Clues to cancer etiology from studies of farmers

,

reporting of po sitive findings. Specifically excluded were studies of individual cancers in which negative findings are likely to go unreported. In a meta-analysis we summed the observed and expected numbers for specific cancers, total mortality, and ischemic heart disease among farmers to create meta-relative risk (MRR) estimates to minimize the influence of unusual chance find ings from individual studies on the overall interpretat ion. This procedure weights the contribution of each study by its size. Statistical significance was evaluated using 950/0 confidence intervals, which were calculated for the MRR estimates using the procedures for a Poisson variable by Breslow & Day (6). We ha ve also pres ented the range of relative risks observed among the studies, the number of studies with relative risks exceeding unity, the number of statistically significant excesses, the number of studies with relative risks less than unity, and the number with significant deficits to evaluate the distribution, magnitude, and consistency of the individual relative risks . This approach weights the contribution of each investigation equally. The statistical significance of the distribution of relative risks greater than and less than 1.00 was evaluated by a sign test (7). See figure 1 for a graphic presentation of the relative risks and 95% confidence intervals for selected cancers.
In analyses such as this it is desirable to be able to adjust for confounding and bias that ma y occur in the data sets used . Since we did not ha ve access to the raw data, such an adjustment was not possible. It seems unlikely, however, that a common set of problems would exist in the various data sets, which were assembled in different ways and were from many different countries.  ., jill! JI+! Figure 1. Relat ive risks and 95% confidence interva ls for selected cancers by study. The vertical bars represent the 95% confidence intervals obtained from the papers , if avallable ; otherwise they were calculated accord ing to Breslow & Day (6).

Results
Table 1 displays summary data from studie s specifically on farmers and from surveys that included evaluations of many occupations in addition to farmers. In these studies, mortality among farmers was usually compared with the risk in the general population. Statistically significant deficits in the MRR estimates occurred for all causes combined, ischemic heart disease, and all cancer combined. For these disease categories, the relati ve risks were less than 1.0 in nearl y every study, most being statistically significant. Significant deficits among farmers occurred for cancers of the lung, esophagus, bladder, colon, liver, and kidney. Few individual studies had relative risks greater than 1.0 for these tumors. MRR estimates significantly greater than 1.0 were obtained for Hodgkin's disease, 210 melanoma, multiple myeloma, leukemia, and cancers of the lip, stomach and pros tate. The excesses, however, were small, and some individual stud ies showed deficits. Although the MRR estimates for non-Hodgkin' s lymphoma and cancer s of the connective tissue and brain were not significantly elevated, more individual studies of farmers showed excesses of these cancers than deficits. Two studies (27,28) included in table 1 evaluated cancer risks among female farmers, but the small numbers observed complicated comparison with the pattern among male farmers . Although the female farmers presumedly operated or worked on farms, no information was available to compare the type of farming performed by women with the type performed by men. Among the women only, for selected sites with at least 10 deaths, the MRR estimates were 1. 23

Discussion
Although farmers comprise one of the largest occupational group s with exposures to potentially hazardous substan ces, agriculture has not been a high prior ity on the occupational health research agenda (I). Issues of exposure and health in agriculture have typically fallen through the cracks of governmental systems in the United States (29) and other countries, since work per- formed by family members is typically not affected by state or federal labor regulations (30). A more intensive focus on agriculture is needed to ensure that farmers have the same protection that is generally provided to other occupational groups. Epidemiologic investigations are important to preventive efforts, as they can provide new leads to the etiology of tumors excessive among farmers. The mortality experience of farmers is favorable in terms of all causes, all cancers combined, and ischemic heart disease. The low rates for cancers of the lung, esophagus, and bladder, as well as heart disease, may be explained by the low prevalence of smoking observed globally among farmers (15,24,(31)(32)(33)(34). In addition, the physical demands on farmers may account for their low body fat and high levels of physical fitness (35), which in turn may contribute to lower risks for heart disease and colon cancer (35)(36)(37)(38)(39)(40). Dietary factors (such as high intake of fresh fruits and vegetables), residence in areas with little air pollution, and selective migration may influence the deficits of cancer observed, but these factors have not been evaluated among farmers.
In contrast to the deficits for most major disease categories, farmers had significantly elevated risks for leukemia, multiple myeloma, Hodgkin's disease, melanoma, and cancers of the lip, prostate and stomach. Mortality from non-Hodgkin's lymphoma and cancers of connective tissue and brain was increased in most studies, although the MRR estimates were not significant. These tumors do not fall into an obvious grouping, other than the fact that they are not strongly associated with smoking. They vary in frequency, histology, and prognosis. The excesses among farmers for a few specific cancers, against a background of low risks for most cancers and nonneoplastic disease, suggest a role for work-related exposures. These patterns may have broader public health implications, since several of the high-rate tumors among farmers also appear to be increasing in the general population of many developed countries (5). Of particular interest are the rising rates for multiple myeloma, non-Hodgkin's lymphoma, melanoma, and cancer of the brain. Prostate cancer has increased especially in England and Wales (41) and the United States (42). Some of these trends may partially result from improvements in diagnosis and reporting, which affect incidence rates to a greater extent than mortality statistics (42). However, it is noteworthy that, in the United States, the rising mortality rates for non-Hodgkin's lymphoma, multiple myeloma, and leukemia have been the most pronounced in rural agricultural areas of the central region of the country (43). Although stomach cancer has been declining in all developed countries, the rates for tumors arising in the gastric cardia have increased remarkably for unknown reasons (44,45).
For cancers showing excessrisks among farmers, the MRR estimates are not large. It is perhaps not surprising that the relative risks from these summary mea-212 sures were small because the broad occupational category of farmer was all that was available for these analyses. Since not all farmers have the same exposures, combining those with different exposures would tend to dilute the effects of relevant exposures and bias risk estimates toward the null (46). The potential magnitude of such a dilution effect can be illustrated with data from a recent study in Iowa and Minnesota (47). Among the 698 population-based referents who ever lived on farms, 110 (160/0) never used insecticides and 344 (490/0) never used herbicides. The proportion of farmers who used specificclassesof pesticideswas even smaller. Approximately 40% of the farmers used phenoxy acid herbicides and 20% used organochlorine insecticides, the two most frequently used pesticide classes. Even if these chemicals were strong risk factors for a particular cancer, analyses based simply on the occupational title of farmer could seriously underestimate the relative risks. For example, if the 40% of the farmers who used phenoxy acid herbicides had twice the risk of some disease as nonusing farmers, an analysis that considered all farmers exposed would yield a relative risk of 1.4. If only 20% of the farmers used the hazardous substance of interest (eg, organochlorine insecticides), the relative risk for farmers would be 1.2. These examples represent classes of chemicals, so the dilution effect would be more dramatic for specific chemical agents. Classifying farmers by specific exposures is clearly preferable, but such a classification was not available from the surveys included in this review.
To overcome this bias, future studies of farmers should focus on carcinogenic risks from specific exposures, including pesticides, fertilizers, fuels and engine exhausts, organic and inorganic dusts, solvents, ultraviolet light, and zoonotic viruses. Such studies may provide explanations for the rising incidence of certain cancers (5,42) and the carcinogenic potential of environmental agents that are much more difficult to evaluate in the general population.
Several etiologic clues to farming-related cancer already exist. Because of the outdoor nature of their work, farmers may have considerable exposure to ultraviolet light, a strong risk factor for melanoma and cancer of the lip (48). Exposure to pesticides, particularly phenoxy acid herbicides, has been linked to non-Hodgkin's lymphoma (49)(50)(51)(52)(53) and soft-tissue sarcoma (54)(55)(56)(57), although risk estimates vary widely among studies and in some the risk was not increased at all (3). Insecticides have been associated with leukemia (47,58), multiple myeloma (59,60), and brain cancer (61). While the use of fertilizers has not been evaluated in relation to cancer risk among farmers, nitrates have been related to stomach cancer in some epidemiologic and experimental investigations (62). The growing contamination of drinking water sources with nitrates in agricultural areas (63) makes this an issue of special concern. In addition, a study of Canadian farmers reported an association between non-Hodgkin's lymphoma and expenditures on fuels and oils, which may reflect exposure to these substances or to engine exhausts (25). A role for fumigants and organic dusts is suggested by the elevated risks of non-Hodgkin's lymphoma among grain handlers in the United States (64) and stomach cancer among grain farmers in China (65). Farmers may also come into contact with animals infected with oncogenic viruses. Excessesof acute lymphocytic leukemia were reported among farmers residing in Iowa counties where dairy herds were infected with the bovine leukemia virus (66), but there is no serologic evidence suggesting transmission to humans (67,68). However, other oncogenic viruses are prevalent in farm animals, and excessrisks of soft-tissue sarcoma, non-Hodgkin's lymphoma, and leukemia have been observed among slaughterhouse workers (3) and veterinarians (2). These occupations are interesting because of the frequent contact with animals and animal products, but only limited exposure to agricultural chemicals.
Although agricultural pesticides have been linked to certain cancers, the mechanisms of action are obscure. Some pesticides are genotoxic . In an evaluation of genetic damage from 65 pesticides in 14 in vivo and in vitro tests, nine were active in most tests, 26 were active in several tests, and 30 were inactive in all tests (69). Some pesticides may operate through epigenetic pathways, including the immune system. In experimental studies, pesticides have caused a variety of immune defects including decreased host resistance to infection, thymus atrophy, suppressed T-cell activity, enhanced B-and T-cell immune response, and contact hypersensitivity (70). Newcombe (71) recently proposed that organophosphates may play a role in oncogenesis through their inhibition of serine esterases, which are critical components in the cytolytic activities of T lymphocytes and natural killer cells.
Although only limited data are available on immune alterations from human pesticide exposure (70), it is well documented that patients with naturally occurring or medically induced immunodeficiencies experience striking excesses of non-Hodgkin's lymphoma (72)(73)(74)(75)(76)(77). In addition, leukemia and stomach cancer appear to be excessive among persons with primary immunodeficiency syndromes; soft-tissue sarcomas, melanoma, and squamous carcinomas of the skin and lip occur in renal transplant recipients (75,76); and brain and skin cancers occur among bone-marrow transplant recipients (77). Since several of the tumors seen in immunocompromised subjects, including non-Hodgkin's lymphoma, appear to be excessive among farmers, it is possible that agricultural exposures affect cancer risks through immunologic perturbations that remain to be identified.
Clues to carcinogenicexposures among farmers have come primarily from case-referent interview studies involving soft-tissue sarcomas and cancers of the lymphatic and hematopoietic system. Because they are typically independent operators, farmers are able to provide considerable detail about their work practices and chemical exposures so that case-referent studies appear well-suitedfor evaluating cancers (eg, stomach, brain, and prostate) that have received little attention. The infrequent use and the changing patterns of the use of agricultural chemicals suggest, however, the utility of prospective studies as well. By incorporating environmental and biologic measures with detailed interview data, prospective studies could improve precision in exposure assessment, identify early markers for disease, eliminate the possibility of case-recall bias, and help to develop and, evaluate preventive approaches. Such studies would need to be very large, however, because the cancers of interest are rare. Once established, however, the study population could also be used to investigateoutcomes other than cancer, such as neurotoxicity, immunotoxicity, and birth or developmental defects, which are also of concern from pesticide exposure (78). In the United States, the National Cancer Institute, in collaboration with the Environmental Protection Agency and the National Institute for Environmental Health Sciences, is undertaking such a prospective study.
Most epidemiologic studies to date have focused on male operators of family farms. More attention should be given to migrant workers and farm laborers, who may experience greater exposures to some chemicals than owners or operators. Furthermore, studies in tropical countries would be useful because of climatic conditions that result in heavier use of pesticides than in temperate areas. More data are also needed on cancer risks among women who are farmers and spouses of farmers. Exposures among spouses and other dependents who are not directly engaged in farm work, although lower than among farmers and farm workers, may be greater than exposures among the general population. Studies that include dependents with indirect exposures could provide insights into risks the general public may face.
In summary, despite low risks for most major causes of death, farmers tend to be at higher risk for cancers of the lip, melanoma, brain, prostate, stomach, connective tissue, and lymphatic and hematopoietic system than the general population. The incidence and mortality rates for many of these tumors are increasing in the general population of developed countries, although there is much controversy about the relative impact of diagnostic and reporting changes versus environmental hazards on the upward trends (5,79). Several aspects of our analysis limit interpretation. We combined results from studies from several countries that may differ regarding the time period covered, type of agriculture practiced (which would result in different exposures), comparison population used, and ethnic and sociodemographic characteristics of the populations surveyed. These differences undoubtedly introduce "noise" into the summary risk estimates, and they would tend to obscure real patterns. On the other hand, because 26 outcomes were considered, chance effects cannot be entirely ruled out for some of the MRR estimates. However, the MRR estimates were significantly less than 1.0 for nine causes of death and significantly elevated for seven, far more than would be expected from chance. Consistency of the findings across the various studies provides some protection against purely chance results. Despite these limitations the data suggest that farmers experience excesses for certain tumors. Analytic studies aimed at detecting carcinogenic exposures in the agricultural setting are needed to provide clues as to which factors may contribute to these excesses. Such investigations may also identify environmental factors which may contribute to rising incidence rates for corresponding tumors in the general population.