Non-Hodgkin's lymphoma and agricultural practices in the prairie provinces of Canada.

1994;20:42-7. OBJECTIVES -_. The aim of this study was to provide an update of a cohort study (1971-1985) that previously reported a significant trend in the risk of non-Hodgkin's lymphoma among male Saskatchewan farm operators according to fuel-oil expenditures and herbicide spraying for farms less than 1000 acres (2570 hectares) by including two additional Canadian prairie provinces, two ad ditional years of follow-up, and data from the 1981 Census of Agriculture. METHODS - Information on farmers from 1971 records of the Census of Agriculture was linked to 1971 records of the Census of Population, to 1981 records of the Census of Agriculture, and to death records. Poisson regression was used to estimate risks according to herbicide spraying and fuel and oil expenditures. RESULTS The addition of a further two years of follow-up resulted in lower risk estimates associated with her bicide spraying for Saskatchewan. No excess risk was observed between herbicide spraying and non Hodgkin's lymphoma for Alberta or Manitoba in the 1971 data. However, a significantly increased risk of non-Hodgkin's lymphoma according to acres sprayed with herbicides was observed for the three provinces combined when the herbicide spraying data from the 1981 Census of Agriculture was used [~380 acres (~939 hectares) sprayed, rate ratio 2.11, 95% confidence interval 1.1-3.9]. CONCLUSIONS - Although the current results are not entirely consistent with the original Saskatch ewan analysis, they support the overall finding of an association between herbicides and risk of fatal non-Hodgkins lymphoma. Prospective cohort studies are needed to overcome the limitations of exist ing epidemiologic studies.

A recent analysis of Saskatchewan farmers (9) noted an increased risk of non-Hodgkin's lymphoma mortality with fuel and oil expenditures and with increasing numbers of acres sprayed with herbicides on farms of less than 1000 acres (2570 hectares). Using data for all three Canadian prairie provinces and adding two years of follow-up plus additional census data, this report updates the study.
The Canadian farm operator study was established to examine the relation between various farm practices and specific cancers. In addition to the study of non-Hodgkin's lymphoma among Saskatchewan farmers (9), other papers have examined cancers of the brain (12), prostate (13), and multiple myeloma (14).

Data sources
The data sources used in this study are described in the original Saskatchewan paper (9). In brief, a cohort of male farm operators was assembled by Statistics Canada, and data from the 1971 Canadian censuses of agriculture and population were linked. The results were then linked to death records contained in the Canadian Mortality Data Base with the use of the Generalized Iterative Record Linkage System (15,16). Death certificates were obtained by Statistics Canada for farmers whose underlying cause of death was recorded as non-Hodgkin's lymphoma. Occupation, as recorded on the death certificates, was used to remove probable false links resulting from the use of the computerized record linkage system.
After the preprocessing of the Central Farm Register and the linking of the agriculture and population censuses, records of 156242 male farm operators (92.9% of all male farm operators active in 1971) from the Canadian prairie provinces of Manitoba, Saskatchewan, and Alberta were available for the mortality linkage. The results of this report are based on the mortality experience of farmers whose age was at least 35 years at some point during the follow-up period (June 1971 through December 1987). A total of ISS 547 men met this criterion.
In addition, the Agriculture Division of Statistics Canada provided a longitudinal file containing infor-matio n on farming practices from the Census of Agriculture from 1966Agriculture from , 1971Agriculture from , 1976Agriculture from , 1981Agriculture from , and 1986. It was created by mergin g data for each farm on the basis of location and the operator's surname. For this analysis , the age of the farm operator on each census was compared, and records were removed if the age of an operator, as given on successive census forms, did not increase by the expect ed five-year increment. For data for a farm operator from the 1981 Censu s of Agriculture to be included, the operator had to be alive and living on the same farm as at the time of the 1971 cens us. Data fro m the Census of Agriculture on the number of acres sprayed with herbicide s and expenditures on fuel and oil applied to the year prior to the census (eg, the 1971 census data for these variables were based on the calendar year 1970) .
Data on herbicide exposure were stratified into an unexposed category plus categories corresponding to Census of Agriculture-specific tertile s. Fuel and oil expend itures were stratified into census-specific quartiles, as an "unexposed" group was not available (almost all farmers reported some expenditure on fuel). Because of small numbers, quartiles two and three were combined in the cross-classified analysis.

Statistical ana lysis
Standardized mortality ratios (SMR) were calculated with the combined Manitoba, Saskatchewan , and Alberta male age-s pecific mortality rates as the refe rence, with confidence intervals based on the Poisson distribution (17).
Models were fitted to the grouped data with the use of Poisson regression analyses (18,19). Farm practices were evaluated for confo unding although Scand J Work Environ Health 1994, vol 20, no I none was known to be an independent risk factor. Models were constructed which included terms for age, calendar-year per iod, fuel-oil expenditures, and the number of acres sprayed with herbicides. Onedegree-of-freedom tests for trend in relative risk with increasing exposure were performed in which a continuous variable was added to a model containing the remaining terms, and the improvement in fit was assessed ( 18).
Because the original Saskatchewan analyses noted an increased risk of non-H odgkin' s lymph oma with increas ing acres sprayed with herbi cides for farms of less than 1000 acres (2570 hectares) only, model s were also constructed in which only farmers of farms less than 1000 acres were included.

Mortality
The mortality rates for major causes of death among the 155547 farm operators aged 35 years or more were significantly below those for men in the provinces as a whole, with an SMR for all-ca use mortality of 0.72 . The observed number of deaths among the farm operators was also significantly lower than expected for non-Hodgkin's lymphoma (SMR =0.79, 95% CI 0.69-0.89).

Non-Hodgkin's lymph oma and fa rming practices
Estimated relati ve risks of death due to non-Hodgkin' s lymphoma accordin g to acres sprayed with herbicides and fuel and oil expenditures in 1970 are displayed in table I for operators of Saskatchewan farms. A small, but not statistically significant, increased risk for acre s sprayed with herbicides was and with log transformation (P = 0.02), it was not significant after control for fuel and oil expenditures (P =0.16).
The estimated relative risks by acres sprayed with herbicides for all three prairie provinces combined were lower than for Saskatchewan alone (table 2).
In Manitoba, the risks were highest among the farmers reporting no herbicide use (table 3); no discernable pattern was observed in Alberta.
Use of the longitudin al file allowed assessment based on farm pract ices reported for 1980, as well as for 1970. For far m operators who lived on the same farm in both 1971 and 1981, according to 1980 farming practices with mortality follow-up from 1981 to 1987, an increased risk of non-Hodgkin's lymphoma mortality associ ated with acreage sprayed with herbicides was observed for the prairies as a whole (table 4). The rate ratio for the highest quartile of 380 or more acres (939 or more hectares) sprayed was 2. 11 (95% CI 1.13-3.93). No difference in risk according to acres sprayed with herbicides was ob-

Discussion
The overall death rates were significantly lower among the male farm operators than among the general male pra irie population. A similar reduction wa s observed in the original Saskatchewan anal ysis (9).
It presumably reflects both the relatively low overall mort ality among farmers previ ously noted in other studies (20) , as well as a failure to identify correctly, with the use of the computerized linkage to the Canadi an Mort ality Data Base, all of the farmers who had died.
The previous analysis of Saskatchewan farmers noted a small , statistically non significant deficit of deaths from non-Hodgkin' s lymphoma (9). However, the present update based on data from all three prairie provinces noted a significant deficit, which reflected both low mortality from this cause among the cohort members in Manitoba and Alberta and increased statistical power.
The present study noted somewhat lower risks associated with herbicide spraying for Saskatchewan than those observed in the original anal ysis, in which mortality wa s followed onl y until the end of 1985 . This change may have been a chance finding, or it may reflect increasing exposure misclassification with longer follow-up . Rate ratios according to fuel and oil expenditures for Saskat chewan farmers were similar in both studies.
Inclusion of data from the additional two prairie prov inces, in particular Manitoba, largely removed the positive relationship noted in the Sask atchewan analysis between herbicide spraying and non-Hodgkin ' s lymphoma mortality. There are the following two possibilities: either the original Saskatchewan findings or the pre sent findings occurred due to chance or increased risks were restri cted to Saskatchewan and , to a much less extent, Alberta. Phenoxy herb icide use, relati ve to other herbicide use, is more common in Saskatchewan than in Manitoba or Alberta (2 1). Amon g prairie farmers reporting phenoxy herbicide use, Saskatchewan farmers were the most likely to use 2,4-dichlorophenoxyacetic acid (2,4-0); the Manitoba farmers were the lea st likel y (22). According to both the 1983 and 1984 national farm surveys, amon g prairi e farm operators who had pesticides applied to thei r farm , those in Sas katchewan were the most likely to have appli ed the pesticides them selves, and Manitoba farmers were the lea st likel y (Agriculture Division , Stati stic s Canada, unpublished data ). However , any exposure-disease relation ship in the present study between 2,4-0 and non-Hodgkin' s lymphoma must be inferred and must, at best, be considered tenuous; the census questionnair e asked about acre s spr ayed in 1970 with herbicides, not about exposure to 2,4-0.
Mortality from non-Hodgkin's lymphoma was significantly elevated among the farmers who reported spraying herbicides on ;::>:380 acre s (; : : >: 939 hectares) in 1980. The magnitude of the risk was comparable to that observed in the original study. Unlike the original Saskatchewan analysis, an increased risk for farmers in the highest exp osure ca tegory of herb icide exposure was not restricted to farms of less than 1000 acres (2570 hectares).
There is only limited evidence suggesting that phenoxy herbicides may be car cinogenic in animals. A statistically significa nt rel ationship bet ween small inte stinal adenocarcinoma in slaughtered sheep and exposure to phenoxy herbi cides wa s noted in a New Zealand study (23) . Companion dogs exposed in residential settings to 2,4-0 were rep orted to be at an increased risk of de veloping malignant lymphomas (24) . A statistically significant increase in astrocytomas was observed in male rats fed high dose s of 2,4-0 (25). However, sever al two-year animal bioass ay studies have failed to note any carcinogenic effe cts (26).
Although an increased risk was observed in our study according to 1970 expenditures for fuel and oil, no significantly increased risk wa s observed according to 1980 expenditures. This finding may refle ct an insufficiently long period of time between exposure and outcome, or it may indicate the changing nature of exposure. For example, enclosed tractor cab s became more common, providing gre ater protection from engine exhaust.
The maj or limitation of thi s study is the potential for exposure misclassification. However , by including data from the 1981 cen sus, the present study has improved the exposure assessment over the earlier anal ysis of Sask atchewan farm operators (9).
Previous cas e-referent studies have been criticized because of the potential for bias res ulting from the use of prox y respondents (27). Thi s was not a concern with the present cohort study. The similarity of our current finding s with those of most case-referent studies (28) suggests that recall bias is unlikely to have produ ced the positi ve herbi cide exposure-response gradie nts obse rved in case -referent studies, and it underscore s the value of coh ort studies of herbicide-exposed workers.