Farming and Prostate Cancer Mortality

Transcription

1 American Journal of Epidemiology Vol. 137, No. 3 Copyright C 1993 by The Johns Hopkins University School ot Hygiene and PubBc Health Printed in U.S.A. AJ rights reserved Farming and Prostate Cancer Mortality Howard Morrison 1-2 David Savitz, 2 Robert Semenciw, 1 Barbara Hulka, 2 Yang Mao, 1 Deirdre Morison, 3 and Donald Wtgle 1 Although farmers appear to be at an increased risk of prostate cancer, trie specific exposures which produce the excess risk remain unexplained. This study was based on a retrospectively assembled cohort of male Manitoba, Saskatchewan, and Alberta, Canada, farmers age 45 years or older identified in the 1971 Canadian censuses of population and agriculture. The cohort was linked to the Canadian National Mortality Database using an iterative computer record linkage system for the period June 1971 to the end of A total of 1,148 prostate cancer deaths and 2,213,478 personyears were observed. Using Poisson regression, the study examined the relation between the risk of dying from prostate cancer and various farm practices as identified on the 1971 Census of, including exposure to chickens, cattle, pesticides, and fuels. A weak, but statistically significant, association was found between number of acres sprayed with herbicides in 197 and risk of prostate cancer mortality. When the analysis was restricted to farmers believed to be subject to the least amount of misclassifteation, the risk associated with acres sprayed with herbicides increased (rate ratio (RR) = 2.23 for 25 or more acres sprayed; 95% confidence interval (Cl) ; test for trend, p <.1). No other farm exposure examined was associated with any detectable pattern of increased or decreased risk. These findings encourage further research to examine the effects of herbicides on prostate cancer. Am J Epidemiol 1993;137:27-8. herbicides; occupational medicine; prostatic neoplasms Cancer of the prostate is the most common cancer, except for nonmelanoma skin cancer, among males in the United States (1) and the second most common in Canada Received for publication January 21, 1992, and in final form August 21, Abbreviations: O, confidence interval; 2,4-D, (2,4-dtehforophenoxy)acetic add; RR, rate ratio 1 Laboratory Centre for Disease Control, Department of National Health and Welfare, Ottawa, Ontario, Canada. 2 Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC. 3 Environmental Health Directorate, Department of National Health and Welfare, Ottawa, Ontario, Canada. Reprint requests to Dr. Howard Morrison, Bureau of Chronic Disease Epidemiology, Health and Welfare Canada, Tunney's Pasture, Ottawa, Ontario, Canada K1A OL2. Supported in part by a training grant from the National Cancer Institute. The authors gratefully acknowledge the contributions of the following persons: Martha Fair, Deborah Jordan- Simpson, and Christine PolKjun of the Canadian Centre for Health Information, Statistics Canada, and the Alberta, Saskatchewan, and Manitoba Vital Statistics Registrars. (2). However, few epidemiologic studies of prostate cancer have implicated specific environmental or occupational exposures (3). The large majority of studies of farmers and prostate cancer have noted an increased risk. Of 24 studies of farming and prostate cancer reviewed by Blair and Zahm (4), 17 showed an increased risk, although only in 1 were results statistically significant. Milham (5) found that ranchers in Washington State appeared to be at a significantly increased risk of prostate cancer. Nonsignificant increases were noted for chicken and dairy farmers. Missouri farmers who raised livestock were also found to be at an increased risk in a study by Brownson et al. (6). A study by Decoufle et al. (7) noted an odds ratio for prostate cancer of 4.8 for dairy farmers. However, this estimate was based on only five cases. 27

2 Farming and Prostate Cancer Mortality 271 Blair and Fraumeni (8) reported a weak but significant positive correlation between county prostate cancer mortality rates and numbers of chickens per county. Delzell and Grufferman (9) noted that proportional mortality ratios for prostate cancer among farmers were highest in counties that reported large numbers of chickens and cattle. A small case-control study of prostate cancer in North Carolina (1) showed a positive association with fanning employment. However, no relation was found with pesticide or gasoline exposure. Although previous studies have typically found an increased risk of prostate cancer among farmers, no specific farm exposures have been implicated. Our study was designed to examine why farmers appear to be at an increased risk of prostate cancer. MATERIALS AND METHODS This cohort study examined the relation between farming practices, sociodemographic variables, and the risk of dying of cancer of the prostate among farmers. A cohort of Manitoba, Saskatchewan, and Alberta, Canada, farm operators was assembled by Statistics Canada by linking records from the 1971 Census of, the 1971 Central Farm Register, and the 1971 Census of Population (figure 1). After preprocessing, records of 156,242 male farm operators were available for the mortality linkage (92.9 percent of male farm operators active in 1971). The 1971 Census of Population consisted of two types of questionnaires, form 2A (short) and form 2B (long). The short form was administered to two thirds of the population, with the remaining one third receiving the long form. The short form contained 19 questions, of which six were concerned with population variables. The long form included 5 questions related to socioeconomic conditions and 2 questions relating to housing, in addition to the 19 questions on the short form. This information was then linked to the 1966, 1976, 1981, and 1986 Central Farm Registers and their corresponding censuses of agriculture in order to examine specific farming practices and exposures reported in the censuses of agriculture. This analysis is based upon data contained in the 1971 and 1981 censuses; herbicide information was not obtained on either the 1966 or 1976 censuses. The 1971 and 1981 censuses of agriculture contained certain questions applicable to 197 and 198, respectively. These questions included the number of acres sprayed with herbicides and insecticides and expenditures paid for fuel and oil for farm operations. Deaths among the cohort for the period were determined by linking this file to the Canadian Mortality Database, which contains death records for all Canadians. Details regarding these data sources have been outlined in previous papers (11, 12). On the basis of a sample in which manual resolution was used, false-positive and false-negative rates were estimated at 8 percent each for the 2A (short form) linkage and 4 percent each for the 2B (long form) linkage (13, 14). Data analysis Person-years and deaths began to accrue from June 1, 1971, for those farmers at least 45 years old at the time of the census or when farmers attained age 45 years at some point during the follow-up period (June 1971 through December 1987). A total of 145,383 fanners reached this age during the follow-up period and contributed persontime to the study. Standardized mortality ratios were calculated, with expected numbers of deaths calculated by applying Canadian prairie (Manitoba, Saskatchewan, and Alberta) age-, calendar period-, and disease-specific mortality rates to the person-years at risk by age and calendar period using the Monson computer program (15). Confidence intervals on standardized mortality ratios were calculated based on the Poisson distribution (16). Poisson regression analysis was used to examine the relation of the following farm variables to risk of dying of cancer of the prostate: acres sprayed with herbicides; dollars spent on fuels and oils for farm purposes

3 272 Morrison et al Central Farm Register 1971 Census of 1971 Census of Population Canadian Mortality Database 1971 Farmers Cohort Farmers cohort with exposures from multiple Censuses Linked Exposure/Mortality File FIGURE 1. Data sources used in the Canadian Farm Operator Study, Census of 1976 Census of 1981 Census of 1986 Census of (as a marker for exposure to fuels and exhaust); acres treated with fertilizers; area of farm; acres in small grains, canola/mustard, and hay/other fodder, number of poultry; number of dairy cattle; and number of beef cattle. The following variables from the Census of Population were examined, as well, as potential confounders: income, education, ethnic origin, and immigrant status. One degree of freedom tests for trend were performed (17). For exposure and potential confounder variables, continuous variables were stratified into four categories. In most cases, these consisted of an unexposed category plus tertiles of the remainder. For some variables, such as fuel and oil expenditures, an "unexposed" group was not available (almost all farmers reported some expenditure on fuel), and quartiles were used. Calendar period and age were considered a priori to be potential confounders and were

4 Fanning and Prostate Cancer Mortality 273 included in all models unless otherwise specified. None of the other variables available for analysis were known to be independent risk factors for prostate cancer, but they were nevertheless assessed as potential confounders. Duration of exposure was approached in two different ways. One was to use data from the 1971 and 1981 censuses of agriculture to develop exposure indices on individual farmers. Individuals identified on the 1981 Central Farm Register were restricted to those farmers who continued farming on the same farm from 1971 to Unfortunately, most farmers from the 1971 census did not appear on later censuses because they had left farming or had bought a different farm. Those who were identified on later censuses clearly remained on the same farm and had continuing exposures, whereas those not found on later censuses had an uncertain exposure history. The second approach was to examine subgroups of farmers whose exposure and outcome statuses were more likely to be accurately classified. False-positive record linkage rates were lower for those farmers who completed the long form (2B) than for those who answered the short form (2A) because better personal identifiers were available from the 2B form. Any etiologic relations for farm exposures should be stronger for farmers with no employees, since employees could presumably be engaged in the activities that incur exposures. Accuracy in imputing exposures should also be better for farmers who reported no custom expenses (work which a farmer hired outside individuals to perform) for insecticides, herbicides, fertilizers, and, to a lesser extent, fuels and oils. RESULTS Standardized mortality ratios for farmers age 45 years or more were calculated. Significantly reduced standardized mortality ratios were noted for most cancer sites examined, including cancers of the buccal cavity and pharynx, stomach, colon, rectum, pancreas, lung, prostate, and bladder, and non-hodgkin's lymphoma. The standardized mortality ratio for all-cause mortality was.71 (95 percent confidence interval (CI).7-.72), whereas that for prostate cancer was.92 (95 percent CI ). Standardized mortality ratios for prostate cancer decreased that increasing age, falling from 1.44 (95 percent CI ) for the age group years to.63 (95 percent CI ) for the oldest age group (age 85 years or more). Results from the Poisson regression analysis are presented in tables 1 through 4. Unless otherwise stated, all mortality rate ratios presented have been adjusted by 5-year age group and calendar period. No striking pattern was apparent when risk of prostate cancer was examined by acres in various crops (table 1). No consistent change in risk was noted according to numbers of animals. When the entire cohort was analyzed, a small statistically nonsignificant decreased risk was noted for the highest exposure categories of dairy cattle (rate ratio (RR) =.78, 95 percent CI.6-1). A small increased risk was noted according to the number of beef cattle (RR = 1.17, 95 percent CI ). Restriction of the analysis to the one-third sample resulted in decreased odds ratios for the highest exposure levels of dairy cattle (RR =.62, 95 percent CI.38-) and poultry (RR =.79, 95 percent CI.59-5). Number of acres sprayed with herbicides was associated with an increased risk of prostate cancer. No consistent pattern was apparent for fuel and oil expenses, fertilizers, or insecticides. Restricting the analyses to younger farmers and to farmers who reported no employees and no custom expenses (data not shown) did not have any apparent effect on the exposure-response relation of either fertilizers or fuels and oils. There was some evidence of minor confounding of these variables by herbicide exposure. However, no rate ratios for fuels and oils or fertilizers were statistically significant for any of the models, nor was any exposureresponse relation apparent. To investigate the effect of confounding on the relation between herbicide spraying

5 274 Morrison et al. TABLE 1. Polsson regression analysis of farmers age 45 years or older, prostate cancer, adjusted* mortality rate ratios, Canadian prairie Canda/mustard ( 12 vs. acres) Hay/fodder ( 7 vs. acres) Other crops ( 8 vs. acres) Small grains ( 38 vs <13 acres) Farm size ( 1, vs. <5 acres) Dairy cattle ( 1 vs. ) Beef cattle ( 5 vs. ) Poultry ( 1 vs. ) Insecticides ( 15 vs. acres) Herbicides ( 25 vs. acres) Fuel/oil ( $9 vs. <$35) Fertilizer ( 25 vs. acres) Custom expenses ( $ vs. $) Agricultural workers ( 2 vs. ) RRt Entire cohort AD models were adjusted for 5-year age group and calendar period. t RR, rate ratio; Cl. confidence Interval. and prostate cancer, we constructed models in which adjustments were made for potential confounders. Models also were constructed that restricted analysis to farmers most likely to be exposed to herbicides (those with no custom expenses and no agricultural employees) or to phenoxy herbicides (Saskatchewan). Results in table 2 are based on analysis of herbicide exposure and prostate cancer using the one-third sample. Mortality rate ratios for herbicide exposure tended to be higher than those obtained when the analysis was based on the entire cohort (table 1). As with the entire cohort, there was no evidence of confounding. Restriction of the analysis to those most likely to have been exposed tended to increase the size of the mortality rate ratios for the highest exposure category, with multiple restrictions yielding rate ratios of in the most highly exposed stratum. No statistically significant interactions were detected. Risks by herbicide spraying were higher in Manitoba and Alberta and lower in Saskatchewan; however, an increased risk was observed for the highest exposure category for all three provinces. 95% at RR One-ttilrd sample 95% ci A stratified analysis compared mortality among farmers by exposure categories in 1971 and Farmers identified on both the 1971 and 1981 censuses of agriculture were included, with mortality follow-up for the period Results are displayed in table 3. Rate ratios of 1.32 and 1.11 were observed for herbicides and for fuels and oils, respectively, comparing farmers in the highest category for both 1971 and 1981 with farmers in the lowest exposure categories for both censuses. Neither rate ratio was statistically significant, reflecting not only the modest increases in risk, but a dramatic loss of power compared with an analysis based on follow-up from 1971 onward. Prostate cancer mortality by 198 herbicide usage is displayed in table 4. No relation was observed between acres sprayed in 198 and prostate cancer mortality ( ). DISCUSSION This large cohort study suggests that exposure to herbicides results in an increased

6 Farming and Prostate Cancer Mortality 275 TABLE 2. Poisson regression analysis of prostate cancer, adjusted* mortality rate ratios according to herbicide use, one-third sample of Canadian prairie farmers age 45 years or older, , restricted to subgroups Restricted to No restriction No hired help No custom expenses No hired help and no custom expenses Age years Age years, no custom expenses, and no employees No. of deaths Acres sprayed > >25 2:25 2=25 Al models were adjusted for 5-year age group and calendar period, t RR, rate ratio; Cl, confidence Interval. RRt 95% Clt Test for trend, p = Test for trend, p = Test for trend, p = Test for trend, p = Test for trend, p = Test for trend, p <.1 The possibility that finding an exposureresponse relation between herbicides and prostate cancer reflects some undetected (and hence, uncorrected) bias cannot be discounted. This is made particularly difficult because of the absence of clear risk factors for prostate cancer (except for age and family history). The only other likely possible confounder that was uncontrolled for in this study was diet; there is evidence that a high- risk of dying from prostate cancer. There is evidence of an exposure-response relation between herbicide exposure and prostate cancer mortality. Tests for trend were statistically significant. Adjustment for the potentially confounding effects of other farm exposures did not alter the risk relation. No other farm exposure examined was associated with any clear pattern of increased or decreased risk.

7 276 Morrison et al. TABLE 3. Polsson regression analysis of prostate cancer by chemical exposure, adjusted mortality rate ratios,* Canadian prairie farmers age 45 years or older, , as identified on the 1981 Census of Variable Herbicides (1971/1981) Low/low High/high Fuels/oils (1971/1981) Low/low High/high Adjusted for 5-year age group. t RR, rate ratio, Cl, confidence Interval. Person-years 15, , , ,817. TABLE 4. Poisson regression analysis of prostate cancer adjusted* mortality rate ratios according to 198 herbicide use, Canadian prairie farmers identified on the 1971 Census of, age 45 years and older, Acres of herbicide sprayed >25 No. of deaths RRt % Clt * The mode) was adjusted for 5-year age groups, t RR, rate ratio; Cl, confidence Interval. fat diet predisposes men to developing prostate cancer (3, 18), and farmers in this cohort as a group ate a relatively high-fat diet (13). However, for dietary fat to explain the relation between herbicides and prostate cancer, fat consumption for farmers would have to increase with increasing acres sprayed with herbicides. Herbicide exposure was inferred from selfreported farm practices from the 1971 Census. However, it is not known who applied the herbicides, but merely how many acres were treated with herbicides. The extent of use of protective equipment when applying herbicides is not known. No information is available regarding actual absorbed doses of herbicides. However, reasonable agreement was found between what was recorded for individual farm operators on various censuses of agriculture with regard to exposure variables such as the number of cattle and No. of deaths RRt %Clt acres sprayed with herbicides, suggesting that exposures over time were not badly misclassified (13). An approach to reducing exposure misclassification was to restrict the analysis to those farmers with no employees and/or those who reported no expenses for paid assistance, which could include herbicide spraying. Such individuals should be more likely to have applied the herbicides themselves. Risk relations were strengthened when the analysis was restricted to this subgroup. Herbicide exposure was also examined when the analysis was restricted to younger farmers (ages years). An exposureresponse relation may be stronger among younger farmers because there is likely to be less misclassification of both exposure and cause of death among the young. It is reasonable to assume that exposure histories may be less accurate among the elderly. A sizeable minority of farm operators were over age 7 years at the time of the 1971 Census of ; some were over age 9. It is likely in these cases that these farmers experienced limited farm exposures. Death certificates are also less likely to be accurate among the very old. A second reason why an analysis restricted to younger farmers may result in stronger exposure-response relations is the "survivor" effect. Numerous studies have noted that risk relations often are stronger among the young (19, 2). It is presumed that this oc-

8 Farming and Prostate Cancer Mortality 277 curs because exposed individuals who survive to old age may be less susceptible to the exposure agent if those susceptible have already succumbed. In this study, risk measures for farmers according to acres sprayed with herbicides were higher when the analysis was restricted to younger farmers. It is also possible to examine the situation in which nondifferential misclassification of disease status is lessened. It was predicted that there should be less misclassification when the analysis is restricted to the onethird sample of farmers who completed the long form of the census of population. Farmers who completed the long form provided more personal identifiers, which should have resulted in better record linkage (and, hence, fewer false-positive links). In the case of herbicide exposure, restricting the analysis to the one-third sample increased the strength of the exposure-response relation. The longitudinal data were used to compare farmers who were in the highest herbicide exposure category for both the 1971 and 1981 censuses with farmers in the lowest herbicide category for both censuses. Mortality rate ratios for the entire cohort were comparable with those noted when the analysis was restricted to farmers who completed the long form of the Census of Population. However, no relation was observed between acres sprayed with herbicides in 198 and risk of prostate cancer mortality from There are several possible explanations for this apparent lack of consistency. One is that the exposure-response relation which was observed using either 1971 or 1981 data resulted from chance. Another explanation is that recent exposures are irrelevant, either because herbicides act as initiators or because the nature of herbicides used on the Canadian prairies has changed over the last 2 years. Since the late 196s, there has been a shift away from broadleaf herbicides such as (2,4-dichlorophenoxy)acetic acid (2,4-D) toward wild oat herbicides (21). Changes in herbicide use patterns will tend to obscure risk relations, assuming that some, but not all, such agents are carcinogenic. Phenoxy herbicides (primarily 2,4-D) were the most commonly used herbicides on the Canadian prairies during the study period (21). If exposure to phenoxy herbicides resulted in an increased risk of prostate cancer, the risk relation should have been strongest in Saskatchewan. Although the differences were not large, phenoxy herbicides comprised a larger percent of total herbicides sprayed in Saskatchewan during the study period than in Alberta and Manitoba. Risks by herbicide exposure were lower in Saskatchewan, suggesting that phenoxy herbicides may not be the responsible agent. However, an increased risk was only associated with 197 and not with 198 herbicide spraying. Given the trend in herbicide use away from phenoxy herbicides and toward other herbicides, this is evidence that phenoxy herbicides may be responsible. To our knowledge, this is the first study to have sufficient power to examine adequately the issue of herbicide exposure and risk of prostate cancer, as only small studies have been reported previously. Of five studies of manufacturing cohorts that reported an association between phenoxy herbicides and prostate cancer, four were weakly positive (table 5). Two studies of pesticide applicators were weakly negative, while one small study of herbicide applicators was positive. A case-control study by Checkoway et al. (1) noted that herbicides could not account for the association that they found between prostate cancer and farming. Small numbers hampered this study; only three cases and three controls reported exposure to herbicides. Fewer than 2 percent of the farmers in this study reported exposure to pesticides, herbicides, or gasoline. This suggests either significant underreporting of exposure to these common farm agents or a farm population markedly different in its exposures from those of the present study. A study by Le Marchand et al. (3) noted a small increased risk of prostate cancer associated with self-reported pesticide exposure. There is insufficient evidence from animal studies to determine if phenoxy herbicides are carcinogenic. A New Zealand study

9 278 Morrison et al. TABLE 5. Prostate cancer mortality among cohorts of male workers employed in the manufacture or application of pesticides Authors (reference no.) Observed deaths Expected deaths FIR* Population Coggon et at. (22) Bond et al. (23) Lynge (24) Cook et al. (25) Ott et al. (26) Rlihlmakl et al. (27) Blair et al. (28) Wlklund et al. (29)t Manufacture of MCPA* Manufacture of phenoxy herbicides Manufacture of phenoxy herbicides Manufacture of phenoxy herbicides Manufacture of 2,4,5-T* and chlcrophenols Herbicide applicators Pesticide applicators Pesticide applicators RR, estimated relative risk; MCPA, 2-methyl-4-cfi)oropheooxyac8tic acid; 2,4,5-T, 2,4,5-trlchlorophenoxyacetlc acid, t Incident cases. noted a statistically significant relation between small intestinal adenocarcinoma in slaughtered sheep and exposure to phenoxy acid (31). Companion dogs exposed residentially to 2,4-D have been reported to be at an increased risk of malignant lymphomas (32). Evidence of a genotoxic effect is equivocal (31). Several 2-year animal bioassay studies have failed to note any carcinogenic effects (33). The International Agency for Research on Cancer has concluded that there is limited evidence that phenoxy acid herbicides are carcinogenic in humans, with inadequate evidence in animals (31). The nature of this study precludes directly addressing the question of whether farmers as a whole are at an increased risk of prostate cancer. The nature of record linkage with incomplete personal identifying information means that using an external control group will underestimate risk. Farm operators not identified as having died were assumed to still be alive. Given that 1-15 percent of deaths were likely to have been missed (11), it is likely that this cohort probably experienced a small increased risk of dying of prostate cancer relative to other male residents of the Canadian prairies. It has been speculated previously that farmers may be at an increased risk of prostate cancer because of exposure to fertilizers, possibly contaminated with cadmium (34). This is not borne out by our study, which found a slight decrease in risk for farmers who used fertilizers. Other studies have noted associations between prostate cancer and chickens (8, 35) and cattle (5, 6, 35). No association was detected in this study for chickens. A very small, statistically nonsignificant increased risk was associated with the highest quartile of exposure to beef, but not to dairy, cattle. The nature of this study prevents concluding that farm exposures other than herbicides are unrelated to the risk of dying of prostate cancer. The main limitation of the study, nondifferential exposure misclassification, is likely to have biased any positive or negative farm exposure risk estimates toward the null. This may mean that other risk factors due to farm exposure were not detected because of exposure misclassification coupled with relatively low risks. However, restricting the analysis to subgroups of farmers less likely to be misclassified did not strengthen the exposure-response relation for any exposure other than herbicides, suggesting that other exposures were not likely to be related to prostate cancer risk. Even given the limitations of this study, it is superior to previous studies both in size and information on farm and sociodemographic variables available on individual farmers. Study findings of an association between herbicides and prostate cancer should be considered tentative because of the relatively low increases in risk and because an association has not been noted previously. However, the cohort design obviates recall bias concerns. The positive association between

10 Farming and Prostate Cancer Mortality 279 herbicides and prostate cancer is not likely to have resulted from the types of biases to which a study of this nature is vulnerable. REFERENCES 1. American Cancer Society. Cancer facts and figures New York: American Cancer Society, National Cancer Institute of Canada. Canadian cancer statistics, Toronto, Ontario, Canada: National Cancer Institute of Canada, Flanders WD. Review: prostate cancer epidemiology. Prostate 1984;5; Blair A, Zahm SH. Cancer among farmers. Occup Med 1991;6; Milham S Jr. Occupational mortality in Washington State, Cincinnati, OH: National Institute for Occupational Safety and Health, (DHEW publication no. (NIOSH) B). 6. Brownson RC, Chang C, Davis JR, et al. Occupational risk of prostate cancer a cancer registrybased study. J Occup Med 1988;3: Decoufle P, Stanislawczyk K, Houten L, et al. A retrospective survey of cancer in relation to occupation. Cincinnati OH: National Institute for Occupational Safety and Health, (DHEW (NIOSH) publication no ). 8. Blair A, Fraumeni JF Jr. Geographic patterns of prostate cancer in the United States. J Natl Cancer Inst 1978; 61: Delzell E, Grufferman S. Mortality among white and nonwhite fanners in North Carolina, Am J Epidemiol 1985;121:391^*O2. 1. Checkoway H, DiFerdinando G, Hulka BS, et al. Medical, life-style, and occupational risk factors for prostate cancer. Prostate 1987; 1: Wigle DT, Semenciw RM, Wilkins K, et al. Mortality study of Canadian male farm operators: non- Hodgkin's lymphoma mortality and agricultural practices in Saskatchewan. J Natl Cancer Inst 199; 82: Morrison HI, Semenciw RM, Morison D, et al. Brain cancer and farming in western Canada. Neuroepidemiol (in press). 13. Morrison HI. Prostate cancer and farming. PhD. dissertation. Chapel Hill, NC: University of North Carolina at Chapel Hill, Jordan-Simpson D, Fair ME, Poliquin C. Canadian farm operator study: methodology. Health Rep 1992: Monson R. Analysis of relative survival and proportional mortality. Comput Biomed Res 1974;7: Breslow NE, Day NE, eds. Statistical methods in cancer research. Vol. 2. The design and analysis of cohort studies. IARC scientific publication no. 82. Lyon, France: International Agency for Research on Cancer, 1987: Breslow NE, Day NE, eds. Statistical methods in cancer research. Vol. 1. The analysis of case-control studies. IARC scientific publication no. 32. Lyon, France: International Agency for Research on Cancer, 198: Slattery ML, Schumacher MC, West DW, et al. Food-consumption trends between adolescent and adult years and subsequent risk of prostate cancer. Am J Clin Nutr 199O;52: Semenciw RM, Morrison HI, Mao Y, et al. Major risk factors for cardiovascular disease mortality in adults: results from the Nutrition Canada Survey cohort. Int J Epidemiol 1988; 17: United States Surgeon General's Office. The health consequences of smoking: cardiovascular disease. A report of the Surgeon General. Rockville MD: US Department of Health and Human Services, Office on Smoking and Health, Manitoba Department of. Agricultural statistics: annual herbicide use surveys. Winnipeg, Manitoba, Canada: Manitoba Department of, Coggon D, Pannett B, Winter PD, et al. Mortality of workers exposed to 2 methyl-4 chlorophenoxyacetic acid. Scand J Work Environ Health 1986; 12: Bond GG, Wetterstroem NH, Roush GJ, et al. Cause specific mortality among employees engaged in the manufacture, formulation, or packaging of 2,4-dichlorophenoxyacetic acid and related salts. Br J Ind Med 1988;45: Lynge E. A follow-up study of cancer incidence among workers in manufacture of phenoxy herbicides in Denmark. Br J Cancer 1985;52: Cook RR, Bond GG, Olson RA, et al. Evaluation of the mortality experience of workers exposed to chlorinated dioxins. Chemosphere 1986; 15: Ott MG, Olson RA, Cook RR, et al. Cohort mortality study of chemical workers with potential exposure to the higher chlorinated dioxins. J Occup Med 1987;29: Riihimaki V, Asp S, Hernberg S. Mortality of 2,4- dichlorophenoxyacetic acid and 2,4,5-trichlorophenoxyacetic acid herbicide applicators in Finland: first report of an ongoing prospective cohort study. Scand J Work Environ Health 1982;8: Blair A, Grauman DJ, Lubin J, et al. Lung cancer and other causes of death among licensed pesticide applicators. J Natl Cancer Inst 1983;71: Wiklund K, Dich J, Holm LE, et al. Risk of cancer in pesticide applicators in Swedish agriculture. Br J Ind Med 1989;46: Le Marchand L, Kolonel LN, Yoshizawa CN. Lifetime occupational physical activity and prostate cancer risk. Am J Epidemiol 1991; 133: International Agency for Research on Cancer. Monographs on the evaluation of the carcinogenic risk of chemicals to humans. Supplement 7. Overall evaluations of carcinogenicity: an updating of IARC monographs volumes 1 to 42. Lyon, France: International Agency for Research on Cancer, Hayes HM, Tarone RE, Cantor KP, et al. Casecontrol study of canine malignant lymphoma: positive association with dog owner's use of 2,4-dichlorophenoxyacetic acid herbicides. J Natl Cancer Inst 1991;83: Bond GG, Bodner KM, Cook RR. Phenoxy herbicides and cancer insufficient epidemiologic evi-

11 28 Morrison et al. dence for a causal relationship. Fund Appl Toxicol 1989; 12: Schwartz GG, Hulka BS. Is vitamin D deficiency a risk factor for prostate cancer? (Hypothesis). Anticancer Res 199;1: Blair A, Malker H, Cantor KP, et al. Cancer among farmers: a review. Scand J Work Environ Health 1985; 11:397^17.