEM that had complete agreement in 52% (n 5 880) of scores and –1 level for the remainder 48% (n 5 812). These final scores are shown and compared in Table 5. Where there remained a difference of one level, the final agreed TEM used the maximum of the two assessments. Table 6 compares the final assigned scores in the TEM, for inhalation and dermal exposures. The distribution of each of these is seen in the marginal totals. In all, 30% of all dermal assessments were assigned a zero score, while none of the inhalation assessments were assigned a zero. Overall, there was agreement in 274 cases (32%), while 17% Table 5. Comparison of final scores assigned to tasks by each hygienist individually (S.S., M.v.T.), following consultation Inhalation Inhalation exposure score by M.v.T. Inhalation exposure score by S.S. None Very Low Low Medium High Total None Very low Low Medium High Total
0 0 0 0 0 0
40 205 173 0 0 418
0 45 83 72 0 200
0 0 113 41 27 181
0 0 0 19 28 47
40 250 369 132 55 846
Dermal Dermal Dermal exposure score by M.v.T. exposure score by S.S. None Very Low Low Medium High Total None Very low Low Medium High Total
253 1 0 0 0 254
0 32 66 0 0 98
0 29 93 30 0 152
0 0 66 96 28 190
0 0 0 101 51 152
253 62 225 227 79 846
Table 6. Cross-classification of final assignments of inhalation and dermal exposure scores Final inhalation Final dermal exposure score exposure score None Very low Low Medium High Total No exposure Very low
0
0
0
0
0
0
112
25
52
26
30
245
Low
89
8
113
60
31
301
Medium
52
0
23
84
67
226
High
0
0
0
22
52
74
Total
253
33
188 192
180
846
agreement would be expected by chance: the kappa statistic was 0.186, representing ‘slight’ agreement (Landis and Koch, 1977). This reflects the varying importance of these exposure routes for the different pesticide groups. Trends in inhalation and dermal exposure intensities over time The TEMPEST database shows trends in pesticide exposure over time. This is seen in the introduction of a growing number and type of pesticides from the 1970s onwards and also in terms of the assigned exposure scores. Fig. 2 shows the proportion of tasks assigned to each of the inhalation and dermal exposure scores. There was an increase in the number of pesticide application tasks scored as medium or high for dermal exposure from the 1940s through to the 1960s, with .80% of tasks being rated as medium or high. From 1970 onwards, with increasing knowledge and regulation, this decreased until the last decade when .96% of tasks had dermal exposures scored as low, very low, or zero. For inhaled exposure, the picture was similar. Fig. 2 also shows that two-thirds of tasks were scored as medium or high in the 1960s and 1970s while this had reduced to 12% by the 1990s and then to only 2% by the 2000 decade. Worked example We illustrate the use of the TEMPEST database in an exposure reconstruction scenario in a worked example. An individual who reported being a soft-fruit (raspberry) farmer between 1955 and 1989 is likely to have applied pesticides for weed, insect, and fungi control. The database indicates that pesticides were generally only introduced in this sector from the 1970s onwards so we only need to consider the individual’s exposure from 1970 to 1989 (in TEMPEST, the two decades beginning 1970 and 1980). The TEM for the three tasks in these two decades extracted from TEMPEST is shown in table 7. The most common pesticide class used for insect control in raspberries in the 1970s was the ‘Other’ class of insecticides (71% of applications) with Pyrethrins being the example pesticide for this category. This changes to OP-based insecticides (99% of applications) by the 1980s. The inhaled exposure is scored at 3 (medium exposure, 26–100% of the workplace exposure limit) for Pyrethrins during the 1970s, with no additional contribution from dermal exposure, as insecticides of this class are generally unlikely to be
Task-exposure matrix for pesticides
449
Fig. 2. Trends in the estimated intensity level of dermal and inhalation exposures to pesticides in pesticide application tasks in Scotland over seven decades from 1940. Table 7. Example TEM for soft-fruit (raspberry) farmer 1955–1989 Task
Pesticide group
Pesticide class
Typical pesticide name
Skin 1970s 1980s notation Probability Inhalation Dermal Probability Inhalation Dermal of use (%) score score of use (%) score score
OC
Endosulfan
Yes
9
4
3
0
0
0
OP
Chlorpyrifos Yes
20
3
3
99
3
3
Carbamates
Propoxur
Yes
0
0
0
1
3
3
Other
Pyrethrins
No
71
3
0
0
0
0
Bromacil Diquat dibromide
No Yes
65 28
3 4
0 3
32 31
2 3
0 3
AF36 Raspberries insect control Insecticide
AF37 Raspberries weed control Herbicide
C1 D O
2,4-D
Yes
1
3
3
2
2
3
Other
Diuron
No
6
3
0
35
2
0
Propoxur Captan
Yes Yes
0 100
0 3
0 3
0 100
0 2
0 3
AF38 Raspberries fungi control Fungicide
Carbamates Other
absorbed across the skin barrier. Inhaled exposure is also scored 3 for the OP-based insecticides used in the 1980s with dermal exposure and uptake also scored 3 (medium exposure, 11–100 mg h1) during this decade. A similar exercise can be carried out for the other raspberry pesticide applications for fungi and weed control. With information on how often these tasks were carried out per year, an estimate
can be made of the cumulative exposure experienced by the individual over their working lifetime. DISCUSSION
This study constructed TEMPEST, a TEM for specific pesticide classes in Scotland covering the period 1945–2005. The development of the TEM has
450
F. D. Dick et al.
highlighted the information to be gathered during any future Scottish epidemiological study of Parkinson’s disease and pesticide exposure and provides a template for developing a questionnaire to collect data that will allow more accurate exposure reconstructions. TEMPEST has a number of strengths and limitations. Strengths of this approach are that explicit pesticide class-specific semi-quantitative exposure estimates are provided and could be validated in future studies. The construction of the matrix required that expert decisions be made as to the most likely agent employed, the likely application method, and the probable quantities used. These decisions were informed, so far as possible, by available data on pesticide usage, application methods, and crops grown. However, expert assessments, in particular for earlier decades, may have resulted in some degree of exposure misclassification in the TEM. One challenge in undertaking epidemiological studies on the health effects of pesticides is the generation of valid estimates of exposure in a heterogeneous group of workers: farmers. In the absence of valid personal biomonitoring results, investigators may use work histories, a JEM, expert assessment, or self-reported exposure to determine pesticide exposure. Both JEMs and expert assessment of individual work histories ultimately rely on expert judgement. A major concern with exposure estimation is exposure misclassification and the consequent loss of statistical power. It is rarely possible to quantify the extent of such misclassification that, if non-differential, will usually bias any association towards the null. It has been argued that the magnitude of exposure misclassification when using a JEM is dependent on the population prevalence of the exposure of interest and the threshold for defining exposure (Kauppinen et al., 1992). Simulations of misclassification errors suggest that well-designed JEMs perform close to expert assessments of exposure, based on individual job histories (Bouyer et al., 1995). The issue of exposure misclassification when using a generic exposure assessment was explored by Acquavella and colleagues (Acquavella et al., 2006) in the Farm Family Exposure Study. That study compared pesticide exposure estimates generated using an exposure algorithm (Dosemeci et al., 2002), developed for the Agricultural Health Study, with the results of urinary biomonitoring, and found only moderate correlations. The authors concluded that a generic exposure assessment reliant on questionnaire data would likely result in significant exposure misclassification. Experts may disagree in their exposure assessments: one study (Garcı´a et al., 2000) found that
the agreement between experts, both with regard to pesticide exposure and exposure intensity, was only fair (kappa of 0.36 and 0.39, respectively) and for some situations, the agreement between experts was little different from chance. Another study explored experts’ subjective ratings of likely pesticide exposure and found that, while occupational hygienists and pesticide experts could rank pesticide exposures in fruit growing, there was significant variation in the estimates generated by experts from the same professional group (de Cock et al., 1996). The authors concluded that such expert assessments should be combined with data on task duration and information on the physicochemical properties of pesticides when generating exposure estimates. The latter conclusion received support when the Farm Family Exposure Study (Acquavella et al., 2006) showed that the type of chlorpyrifos formulation (liquid versus granular) influenced the correlation between algorithm predicted pesticide exposure (Dosemeci et al., 2002) and biomonitoring results. Pesticide exposures vary over time reflecting many factors including changing work practices, the efficacy, availability and relative cost of agents, pesticide formulation (Acquavella et al., 2006), the geographical region, and the size of the agricultural holding (Wood et al., 2002). Given the diversity of farming, then, JEM, crop-, or task-exposure matrices developed in one country or region may not be valid if applied, unmodified, in another. Efforts to establish the validity of both expert assessment and JEMs have been hampered by the lack of historical pesticide usage data, the substantial cost of biomonitoring, intra, and inter-individual variability of exposures, and the lack of biomarkers of long-term exposure (OC are a notable exception owing to their biopersistence). One study combined self-administered questionnaires and workplace surveys to assess pesticide exposure in greenhouse workers and found significant variations in exposure both between work sites and between workers (Tielemans et al., 2007). Some studies have employed repeated measures of personal pesticide exposure and found higher variability of exposure within workers than between workers (de Cock et al., 1998; Simcox et al., 1999). There is a need for more research into the validity of JEM and task-exposure matrices. Pesticide exposure control has improved in the last seven decades. In addition to changes in pesticide type, there have been significant changes to application methods, driven by efficiency in application of an expensive material and, more recently, by human and environmental health concerns. Exposures to most chemicals (more generally than pesticides)
Task-exposure matrix for pesticides
appear to have decreased by 2–10% per year (Creely et al., 2007). Our exposure estimates reflected these changes. For example the percentage of assessments in the highest two exposure intensity levels (3 and 4) fell from 54% of all assessments in the 1950s decade to 1% in the 2000s decade. Assumptions were made about application methods, personal protective equipment use, and knowledge influencing worker behaviour. Hand-held applications in arable farming reduced with larger fields and more spraying using tractors. Ventilated tractor cabins with air filtration are now widespread. Similarly, protective clothing is now the norm for workers applying large quantities of pesticide in all sectors. Application methods changed least in the hobby sector although formulations changed, concentrations decreased, and the potential for dermal uptake fell. The results from exposure assessment methods used in the original Geoparkinson study (Dick et al., 2007) differed from those generated by applying TEMPEST. The Geoparkinson study used a simple JEM to quantify exposure to any pesticide and then modified this base estimate using any relevant data from the exposure interview (Semple et al., 2004). These exposure modifiers included descriptions of personal protective equipment use, quantities of pesticide used, and reports of symptoms related to pesticide use. TEMPEST was more detailed in that it allowed particular pesticide classes to be assigned to a task and provided exposure intensities for each pesticide class for both dermal and inhalation routes. Importantly, however, TEMPEST did not take into account any individual information on jobs obtained from interview of a study participant. Incorporation of such data would allow modification of the TEMPEST-generated exposure assessments. We would recommend that any future pesticide epidemiological studies use a locally generated version of TEMPEST combined with personally derived work-history exposure modifiers, in order to capture inter- and intra-worker variability. CONCLUSION
This project successfully developed a TEM for specific pesticide categories in Scotland for the period 1945–2005. One area for future work is the validation of TEMPEST against measured exposures in field conditions. TEMPEST was constructed for Scotland and may not be directly applicable to other countries with different work practices, crops, and pests (van Tongeren et al., 2002) but could be adapted, given appropriate data, to produce pesticide TEMs for other
451
countries or areas. TEMPEST is also likely to offer improved pesticide exposure estimation over generic JEMs and could be employed in any future epidemiological studies examining the relationship between ill health and pesticide exposure in Scotland. FUNDING
Department for Environment Food and Rural Affairs (PS2610). Acknowledgements—Information on The TEMPEST Study— epidemiology of the association between pesticides and Parkinson’s Disease (final report available at http://randd.defra .gov.uk/Default.aspx?Menu5Menu&Module5More&Location 5None&Completed50&ProjectID513742). Further information on TEMPEST is available at http://www.iom-world.org/research /tempest/. We gratefully acknowledge the contribution of the many experts from industry and government agencies who gave so freely of their time to assist us in developing the TEM.
REFERENCES Acquavella JF, Alexander BH, Mandel JS et al. (2006) Exposure misclassification in studies of agricultural pesticides: insights from biomonitoring. Epidemiology; 17: 69–74. Blair A, Zahm SH. (1993) Patterns of pesticide use among farmers: implications for epidemiologic research. Epidemiology; 4: 55–62. Bouyer J, Dardenne J, Hemon D. (1995) Performance of odds ratios obtained with a job-exposure matrix and individual exposure assessment with special reference to misclassification errors. Scand J Work Environ Health; 21: 265–71. Brouwer DH, Brouwer EJ, van Hemmen JJ. (1994) Estimation of long-term exposure to pesticides. Am J Ind Med; 25: 573–88. Brown TP, Rumsby PC, Capelton AC et al. (2006) Pesticides and Parkinson’s disease—is there a link? Environ Health Perspect; 114: 156–64. Cattani M, Cena K, Edwards J et al. (2001) Potential dermal and inhalation exposure to chlorpyrifos in Australian pesticide workers. Ann Occup Hyg; 45: 299–308. Creely KS, Cowie H, Van Tongeren M et al. (2007) Trends in inhalation exposure—a review of the data in the published scientific literature. Ann Occup Hyg; 51: 665–78. Daures JP, Momas I, Bernon J et al. (1993) A vine-growing exposure matrix in the He´rault area of France. Int J Epidemiol; 22 (Suppl 2): S36–41. de Cock J, Heederik D, Kromhout H et al. (1998) Exposure to captan in fruit growing. Am Ind Hyg Assoc J; 59: 158–65. de Cock J, Kromhout H, Heederik D et al. (1996) Experts’ subjective assessment of pesticide exposure in fruit growing. Scand J Work Environ Health; 22: 425–32. Dick FD, De Palma G, Ahmadi A et al. (2007) Environmental risk factors for Parkinson’s disease and parkinsonism: the Geoparkinson study. Occup Environ Med; 64: 666–72. Dosemeci M, Alvanja M, Roland AS et al. (2002) A quantitative approach for estimating exposure to pesticides in the Agricultural Health Study. Ann Occup Hyg; 46: 245–60. Garcı´a AM, Orts E, Esteban V et al. (2000) Experts’ assessment of probability and level of pesticide exposure in agricultural workers. J Occup Environ Med; 42: 911–6.
452
F. D. Dick et al.
HRAC. (2005) Classification of herbicides according to mode of action. Available at http://www.plantprotection.org/HRAC/. Accessed 31 January 2008. Hertzman C, Teschke K, Dimich-Ward H et al. (1988) Validity and reliability of a method for retrospective evaluation of chlorophenate exposure in the lumber industry. Am J Ind Med; 14: 703–13. HSE. (2005) Workplace exposure limits. EH40/2005. Sudbury, Suffolk, UK: HSE Books. ISBN 0717629775. Kauppinen TP, Mutanen PO, Seitsamo JT. (1992) Magnitude of misclassification bias when using a job-exposure matrix. Scand J Work Environ Health; 18: 105–12. Landis JR, Koch GG. (1977) The measurement of observer agreement for categorical data. Biometrics; 33: 159–74. London L, Myers JE. (1998) Use of a crop and job specific exposure matrix for retrospective assessment of long term exposure in studies of chronic neurotoxic effects of agrichemicals. Occup Environ Med; 55: 194–201. MacFarlane EM, Glass DC, Fritschi L. (2009) Is farm-related job title an adequate surrogate for pesticide exposure in occupational cancer epidemiology? Occup Environ Med; 66: 497–501. Miligi L, Settimi L, Masala G et al. (1993) Pesticide assessment: a crop exposure matrix. Int J Epidemiol; 22 (Suppl 2): S42–5. Samanic CM, De Roos AJ, Stewart PA et al. (2008) Occupational exposure to pesticides and risk of adult brain tumors. Am J Epidemiol; 167: 976–85. Semple SE, Dick F, Cherrie JW. (2004) Exposure assessment for a population-based case-control study combining
a job-exposure matrix with interview data. Scand J Work Environ Health; 30: 241–8. Simcox NJ, Camp J, Kalman D et al. (1999) Farmworker exposure to organophosphorus pesticide residues during apple thinning in Central Washington State. Am Ind Hyg Assoc J; 60: 752–61. Teschke K, Olshan AF, Daniels JL et al. (2002) Occupational exposure assessment in case-control studies: opportunities for improvement. Occup Environ Med; 59: 575–94. Tielemans E, Bretveld R, Schinkel J et al. (2007) Exposure profiles of pesticides among greenhouse workers: implications for epidemiological studies. J Expo Sci Environ Epidemiol; 17: 501–9. Tielemans E, Heederik D, Burdorf A et al. (1999) Assessment of occupational exposures in a general population: comparison of different methods. Occup Environ Med; 56: 145–51. van Tongeren M, Nieuwenhuijsen MJ, Gardiner K et al. (2002) A job-exposure matrix for potential endocrine-disrupting chemicals developed for a study into the association between maternal occupational exposure and hypospadias. Ann Occup Hyg; 46: 465–77. Wood D, Astrakianakis G, Lang B et al. (2002) Development of an agricultural job-exposure matrix for British Columbia, Canada. J Occup Environ Med; 44: 865–73. Young HA, Mills PK, Riordan D et al. (2004) Use of a crop and job specific exposure matrix for estimating cumulative exposure to triazine herbicides among females in a casecontrol study in the Central Valley of California. Occup Environ Med; 61: 945–51.
