Background This paper describes methods developed to assess occupational exposure to pesticide active ingredients and chemical groups, harmonised across cohort studies included in the first AGRICOH pooling project, focused on the risk of lymph-haematological malignancies.
Methods Three prospective agricultural cohort studies were included: US Agricultural Health Study (AHS), French Agriculture and Cancer Study (AGRICAN) and Cancer in the Norwegian Agricultural Population (CNAP). Self-reported pesticide use was collected in AHS. Crop-exposure matrices (CEMs) were developed for AGRICAN and CNAP. We explored the potential impact of these differences in exposure assessment by comparing a CEM approach estimating exposure in AHS with self-reported pesticide use.
Results In AHS, 99% of participants were considered exposed to pesticides, 68% in AGRICAN and 63% in CNAP. For all cohorts combined (n=316 270), prevalence of exposure ranged from 19% to 59% for 14 chemical groups examined, and from 13% to 46% for 33 active ingredients. Exposures were highly correlated within AGRICAN and CNAP where CEMs were applied; they were less correlated in AHS. Poor agreement was found between self-reported pesticide use and assigned exposure in AHS using a CEM approach resembling the assessment for AGRICAN (κ −0.00 to 0.33) and CNAP (κ −0.01 to 0.14).
Conclusions We developed country-specific CEMs to assign occupational exposure to pesticides in cohorts lacking self-reported data on the use of specific pesticides. The different exposure assessment methods applied may overestimate or underestimate actual exposure prevalence, and additional work is needed to better estimate how far the exposure estimates deviate from reality.
Statistics from Altmetric.com
What this paper adds
To investigate associations between rare pesticide exposures and rare health outcomes, pooling of data from agricultural cohort studies is a necessity.
Occupational exposure to 14 pesticide chemical groups and 33 active ingredients was assessed across three cohort studies included in the AGRICOH pooling project on lymph-haematological malignancies. Country-specific crop-exposure matrices (CEMs) were created to assign exposure in the two cohorts that did not collect data on the use of specific pesticides.
Pesticide exposures were highly correlated where CEMs were applied and agreement between self-reported pesticide use and assigned exposure using a CEM approach was poor.
Pooling of data from different agricultural cohort studies with distinct designs and detail regarding pesticide exposure is possible. However, results from subsequent epidemiological analyses should be interpreted cautiously due to potentially substantial exposure misclassification.
Agricultural workers are of specific interest in occupational epidemiology as they show decreased risk of some diseases and excess risk for others, possibly due to experiencing a wide range of specific work-related exposures (eg, pesticides, diesel exhaust, dust and endotoxins) and conditions (eg, physical activity, outdoor work).1–5 Pesticides are one of the most studied exposures for farmers and farmworkers. The biological activity of pesticides may impact both target pests and human health and there are plausible hypotheses on mechanisms through which pesticides could be involved in different adverse health effects.
Accurate assessment of exposure to pesticides is critical for epidemiological studies to further investigate suggested associations with health outcomes, but this is a major challenge. Biological measurements often do not allow quantification of past pesticide exposure, as most pesticides and their metabolites are not persistent in the human body. Many studies have used occupational classifications (eg, farmer, farmworker) as a proxy for exposure to pesticides, or assessed exposure in broad classifications such as ‘herbicides’. However, such metrics of pesticide exposure are crude and often not specific enough to offer useful information on the chemicals that represent potential health risks. A few attempts have been made to analyse data considering exposure to chemical groups of pesticides or individual active ingredients, and some interesting findings have emerged from these studies.6 ,7 For example, results from the US Agricultural Health Study (AHS) are suggestive of associations between specific pesticides and some cancers, including leukaemia, non-Hodgkin lymphoma and multiple myeloma.8
Individual studies often lack statistical power to investigate associations between specific pesticide exposures and health outcomes due to rare exposures (eg, use of infrequently applied pesticides), rare health outcomes (eg, cancer subtypes, specific neurological or autoimmune diseases) or a combination of both.9 Therefore, the AGRICOH consortium was established with the aim of promoting and sustaining collaboration between agricultural cohort studies and enabling data sharing and pooled analyses.10 The AGRICOH consortium currently consists of 28 prospective cohort studies from 12 countries.11 However, the type of pesticide exposures or exposure proxies investigated varies from one study to another. These differences in exposure assessment, and further variation in agricultural practices between and within countries, pose a challenge for pooling of data across studies.
Here, we describe the development of cohort-specific methods to assess exposure to selected pesticide chemical groups and individual active ingredients in three prospective cohort studies from France, Norway and the USA, included in the AGRICOH pooling project on lymph-haematological malignancies. The information on pesticide exposures collected in each cohort is discussed, as well as the harmonisation efforts resulting in a set of common exposures and their prevalence across the cohorts. Furthermore, potential exposure misclassification arising from these cohort-specific exposure assessment methods is explored. The focus on lymph-haematological malignancies guided the selection of pesticide chemical groups and active ingredients. Risk estimates of lymph-haematological cancer and pesticide exposure, derived from the exposure assessment described, will be reported in separate publications. Nevertheless, the exposure assessment methods developed for this project will be applied in other pooling projects within the AGRICOH consortium.
Description of the cohort studies
The three prospective cohort studies included in this pooling project on lymph-haematological malignancies were: the US Agricultural Health Study (AHS),12 the French Agriculture and Cancer Study (AGRICAN)13 and the Cancer in the Norwegian Agricultural Population (CNAP) Study14 (table 1). These three cohort studies were selected within the AGRICOH consortium because of available data on the incidence of lymph-haematological malignancies, and the opportunity to assess exposure to individual pesticides within these cohorts.
AHS is a prospective cohort study which collected data on 52 394 licensed private pesticide applicators and 32 346 of their spouses in Iowa and North Carolina, USA, and on 4916 commercial applicators in Iowa. Details of the AHS design have been described elsewhere.12 In brief, pesticide applicators were enrolled between 1993 and 1997 by completing self-administered questionnaires15 that asked, among other things, about past and current personal use of 50 commonly used pesticide products based on a predefined list of chemicals (phase I). Five years following enrolment, a follow-up questionnaire was administered (phase II) asking about crops cultivated, animals raised and pesticides used (in an open-ended question) during the reference year, which was the most recent year a participant was farming.
AGRICAN is a prospective cohort study of 181 747 participants affiliated with the Mutualité Sociale Agricole (MSA), the French agricultural health insurance. Details of the AGRICAN cohort have been described previously.13 In brief, participants were enrolled between 2005 and 2007 by completing a mailed self-administered questionnaire. Participants were asked about the cultivation of 13 crops and different tasks, including pesticide treatment, performed on 11 of these crops.
The CNAP prospective cohort study of Norwegian farm holders and their families was compiled by linking data on farm characteristics and production from the compulsory agricultural censuses (between 1969 and 1989). Details of this cohort have been described elsewhere.14 In brief, a total of 147 134 farm holders and 98 759 spouses have been included in the cohort. Eligibility criteria for participation in the censuses have changed over the years. In each census, information on the cultivation of different crops at the farm during the preceding year was collected, using a prompted list. In addition, the 1969 census collected data on pesticide purchases and the 1979 census on the presence of pesticide spraying equipment at the farm.
Selection of participants
To harmonise the participants included in the pooled analysis, male and female private pesticide applicators and/or farmers (active or retired) were selected from each of the three cohorts. After exclusions, 51 167 participants from AHS, 127 282 from AGRICAN and 137 821 from CNAP remained (table 1). Participants in AHS were younger at enrolment than participants in AGRICAN. In AHS and CNAP, respectively, 3% and 16% of the cohort were female, compared to 44% in AGRICAN.
Chemical groups and active ingredients
For this pooling project, pesticide chemical groups and active ingredients were selected that were potentially used in more than one of the three countries, and prioritised if there was evidence from the scientific literature for an association with lymph-haematological malignancies. For this purpose, International Agency for Research on Cancer (IARC) monograph evaluations, US Environmental Protection Agency (EPA) assessments of carcinogenicity and the published epidemiological literature were reviewed. In total, 14 pesticide chemical groups and 33 individual active ingredients were selected (table 2). Active ingredients were classified into chemical groups on the basis of their main function (insecticide, fungicide, herbicide) and the classifications listed in the Wood's16 compendium of pesticides and the 199117 and 200918 British Crop Protection Council pesticide manuals. The complete list of the active ingredients contributing to each of the chemical groups per cohort varied, since the registration and use of specific active ingredients differed by country (see online supplementary table S1).
In AGRICAN and CNAP, no self-reported information on the use of individual pesticides was collected. Available data on crop cultivation and indicators of personal pesticide use were used as the basis for the exposure assessment. A number of crops were selected for developing country-specific crop exposure matrices (CEMs), to assign exposures to chemical groups and individual active ingredients. These crops were considered relevant because they were commonly produced in at least two of the three countries, or were major commodities in any of the countries, irrespective of the frequency of cultivation in the other cohorts (eg, vineyards in France).
For AGRICAN, grassland, corn, grains, potatoes, tobacco, orchard crops and vineyards were selected, and for CNAP, grassland, grains, potatoes, orchard crops and greenhouses. The prevalence of cultivation of these crops in each of the cohorts is shown in table 3. Although crop cultivation was not used to assign exposure in AHS, it is presented in table 3 to illustrate differences between the cohorts. In AGRICAN, 276 participants exclusively cultivated other crops than the ones selected for the CEM. Similarly, in CNAP, 33 834 participants were farm-holders who did not cultivate any of the selected crops. These participants were considered unexposed to any of the selected pesticide chemical groups and active ingredients.
Assessment of exposure to pesticides
For AHS, exposure to the specific pesticide active ingredients and chemical groups was based on participants' self-reported use of pesticides. In phase I, participants could indicate the period in which they first used the active ingredient as well as the duration of use of each pesticide product. If a participant reported the use of one of these active ingredients in phase II, the reference year of that participant was used as the last year of use, and the duration of use was recalculated accordingly. If a pesticide was only reported at phase II, then duration of use was calculated as the period between the enrolment year and the reference year. For the chemical groups, ever use, first year of use and duration of use were determined by the self-reported use of any active ingredient considered within the group (see online supplementary table S1).
A country-specific CEM was developed for seven main crops in France (table 3), covering the period 1950–2009, as the last day of follow-up for AGRICAN was 31 December 2009 in this pooling project. This CEM lists the first and last years the chemical groups and active ingredients were potentially used on each crop in France (see online supplementary table S2). For the development of this CEM, data were drawn from an existing French matrix, PESTIMAT,20 which contained information on the registration, sales and recommended use of a selection of chemical groups and active ingredients. For the crops and chemical groups and active ingredients not present in PESTIMAT at the time of the present study, additional work was performed to extract the first and last years of potential use from the data sources underlying PESTIMAT.20 For grassland, only herbicides were considered in the CEM.
For this pooling project, AGRICAN participants were considered potentially exposed to an active ingredient during a year if (1) they declared cultivating a crop, (2) reported personally performing pesticide treatment tasks on this crop and (3) the active ingredient was registered and recommended for use on the crop during that year according to the CEM. Participants who did not report any pesticide treatment tasks were considered unexposed to any pesticide for that crop. All active ingredients classified within a chemical group were considered for determining ever use, first year of use and duration of use for that group (see online supplementary table S1).
A country-specific CEM was developed for five main crops in Norway (table 3), covering the period 1950–2011, as the last day of follow-up for CNAP was 31 December 2011 in this pooling project. This CEM provided the first and last years each chemical group or active ingredient was sold and registered for use on each of the selected crops (see online supplementary table S2). In Norway, the available historical data on farmer pesticide use was restricted to pesticide registration and sales data, which were obtained from the Norwegian Food Safety Authority, and supplemented with expert input if the first year of registration or sales was unknown. All active ingredients classified within a chemical group were considered for setting the first and last years of potential use for a chemical group. For grassland, only herbicides were considered in the CEM (see online supplementary table S1).
Census data on crop cultivation on the farm was restricted to a maximum of five points in time. If a crop was cultivated at the farm in one census as well as the follow-up census, it was assumed that the crop was cultivated for the whole period between the censuses. Otherwise, the median year between the two censuses was assigned as the end or start year for cultivation of that crop (see online supplementary table S3). No crop cultivation was assigned to a participant for the years he/she was under 18 years of age. In the Norwegian census data, two pesticide use indicators were available at the farm level: (1) the amount of money spent on purchasing pesticides (1969 census) and (2) the presence of pesticide spraying equipment on the farm (1979 census). When either of these two indicators was positive, the participant was considered a likely pesticide applicator. Participants who only participated in the 1989 census (n=5852) or the horticultural censuses (n=421), and thereby lacked either of these pesticide use indicators, were also considered likely pesticide applicators.
For this pooling project, CNAP participants were considered to be potentially exposed to an active ingredient during a year if (1) they cultivated a crop, (2) were considered pesticide applicators and (3) the active ingredient was sold and registered on the crop during that year according to the CEM. All active ingredients classified within a chemical group were considered for determining ever use, first year of use and duration of use for that group (see online supplementary table S1).
Imputation of missing data
The imputation of missing data in AHS has been described previously.21 For AGRICAN, missing data on crop cultivation and pesticide treatment tasks on crops were imputed using multiple imputation by chained equations (MICE).22 All imputations were performed five times and the imputed data were combined using Rubin's Rules.23 No data were imputed in CNAP as the agricultural censuses were mandatory and thereby complete.
Comparison of exposure assessment methods
No external data were available to validate the exposure assessment methods used in this pooling project. However, we did attempt to compare exposure estimates generated using the different methods applied. In the AHS phase II questionnaire, participants provided information on both pesticides applied and crops cultivated during the reference year. These data allowed for a comparison between pesticide exposure assigned using a CEM approach based on crop cultivation and self-reported pesticide use in the reference year.
For the purpose of this comparison, a CEM was developed on the basis of pesticide registration data from the USA, for the period corresponding to the phase II reference years (1992–2002). This CEM will not be used in the epidemiological analyses, as for AHS exposure is based on the participants' self-reported use of pesticides. Eight crops were selected on the basis of cultivation in either of the other two cohorts (grassland, corn, grains, potatoes, tobacco, orchard crops and vineyards). Soya beans were added as they are a major crop in Iowa and North Carolina, USA. Eleven pesticide active ingredients with relatively low, medium and high self-reported use among AHS participants were selected for this CEM (see online supplementary table S4).
In AGRICAN, participants were considered exposed to the active ingredients in the CEM for a specific crop, when cultivating the crop and performing crop-specific pesticide treatment tasks. To approximate this approach, AHS phase II participants were classified as exposed to active ingredients registered for use on the crop cultivated during the reference year, if they reported having applied any pesticide to the relevant crop (approach 1). For CNAP, exposure to pesticides was assigned on the basis of purchase of pesticides and/or the presence of pesticide spraying equipment on the farm. AHS phase II participants who reported any pesticide application were considered exposed to all active ingredients registered on the crop during the reference year according to the CEM (approach 2).
For each active ingredient, agreement between the self-reported use and the two CEM exposure assignments was estimated by the percentage raw agreement and Cohen's κ score.24
Prevalence of pesticide exposure
In AHS, 99% of the participants reported ever using any pesticide. In AGRICAN, 68% of the participants reported performing pesticide treatment tasks on any of the selected crops, and in CNAP, 63% of the participants were considered a likely pesticide applicator based on the available farm-level pesticide use indicators. The prevalence of exposure could differ substantially between the cohorts, depending on the chemical group or active ingredient (table 4). For example, organophosphate insecticides were used by 93% of AHS participants, but exposure was assigned to only 64% of AGRICAN participants and 42% of CNAP participants. In contrast, exposure to dithiocarbamate and phthalimide fungicides was highly prevalent in AGRICAN (64% and 60%, respectively), while fewer AHS participants (12%) reported using either of these chemical groups. However, for some chemical groups, the overall prevalence of exposure was relatively similar. Exposure prevalence among women participants was lower than among male participants in all cohorts (see online supplementary table S5). In AHS, for some pesticides data were only collected in phase II (eg, MCPA) and the prevalence therefore referred to the phase II reference year only, and is lower than the (lifetime) prevalence of exposure in AGRICAN and CNAP. The distribution of exposure duration in the cohorts can be found in the online supplementary table S6.
Overall, low to moderate Pearson correlations were found in AHS between the self-reported duration of use of pesticide chemical groups (median r=0.08) or active ingredients (median r=0.07) (see online supplementary tables S7.1 and S8.1). For AGRICAN and CNAP, where a CEM was used to determine potential exposures, correlations between the assigned duration of exposure were high for chemical groups (median AGRICAN r=0.80, CNAP r=0.77) and active ingredients (median AGRICAN r=0.71, CNAP r=0.55) (see online supplementary tables S7.2, S7.3, S8.2 and S8.3). Correlations were especially high between active ingredients within some the chemical groups, for example, within the phenoxy herbicides (median AGRICAN r=0.91, CNAP r=0.99, AHS r=0.24) and pyrethroid insecticides (median AGRICAN r=0.86, CNAP r=0.95, AHS r=0.51).
Comparison of exposure assessment methods
Generally, <10% of AHS participants reported the use of the active ingredients selected for the CEM comparison during the phase II reference year, except for the herbicides glyphosate and metolachlor (table 5). Using either of the two CEM approaches, the majority of the AHS participants were classified as potentially exposed to the active ingredients during the reference year. Agreement between the self-reported use and the assigned exposure was poor for both approach 1, which resembled the exposure assignment in AGRICAN (κ −0.00 to 0.33), and approach 2, which resembled the exposure assignment in CNAP (κ −0.01 to 0.14). The CEM approaches led to a higher exposure prevalence compared to the self-reports, with minimal differences between approaches 1 and 2.
We developed cohort-specific methods to assess exposure to a harmonised set of pesticide chemical groups and individual active ingredients through declared or presumed application in three prospective cohort studies with marked differences in design and detail regarding pesticide exposure. Self-reported pesticide use was used to derive exposure estimates for the AHS participants. In AGRICAN and CNAP, time-specific information on crop cultivation was present and cohort-specific CEMs were developed to estimate pesticide exposures at the active ingredient or chemical group level.
The observed differences in prevalence of exposure to the chemical groups and active ingredients between the cohorts could be due to a wide range of factors, including differences in the (number of) crops being cultivated in each country, the pesticides registered for use over time, the age and gender distribution of the participants in each cohort, or the different exposure assessment methods used for each cohort. The estimated pesticide exposure prevalence in AGRICAN and CNAP was generally higher than in AHS, among participants considered to be pesticide applicators (68% in AGRICAN and 63% in CNAP). In AGRICAN, fruit growing and vineyards were far more common than in the other cohorts, which can in part explain the higher exposure prevalences as many of the selected active ingredients and chemical groups were registered on these crops. A relatively high prevalence of exposure to organochlorine insecticides was found in AGRICAN, which could be due to the type of crops cultivated and the older age of the participants in relation to the years these pesticides were registered. In AHS, a high prevalence of exposure to chloroacetanilide, dinitroaniline and triazine herbicides was found, which may be related to the use of these herbicides on corn and soya beans, which were predominant crops in AHS. The higher prevalence of exposure to organophosphate insecticides in AHS might be the result of differences in pesticide registration and the use of these insecticides on livestock, which was part of the self-reported insecticide use in AHS, but was not considered in the CEMs for AGRICAN and CNAP.
In AGRICAN and CNAP, correlations between the exposure estimates were high. Roughly 20% of the correlation coefficients between the active ingredients exceeded 0.80, which will make it difficult to attribute health effects, if any, to exposure to individual agents. Mixed exposures and co-occurring exposures to pesticide active ingredients are given in many agricultural settings.25 ,26 However, in AHS, correlations between exposures to the pesticide chemical groups and active ingredients were low overall. This might be explained by the higher specificity of self-reported information compared to CEMs, applicators using only a limited set of pesticides on the crops cultivated, differences in the timing of data collection (which is restricted to phase II for a number of pesticides) or applications on livestock included in the self-reported insecticide use.
CEMs have been successfully applied in epidemiological studies to estimate (potential) exposure to pesticides among agricultural populations.27–29 These studies generally incorporated detailed data to estimate personal pesticide use, such as the intensity, probability and frequency of use or specific tasks performed. In AGRICAN, data on performing crop-specific pesticide treatment tasks were available, but in CNAP, farm-level variables were used as an indicator for personal pesticide application by the farm holder. The CEMs developed in this project assume a 100% probability of use when an active ingredient was registered and recommended for use on a crop (AGRICAN) or registered for a crop and sold (CNAP). Several studies have indicated that only a fraction of registered pesticides are regularly used by farmers,30 ,31 and their use will depend, among others, on weather conditions, prevalence of pests, regional preferences and costs. Recent Norwegian survey data indicated that pesticides were applied to potatoes on only 66% of farms, depending on the potato acreage, and only a fraction of farmers applied any herbicides to grassland.32 Therefore, including information on the probability of use would be an important improvement of the CEMs developed in this project, to differentiate between active ingredients registered for similar uses during the same time period.
Not all crops cultivated by the participants were included in the CEMs, only those considered most prevalent and relevant in AGRICAN and CNAP. Therefore, some exposures to the selected pesticides, associated with excluded crops or livestock, will not be considered. Data from AHS phase II indicate that the underestimation of exposure from not accounting for livestock or poultry applications seems to be minor. Less than 2% of these participants have only livestock or poultry, and for those having both crops and livestock (50%), exposure profiles of a number of selected insecticides (table 5) appear to be similar. Also, exposure through re-entry tasks is not accounted for in any of the three cohorts. In AGRICAN, where the percentage of female participants is especially high, as well as the cultivation of crops likely involving re-entry tasks, exposure to some pesticides might be underestimated. In this study, we have focused on pesticide active ingredients only. It should be acknowledged that pesticide products can also contain multiple additives and solvents, which we do not take into account. Therefore, we cannot exclude the possibility that these chemicals might contribute to potential effects observed in subsequent epidemiological analyses.
When two CEM approaches, approximating the exposure assessment as conducted for AGRICAN and CNAP, were used to assign pesticide exposure to AHS phase II participants, agreement between self-reported exposures and CEM exposures estimates was poor. The CEM approaches developed for this project appeared to overestimate exposure during the reference year compared to self-reports. The comparison also suggested that assigning exposure exclusively to participants who reported pesticide applications on a specific crop (approach 1) only resulted in small improvements, compared to assigning exposure on an overall pesticide use indicator (approach 2). Self-reported exposure information is not a true gold standard. A study among male applicators participating in AHS indicated their ability to produce reliable and reproducible reports of their pesticide use, but the validity of these reports could not be assessed.33 Therefore, it remains unclear to what extent the AHS self-reported data may underestimate or overestimate true pesticide use. All AHS participants in this pooling project were licensed private pesticide applicators and the vast majority reported applying pesticides. Most of them cultivated corn and/or soya beans, on which all of the active ingredients selected for this comparison were registered during the consecutive reference years. Therefore, the AHS phase II data might not be comparable to the situation in the AGRICAN and CNAP cohorts, where not all participants are considered pesticide applicators based on their farming activities (AGRICAN 68%, CNAP 63% vs AHS 99%), lifetime crop cultivation is more diverse (table 3), and pesticide registration of active ingredients appears to be more restricted to individual crops.
We acknowledge that the CEM approach may generate false positive exposure assignments. The lack of specificity of the CEMs is, however, less of an issue for assigning exposure to pesticide chemical groups and active ingredients that are relatively frequently applied. It will lead to more substantial exposure misclassification if the actual prevalence of use is low.34 We assume the misclassification to be non-differential, given that the assignment of exposure is based on occupation and independent of disease status. Non-differential exposure misclassification usually leads to a bias of the estimate towards the null, especially when the strength of the association is modest, as is the case for most pesticide exposures and health effects. This will most likely limit our ability to detect associations, if any, in the two cohorts where CEMs were applied. Results from the epidemiological analyses should be interpreted cautiously.
A strength of this work is that the exposure assessment efforts accommodate the use of information from three different countries and studies, with distinct designs and detail regarding pesticide exposure, in a large pooling project. The large sample size will enable the analysis of associations between rare agricultural exposures and rare health effects. A wide range of chemical groups and active ingredients, selected a priori, have been included, and will allow us to investigate previously studied associations with greater power and other associations for the first time. This is a significant improvement compared to analysing exposure to pesticides as broad categories, or using job titles such as farmer, applicator or farm worker as a proxy for pesticide exposure. Furthermore, the exposure assessment extended back to 1950, thereby covering a substantial fraction of the occupational lifetime of most of the participants.
For future studies applying CEMs, the quality of the exposure estimates will largely depend on the available internally collected exposure data (eg, personal pesticide application on a specific crop) and external information on pesticide registration, sales figures and crop specific usage patterns, which need to be collected separately for each country for the relevant time period. As shown in this study, pesticide registration and sales figures alone are of limited use when the aim is to create CEMs with sufficient specificity to differentiate between exposures to individual active ingredients or chemical groups. More detailed exposure information such as probability and frequency of use is warranted to reduce overestimation of exposure and (non-differential) misclassification. Owing to time constraints and limitations in available resources, these additional factors could not be taken into account in the CEMs in our pooling project at present. Future undertaking of nested case–control studies in AGRICOH would bring the opportunity to collect retrospective exposure data with the necessary depth and in a similar manner in these cohorts. Agricultural studies should work towards more harmonised exposure assessment, using common questionnaires and collecting detailed information on personal application practices, and, given the suspected associations between pesticide exposures and adverse health outcomes, national agricultural censuses would be encouraged to collect more information on the use of (specific) pesticides.
We developed methods to assess occupational exposure to specific pesticide chemical groups and active ingredients for the first pooled study within the AGRICOH consortium. Exposure estimates were generated for 14 chemical groups and 33 active ingredients. Our study illustrates the wide range of chemical agents farmers are exposed to or potentially exposed to in the three agricultural cohorts from the USA, France and Norway. The various exposure assessment methods provided exposure estimates that may overestimate or underestimate actual exposure prevalence. Additional work is needed to better quantify how far these estimates deviate from reality. Limitations of the exposure assessment should be taken into account in planning and interpreting results of the subsequent epidemiological analyses.
The authors would like to acknowledge the Norwegian Food Safety Authority and Norwegian expert for providing information on the historical registration and sales of pesticides in Norway.
Funding This study was funded by a grant from the Office National de l'Eau et des Milieux Aquatique (ONEMA), Plan d'action national ECOPHYTO 2018, Axe 3, Volet 4, France. In addition, this work was funded, in part, by the Intramural Research Program of the National Cancer Institute, National Institutes of Health (Z01-CP010119) and the Ammodo van Gogh travel grant VGP.14/20. We used the following AHS data releases for this analysis: P1REL201209.0 and P2REL201209.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.