Objectives: Accurate assessment of exposure is a key factor in occupational epidemiology but can be problematic, particularly where exposures of interest may be many decades removed from relevant health outcomes. Studies have traditionally relied on crude surrogates of exposure based on job title only, for instance farm-related job title as a surrogate for pesticide exposure.
Methods: This analysis was based on data collected in Western Australia in 2000–2001. Using a multivariate regression model, we compared expert-assessed likelihood of pesticide exposure based on detailed, individual-specific questionnaire and job specific module interview information with reported farm-related job titles as a surrogate for pesticide exposure.
Results: Most (68.8%) jobs with likely pesticide exposure were farm jobs, but 78.3% of farm jobs were assessed as having no likelihood of pesticide exposure. Likely pesticide exposure was more frequent among jobs on crop farms than on livestock farms. Likely pesticide exposure was also more frequent among jobs commenced in more recent decades and jobs of longer duration. Our results suggest that very little misclassification would have resulted from the inverse assumption that all non-farming jobs are not pesticide exposed since only a very small fraction of non-agricultural jobs were likely to have had pesticide exposure.
Conclusions: Classification of all farm jobs as pesticide exposed is likely to substantially over-estimate the number of individuals exposed. Our results also suggest that researchers should pay special attention to farm type, length of service and historical period of employment when assessing the likelihood of pesticide exposure in farming jobs.
Statistics from Altmetric.com
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.
Accurate exposure assessment is a critical factor in epidemiological research in occupational settings.1 2 However, the quality and rigour of occupational exposure assessment in published studies varies widely and lack of adequate exposure definitions and imprecise exposure assessment methods remain a significant limitation in the majority of studies in occupational epidemiology.2–4 Inadequate exposure assessment leading to non-differential exposure misclassification tends to result in underestimation of effects,5 particularly where study power is relatively low.6 This may result in important associations being under-rated or remaining undetected.
Occupational exposure assessment has commonly been based on job titles as a surrogate for occupational exposure, particularly in community based studies.7 Assessment of exposure is particularly challenging in retrospective studies because the exposures of interest may have been experienced decades earlier.8 This is particularly important for studies of diseases with long latent periods such as cancer. Sophisticated retrospective exposure assessment methods have been developed, such as job exposure matrices (JEMs), exposure algorithms and job specific module (JSM) questionnaires.4 7 9 10 However, these are still seldom used in studies of farmers. Many retrospective studies of cancer and occupational exposure to agricultural pesticides have relied on the assumption that farmers and farm workers are, by definition, pesticide exposed.11
Pesticides are a large and heterogeneous group of chemicals exerting a variety of biological effects and are used for many different agricultural applications.12 Different crops or animals farmed in different geographical regions with different pest problems are likely to result in quite different exposure patterns. Therefore, the assumption that farm workers are homogenous in terms of pesticide exposure is likely to be flawed. However, there is limited information about what types of farm workers are more likely to be exposed and factors that predict the exposure remain unclear.13
This paper describes an examination of the exposure assessment conducted as part of a large case–control study of occupational risk factors for prostate cancer and benign prostatic hyperplasia.14–16 Subjects completed computer-assisted telephone questionnaires which included questions about whether they handled pesticides and what types they used. These questionnaires were reviewed by an expert who determined the likelihood of exposure to different pesticides. Hypothesising that farm-related jobs may be highly heterogenous in terms of pesticide exposure, we compared the JSM-informed expert assessed likelihood of pesticide exposure (gold standard) with job title classification in order to see how well farm-related job titles reflect pesticide exposure in a community based study.
The analyses reported in this paper are based on a case–control study which investigated risk factors for prostate cancer and benign prostatic hyperplasia. The methods of recruitment and data collection in this case–control study have been reported previously.14–16 Briefly, cases were recruited from hospital morbidity records through the Western Australian Data Linkage System. Cases were men aged 40–75 years. Controls were randomly selected from the Western Australian electoral roll and frequency matched to the expected case distribution within 5-year age groups. Recruitment rates were over 60% for prostate cancer cases and approximately 40% for controls and benign prostatic hyperplasia cases. Recruitment and data collection occurred during 2001 and 2002.
A self-administered questionnaire was used to collect demographic information, medical history and occupational history. Occupational histories were collected from 1543 subjects. Occupational histories listed all jobs held for 1 year or more over the individual’s working life, including start and end dates, job title/description, industry and employer. Job title, industry and employer information was used to code jobs using the Australian Standard Classification of Occupations (ASCO).17
For 14 specific job categories of interest (carpenter, driver, electrician, plumber, forestry worker, farmer, labourer, machinist, mechanic, miner, fisherman, painter, railway worker and welder), JSMs18 were used to collect further exposure information. To keep interviews to a reasonable length, where a subject’s occupational history had more than fives jobs of interest, JSMs were allocated in a priority order aimed at obtaining the maximum coverage with no more than five JSMs per individual. In particular, where a subject reported several similar jobs, JSMs were not repeated for all jobs if the total number of jobs of interest exceeded five. The JSM specific to farming jobs included detailed questions about pesticide use, frequency and method of use and specific pesticide products used.
An expert occupational hygienist reviewed the occupational histories and the answers to the JSMs and determined the likelihood of exposure to a variety of contaminants as “probable”, “possible” or “no exposure”.19 The expert assessment of jobs without JSM data was informed by the JSM data for similar jobs in the same subject and therefore the multivariate analyses were clustered to account for similarities in jobs held by the same subject.
Each job was assessed for exposure to five categories of pesticides: organophosphates, organochlorides, phenoxyherbicides, other herbicides and other pesticides. For each of these five categories, the expert hygienist rated jobs as “probably exposed”, “possibly exposed” or “not exposed”. For the purpose of the analyses presented in this paper, these five categories were collapsed into a single variable representing exposure to any class of pesticides. Because of the comparatively small number of reported jobs assessed as “possibly exposed” to pesticides (73 jobs), for the analyses presented in this paper the “possibly exposed” jobs have been combined with the “probably exposed” jobs. In order to assess the effect of this amalgamation, a sensitivity analysis was undertaken by repeating the multivariate analysis excluding the “possibly exposed” jobs. Analyses were also repeated using only control subjects’ jobs to test for any case–bias effect.
Farming jobs were defined based on self-reported job titles coded as ASCO17 codes: 1310 (farmers and farm managers), 1311 (mixed farmers (crop and livestock)), 1312 (livestock farmers), 1313 (crop farmers), 4611 (farm overseers), 9921 (farmhands) and 9929 (other agricultural and horticultural labourers).
For the purpose of these analyses, all available information for each farm job was re-examined to determine the employment status (farmer/farm manager or farm labourer/other farm employee) and farm type (predominantly crops, predominantly livestock, mixed crops and livestock) for each job.
Analyses were performed using Stata 9.0. (Stata, College Station, Texas, USA). Multivariate logistic regression models were used to obtain odds ratios and confidence intervals were calculated with robust standard errors using quasi-likelihood estimation.20 Analyses were adjusted for correlation between jobs held by the same individual, using the Stata 9.0 “cluster” function (robust sandwich variance estimator).
Overall, 1543 subjects provided occupational histories (table 1). Of these subjects, 91% were over 55 years of age at the time of recruitment in 2001–2002. Subjects reported a total of 13 048 jobs. Overall, 371 subjects (24.0%) reported at least one farm job and these subjects reported a total of 801 farm jobs. Of these jobs, 511 (63.8%) had JSM data available and 276 (34.5%) had no JSM data; written questionnaire data were available for all jobs. Fourteen (1.0%) jobs had insufficient information for exposure assessment. These 14 farm jobs with incomplete job histories which were missing information about farm type, employment status and job start/end dates, were excluded from the current analysis.
Of the 371 subjects who reported any farming jobs, 86.5% received at least one farmer JSM. Of the 163 subjects who reported only a single farming job, 76.7% received a farmer JSM. Short duration jobs were the least likely to have JSM data available. Among farm jobs of the shortest category of duration (<2 years), only 38.8% had JSM data available and the longer the job the more likely JSM data were available, with 96.0% of jobs of more than 20 years’ duration having JSM data. Also, the expert assessor was much more likely to judge a job as “probably exposed” if JSM data were available. Of jobs judged “probably exposed”, 80.4% (n = 111) had JSM information, while of jobs judged “possibly exposed”, only 57.1% (n = 20) had JSM information.
Amongst all subjects in the study, the average number of jobs per career was 8.5 and average job duration was 5.6 years. However, amongst those who ever had at least one farm job, the average number of jobs per career was 4.9 and the average duration was 7.8 years per job. Overall, 27.0% of farm jobs (n = 210) were of less than 2 years’ duration and these jobs were predominantly held by younger workers: 72.4% were under 30 and 43.8% were under 20. Also, 49.0% of these short duration farm jobs were the subject’s first ever job and 74.3% of them were the worker’s first or second ever job.
Among farming jobs with JSM data, 24.9% of farmer/manager jobs and 24.2% of farm worker jobs were on animal-only farms. In addition, 54.5% of farmer/manager and 51.6% of farm worker jobs were on mixed crop and livestock farms.
The majority of jobs with possible or probable pesticide exposure, (68.8%, n = 174) as assessed by expert review, were farming jobs (table 2). However, most farm jobs (78.3%) were not pesticide exposed. The 79 non-farming pesticide-exposed jobs were a diverse group including jobs associated with ornamental gardens and sports fields (such as landscape gardeners, council workers, nurserymen and green-keepers) as well as urban pest controllers.
Among farm jobs, jobs on crop farms and mixed farms were more likely to be pesticide exposed than jobs on farms where animals only were raised (table 3). Farmers were more likely to be exposed than farm hands and the probability of being exposed increased with duration of job, with age at starting the job, and with more recent starting years.
After adjusting for all other variables in table 3, crop farm jobs remained the most likely to be associated with pesticide exposure, compared with animal-only farm jobs. Jobs on mixed crop and livestock farms were also significantly more likely to be associated with pesticide exposure compared with animal farm jobs. Employment status and age at job commencement were not predictive of pesticide exposure in the multivariate model, however a dose–response relationship appeared evident for increasing job duration. This was stronger when the “possibly exposed” jobs were excluded.
Likelihood of pesticide exposure appeared to increase with more recently commencing jobs, with marked increases particularly after 1980. However, it should be noted that in the periods since 1980 the confidence intervals were wide because there were relatively few jobs in these categories. Jobs commencing in all periods since 1950 were more likely to be associated with pesticide exposure than jobs commenced prior to 1950.
When the jobs assessed as “possibly exposed” (n = 38) were excluded from the multivariate analysis, the strength of all associations increased, although the confidence intervals were wider because of smaller numbers (table 3).
This was a case–control study of prostate cancer, a condition which affects older men near the end of their working careers or in retirement. Therefore, the job histories collected provide a longitudinal view of complete or near complete working careers of a sample of Western Australian men in the second half of the 20th century. Amongst this group, farming jobs featured frequently with nearly one quarter of subjects having ever held a farm job. According to the most recent census data, 3.4% of males in Western Australian were employed in farm jobs on census night,21 but our results suggest that the proportion of the population who have ever worked on a farm is somewhat larger.
Those who had worked in farm jobs tended to have fewer jobs in total and their jobs tended to be of longer mean duration, implying a group of workers with job stability. Although information on jobs of less than 1-year duration was not collected, our data suggest that this agricultural workforce has a large proportion of settled workers. This contrasts with the mainly North American literature on farm worker health, where farm workers are frequently associated with itinerant and often immigrant workers.22 In our study, 27% of farm jobs were held for less than 2 years, however the short-term jobs in this study were predominantly held by young people and they were often the individual’s first ever job, suggesting that the short-term farm workers in this study are demographically different to transitory farm workers observed in North American research.
Although the majority of jobs with likely pesticide exposure were farm jobs, pesticide exposure was far from universal amongst farm jobs, with the majority of farm jobs assessed as having no likelihood of pesticide exposure. Although farming communities in different locations can be expected to differ in the precise proportions of workers who are pesticide exposed, our results challenge the assumption that farm work alone is an appropriate surrogate for pesticide exposure in epidemiological studies. In response to the recommendations of reviewers such as Zahm and Ward,23 there has been a move towards more detailed exposure assessment of farm jobs in recent years4 and our findings suggest that the additional expense of more detailed exposure assessment for agricultural jobs is indeed justified in case–control and cohort studies.2 11
In order to characterise factors predictive of pesticide exposure among farm jobs, we also examined the associations between pesticide exposure and a number of other factors. We found that jobs on farms where crops were grown were more likely to be pesticide exposed compared to farms with animals. This effect was strongest for crop-only farms but was also significant for mixed crop and livestock farms, suggesting that the presence of crops on the farm may be an important factor. This is consistent with research from the Netherlands reporting considerable variation in pesticide exposure between different agricultural sectors.24 In our study, one third of farm jobs were on animal-only farms.
We hypothesised that farmers and managers would be more likely to be exposed than other farm employees. However, any differences between farmers/managers and other farm employees were not evident in the multivariate analyses, leading to the conclusion that employment status in itself may not be an important factor. However, there was a strong dose–response relationship between likely pesticide exposure and longer job duration, and this pattern was stronger still when the “probably exposed” jobs were excluded. It may be that quantitative job duration provided a more accurate summary of the experience, proprietorship and seniority factors relevant to pesticide exposure than our ranking of employment status based on job title. When seeking indicators of exposure among workers on farms, job duration is likely to be a useful metric and possibly more reliable for this purpose than employment rank inferred from job titles in Australia and other similar contexts.
Much of the published literature on pesticides and farm worker health is from North American research where the situation is characterised by the fact that a significant proportion of farm employees are distinct from their employers and managers in terms of race and socioeconomic, educational and citizenship status.22 25–28 For example, in California it is estimated that half of the hired farm workforce are undocumented immigrants.25 However, this study demonstrates that this mainly North American literature on pesticides and farm workers’ health may not be applicable to other contexts.
We found pesticide exposure to be more likely in jobs which commenced in more recent decades compared with jobs before 1950. Jobs commenced in the 1950s and 1960s were more than twice as likely to be pesticide exposed than pre-1950s jobs and this is consistent with the well known dramatic rise in agricultural pesticide use in the period after the Second World War.29 Unexpectedly, however, likelihood of pesticide exposure rose most steeply for jobs since 1980. This is an interesting finding and may reflect a greater proportion of the workforce handling pesticides in recent decades, possibly associated with the increased popularity of newer pesticide products of perceived lower toxicity. However, it may also reflect clearer memory of more recent work activities than historical ones.
A relatively small number of non-farming jobs were assessed as pesticide exposed, including jobs such as urban pest controllers and jobs associated with maintenance of ornamental gardens and sports fields. The classification of all non-farming jobs as non-pesticide exposed would result in the misclassification of fewer than 1% of all jobs in this sample.
When the jobs assessed as “possibly exposed” were excluded from the multivariate analysis, the strength of all associations increased (table 2). This suggests that exposure is indeed less certain in those jobs judged “possibly exposed” compared with the “probably exposed” jobs and this should be taken into account when interpreting results of epidemiological studies where degrees of expert certainty have been used in the exposure assessment.
When analyses were repeated using only the control subjects (n = 599 jobs), the same general pattern of results emerged but with wider confidence intervals reflecting the reduction in statistical power (results not shown), indicating that the results of the analyses reported here are not likely to be biased by case selection.
Misclassification of pesticide exposure can result in errors in estimated risk of disease in epidemiological studies. In this study, using farm-related job title as an exposure surrogate resulted in sensitivity and specificity of 0.69 and 0.95, respectively. Modelling the effect of this misclassification on odds ratio estimations using methods published by White et al30 (results not shown) demonstrated underestimation of 11%–43% in hypothetical case–control analyses where 8%–20% of cases and 5%–15% of controls were truly exposed. When only jobs on farms with crops (crop-only+mixed farms) were considered, the sensitivity rose to 0.75 but specificity was only 0.38, which resulted in a more exaggerated movement of the odds ratio towards 1.0 (26%–67% difference).
For the purpose of this paper we have considered all pesticides as a single exposure category. However, pesticides are a group of diverse chemicals with varying toxicities and human health effects. In research investigating associations with specific health outcomes, an exposure definition such as “exposed to any pesticides” may be inadequately specific. The present analysis demonstrates that even when all pesticides are considered together, the proportion of farm workers with exposure is still low and exposure to specific products would be lower still. This indicates the importance of detailed, pesticide-specific exposure assessment of farm workers in studies investigating particular health outcomes.
One of the strengths of this research is that the subject group was derived from a community-based case–control study. It is therefore reasonable to assume that the study sample is representative of the male workforce in Western Australia. Also, because the case–control study concerned a disease predominantly of late adulthood, we had complete or nearly complete career histories for all subjects. The JSM-assisted expert exposure assessment also provided detailed and specific exposure information for the majority of jobs, providing a high degree of confidence in the exposure assessment.
These results suggest that farm job titles alone are an inadequate surrogate for likely pesticide exposure and that future epidemiological research should emphasise more sophisticated exposure assessment methods. Classification of all farm-related jobs as pesticide exposed is likely to result in substantial over-estimation of the number of exposed workers. In assessing exposure to agricultural pesticides in Australia and similar contexts, researchers should be aware that exposure to pesticides may be more likely in jobs on farms where crops are grown, workers with long lengths of service and workers who commenced in jobs since 1980. Conversely, workers who raise animals and those with short lengths of service and jobs in earlier decades may have a lower likelihood of pesticide exposure.
Researchers should also be aware that although pesticide exposure exists in other sectors, the proportion of pesticide-exposed jobs outside farming is small and therefore classifying non-agricultural jobs as not exposed to pesticides is unlikely to lead to significant misclassification in population-based studies.
Farming job title was found to be a poor indicator of likelihood of pesticide exposure as a substantial proportion of farm jobs may involve no such exposure.
In this population-based study, relatively few pesticide-exposed jobs were found to be outside the farming sector.
Likelihood of pesticide exposure amongst farm workers is associated with employment and workplace factors, particularly farm type, length of service and historical period of employment.
The assumption that farm-related job title can be taken as an indication of pesticide exposure is likely to result in over-estimation of the probability of exposure; detailed exposure assessment is required for accurate pesticide exposure assessment in agricultural jobs.
Classification of non-agricultural workers as unlikely to have pesticide exposure may result in relatively little misclassification in population-based studies.
When assessing pesticide exposure in farm jobs for epidemiological purposes, employment and workplace factors may provide useful clues to likelihood of exposure.
We thank Justine Leavy, Dr Jafar Tabrizi and Dr Gina Ambrosini for their work on the original case–control study and Pam Simpson and Professor Andrew Forbes for statistical advice.
Funding: The original case–control study on which the analyses in this paper are based was funded by Healthway and the BUPA Foundation. Ewan MacFarlane is supported by an NHMRC Postgraduate Research Scholarship. Lin Fritschi is supported by an NHMRC fellowship.
Competing interests: None.