Agricultural pesticide use and pancreatic cancer risk in the Agricultural Health Study Cohort
Pancreatic cancer is a rapidly fatal disease that has been linked with pesticide use. Previous studies have reported excess risks of pancreatic cancer with organochlorines such as DDT, however, many other commonly used pesticides have not been examined. To further examine the potential associations between the use of a number of pesticides and pancreatic cancer, we conducted a case-control analysis in the Agricultural Health Study, one of the largest prospective cohorts with over 89,000 participants including pesticide applicators and their spouses in Iowa and North Carolina. This analysis included 93 incident pancreatic cancer cases (64 applicators, 29 spouses) and 82,503 cancer-free controls who completed an enrollment questionnaire providing detailed pesticide use, demographic and lifestyle information. Ever use of 24 pesticides and intensity-weighted lifetime days [(lifetime exposure days) × (exposure intensity score)] of 13 pesticides was assessed. Risk estimates were calculated using unconditional logistic regression controlling for age, smoking, and diabetes. Among pesticide applicators, 2 herbicides (EPTC and pendimethalin) of the 13 pesticides examined for intensity-weighted lifetime use showed a statistically significant exposure-response association with pancreatic cancer. Applicators in the top half of lifetime pendimethalin use had a 3.0-fold (95% CI 1.3–7.2, p-trend = 0.01) risk compared with never users, and those in the top half of lifetime EPTC use had a 2.56-fold (95% CI = 1.1–5.4, p-trend = 0.01) risk compared with never users. Organochlorines were not associated with an excess risk of pancreatic cancer in this study. These findings suggest that herbicides, particularly pendimethalin and EPTC, may be associated with pancreatic cancer. © 2008 Wiley-Liss, Inc.
Pancreatic cancer is a rare, but rapidly fatal disease. It is currently the fourth leading cause of cancer death among both men and women in the United States.1 Cigarette smoking is the best established risk factor, yet only explains about 25% of the cases.2 Other suggested risk factors for pancreatic cancer include diabetes, obesity, black race, pancreatitis, family history of pancreatic cancer, and possibly, pesticide use.3
Several epidemiological studies have reported excess risks of pancreatic cancer in relation to agricultural occupations.4–6 Exposure to organochlorines, including DDT, DDD and ethylan have been associated with excess risks of pancreatic cancer in 2 case-control studies,7, 8 and significantly higher serum concentrations of organochlorine compounds were found among pancreatic cancer cases when compared with controls.9, 10 To further explore specific pesticide use and pancreatic cancer risk, we conducted a case-control analysis in the Agricultural Health Study (AHS) cohort, one of the largest prospective cohort studies of pesticide applicators and their spouses.
Material and methods
The AHS cohort with over 89,000 participants includes licensed private and commercial pesticide applicators, and spouses of private applicators residing in Iowa and North Carolina. Applicators were enrolled between December 13, 1993 and December 31, 1997 at pesticide licensing facilities in each state; 82% of applicators seeking licensing were enrolled in the study (57,311 applicators). Spouses of enrolled applicators were asked to complete a questionnaire provided by the enrolled applicator, and return it by mail or complete it over the phone, generally within 1 month of applicator enrollment. Of all spouses 75% (32,347 spouses) chose to participate. The study protocol was approved by all appropriate institutional review boards.
Applicators completed a self-administered questionnaire at enrollment eliciting information on demographics, lifestyle factors, medical histories, ever/never use of 50 pesticides, and comprehensive pesticide use (i.e., years and days per year applied pesticides, personal protective equipment, and application methods) on 22 of the 50 pesticides. All enrolled applicators were given a self-administered take-home questionnaire, which collected comprehensive pesticide exposure data on the other 28 pesticides; 40% of the enrolled applicators returned the take-home questionnaire. No meaningful differences in demographics, medical histories or farming practices were found between those who completed and did not complete the take-home questionnaire.11 For spouses of enrolled applicators, information on demographics, lifestyle factors, medical histories, and ever/never use of the same 50 pesticides was obtained from a self-administered questionnaire (81%) or telephone interview (19%). For this analysis, we report results for pesticides with at least 10 exposed pancreatic cases, or at least 5 organochlorine-exposed cases (because of the previous findings for this chemical group), which totaled 24 pesticides examined for ever/never exposure, and 13 pesticides for intensity-weighted lifetime exposure days.
For pesticide applicators, lifetime exposure days through enrollment were calculated as: [(years applied each pesticide) × (days applied each pesticide in an average year)]. Using lifetime exposure days and an intensity score, the intensity-weighted lifetime exposure days was calculated as: [(lifetime exposure days) × (intensity score)]. The intensity score was computed from an algorithm that took into account exposure-modifying factors such as how the pesticide was used and applied, and the protective equipment that was used.12 For spouses, only ever/never pesticide use was available.
Pancreatic cancer identification
Pancreatic cancer (ICD-10 = C25.0–C25.9) cases were identified using population-based state cancer registries. Incident pancreatic cancer cases diagnosed between enrollment and December 31, 2004 (over 9 years of follow-up time) were included in the analysis. Participants with any type of prevalent cancer at enrollment were excluded from the analysis. Vital status was obtained from the state death registries and the National Death Index. Participants who moved out of North Carolina or Iowa were not followed for cancer occurrence after moving.
This analysis included cases with incident primary carcinoma of the pancreas (93 cases: 64 applicators, 29 spouses), and controls without a history of cancer (52,721 applicators, 29,782 spouses). All but 1 of the 93 cases was exocrine tumors, with 47% located in the head of the pancreas. Pancreatic cancer risk in relation to selected characteristics and pesticide use was estimated by unconditional logistic regression controlling for age group (<50, 50–59, 60–69, ≥70), cigarette smoking (never, past, current), diabetes (no, yes), and subject type for ever/never pesticide exposure (applicator, spouse). Pesticide use was assessed using ever/never use for applicators and spouses and intensity-weighted lifetime exposure days for applicators. Intensity-weighted lifetime exposure days was categorized into never use, low use (<median days) and high use (≥median days) using the median value of each pesticide among the controls as the cutoff between low and high use. A test of trend was calculated using an ordinal variable for never, low, and high use. Family history of pancreatic cancer could not be considered a potential confounder because only 10 cases reported a family history of “other cancer,” which may or may not have included pancreatic cancers. SAS version 9.1 (Cary, NC) and the AHS data release PIREL0612 were used to conduct all analyses.
As shown in Table I, age was positively associated with pancreatic cancer among applicators and spouses. Also, compared with applicators who never smoked cigarettes, past smokers had a 1.5-fold (95% CI = 0.8–2.8) risk and current smokers had a 3.3-fold (95% CI = 1.7–6.4) risk. Applicators with diabetes had a 3.3-fold (95% CI = 1.6–6.9) risk when compared with non-diabetics. Among spouses, cigarette smoking, diabetes, and obesity (≥30 kg/m2) were associated with excess, although not statistically significant, risks of pancreatic cancer.
Table I. Odds Ratios and 95% Confidence Intervals for Pancreatic Cancer in Relation to Selected Characteristics among Applicators and Spouses in the Agricultural Health Study, 1993–2004
|Age at interview|
| <50||33,462||(63.4)||10||(15.6)||1.0||–||18,365||(61.7)||4||(13.8)|| || |
|p-trend|| || || || || ||<0.0001|| || || || || ||<0.0001|
| High school or less||28,479||(56.5)||45||(72.6)||1.0||–||11,737||(45.3)||17||(70.8)||1.0||–|
| Beyond high school||21,927||(43.5)||17||(27.4)||0.8||0.4–1.4||14,151||(54.7)||7||(29.2)||0.5||0.2–1.3|
|State of residence|
| North Carolina||18,577||(35.2)||31||(48.4)||0.9||0.6–1.6||9,775||(32.8)||11||(37.9)||0.9||0.4–1.9|
|p-trend|| || || || || ||0.001|| || || || || ||0.28|
|Other tobacco use|
| Past or current||22,981||(43.6)||34||(53.1)||1.0||0.6–1.7||3,348||(11.2)||4||(13.8)||1.3||0.4–3.8|
|Body mass index (kg/m2)|
|p-trend|| || || || || ||0.94|| || || || || ||0.34|
Applicators and spouses had similar risk factors for pancreatic cancer, therefore the risk estimates for ever/never pesticide use are shown for both combined (Table II). Associations for pesticides with at least 10 exposed cases, or at least 5 organochlorine-exposed cases (because of the previous findings for this chemical group), are reported. Among the 24 pesticides examined for ever/never use, ever use of 5 (pendimethalin, trifluralin, EPTC, chlorimuron-ethyl, and heptachlor) were associated with excess risks of pancreatic cancer adjusting for age group, cigarette smoking, diabetes and subject type, with ORs greater than or equal to 1.4, although the associations were not statistically significant. Ever use of DDT and malathion were significantly inversely associated with pancreatic cancer risk. These results were not considerably changed when the associations were examined only among applicators.
Table II. Odds Ratios and 95% Confidence Intervals for Pancreatic Cancer in Relation to Ever/Never Pesticide Exposure1 among Applicators and Spouses in the Agricultural Health Study, 1993–2004
| Permethrin used on crops and animals||69,578||12,925||83||10||0.9||0. 5–1.8|
Associations between intensity-weighted lifetime days of pesticide use and pancreatic cancer risk among applicators are shown in Table III for pesticides with at least 10 exposed cases and 5 cases per exposure group. Of the 13 pesticides examined for intensity-weighted lifetime days of exposure, 2 herbicides, pendimethalin and EPTC, had statistically significant exposure-response associations with pancreatic cancer adjusting for age group, cigarette smoking, and diabetes. For pendimethalin, low users had a 1.4-fold risk (95% CCI = 0.5–3.9) and high users had a 3.0-fold (95% CI 1.3–7.2) risk (p-trend = 0.01) when compared with never users. For EPTC, low users had a 1.8-fold risk (95% CI = 0.7–4.3) and high users had a 2.5-fold (95% CI = 1.1–5.4) risk (p-trend = 0.01) compared with never users. The risk estimates and exposure-response association between intensity-weighted lifetime days of pendimethalin and pancreatic cancer were not measurably changed after adjusting for ever/never use of EPTC (low pendimethalin use: OR = 1.4, 95% CI = 0.5–3.9; high pendimethalin use: OR = 2.6, 95% CI = 1.0–6.5; p-trend = 0.03). In contrast, the association between intensity-weighted lifetime days of EPTC and pancreatic cancer after adjusting for ever/never use of pendimethalin was attenuated and no longer statistically significant; however, this model was limited to the subset who completed the take-home questionnaire (40% of the applicators) as data for pendimethalin was only available from the take-home questionnaire. Restricting to participants who completed the take-home questionnaire (5 EPTC exposed cases), the association between EPTC and pancreatic cancer (not adjusting for pendimethalin) was attenuated and not statistically significant (data not shown). This suggests the attenuation may not be due to confounding by pendimethalin, but rather the reduced number of cases. To further explore possible confounding by multiple pesticide use, the following additional analyses were conducted. Among applicators in the AHS cohort, pendimethalin and EPTC were not highly correlated to each other (r = 0.19) or to the other pesticides examined for intensity-weighted lifetime days of exposure (r < 0.36). Also, we considered the lifetime use of all pesticides among applicators, and found no considerable change in the associations for pendimethalin and EPTC (less than 5% change in risk estimates (data not shown)), suggesting that confounding because of multiple use of more than 1 pesticide was not likely.
Table III. Odds Ratios and 95% Confidence Intervals for Pancreatic Cancer in Relation to Intensity-Weighted Pesticide Exposure1 among Applicators in the Agricultural Health Study, 1993–2004
|p-trend|| || || ||0.01|
|p-trend|| || || ||0.70|
|p-trend|| || || ||0.01|
|p-trend|| || || ||0.72|
|p-trend|| || || ||0.12|
|p-trend|| || || ||0.80|
|p-trend|| || || ||0.28|
|p-trend|| || || ||0.34|
|p-trend|| || || ||0.87|
|p-trend|| || || ||0.85|
|p-trend|| || || ||0.32|
|p-trend|| || || ||0.09|
|p-trend|| || || ||0.38|
In this case-control analysis of pancreatic cancer in the AHS, we found statistically significant exposure-response associations for 2 herbicides, pendimethalin and EPTC. Our finding of an excess pancreatic cancer risk for these 2 herbicides, but not for insecticides, is consistent with the results from a case-control study of pesticide exposed workers that found an excess pancreatic cancer risk with herbicide, but not insecticide, use.6 Pendimethalin (N-[1-ethylpropyl]-2,6-dinitro-3,4-xylidine) is a widely used dintroaniline herbicide, classified as a Group C possible human carcinogen by the United States Environmental Protection Agency (U.S. EPA).13 It has been associated with an increased risk of lung14 and rectal cancers15 among applicators in the AHS, and with thyroid tumors in rodents.16 The thiocarbamate EPTC (S-ethyl-N,N-dipropylthiocarbamate) is a Group E non-likely carcinogen with moderate acute toxicity based on the U.S. EPA classification.13 It has been linked with an increased risk of colon17 and prostate cancer18 in the AHS, as well as NHL among farmers using the herbicide for less than 7 years.19 Because of the small number of pancreatic cancer cases, previous studies of pendimethalin15 and EPTC18 in the AHS did not examine pancreatic cancer risk.
The biological mechanisms by which pendimethalin or EPTC may be linked with pancreatic cancer have not been described. Pendimethalin contains N-nitrso-compounds or nitrosamine impurities,20 and EPTC has been classified as a nitrosatable pesticide (able to form N-nitroso compounds in reaction with nitrite).21 Nitrosamines are potent animal carcinogens and are considered suspected human carcinogens.22 Nitrosamine compounds, such as those found in tobacco products, have been implicated as significant causes of cancer, including pancreatic cancer.23, 24 Although regulations were implemented by the US EPA to reduce nitrosamine contamination in pesticides in the 1980s,20 it may be biologically plausible that moderate nitrosamine exposure in pesticides before the US EPA reduction and subsequent lower levels of nitrosamine exposure may have a carcinogenic effect in regularly exposed individuals. To note, 3 other pesticides, trifluralin, atrazine, and dicamaba, which are also nitrosatable pesticides, were not associated with pancreatic cancer. Further investigation of the mechanisms of individual pesticides is warranted.
In contrast to previous findings, we found no association between organochlorines and pancreatic cancer, as well an inverse association for DDT. It is possible that the inverse association for DDT may in part be due to a healthy survivor effect among those who had used DDT before it was banned in the 1970's. The case-control study of pancreatic cancer conducted by Garabrant et al. that found statistically significant effects of DDT, DDD, and ethylan7 was conducted among chemical manufacturing workers who likely had higher levels of exposure to the pure product than those who used or applied these pesticides. Furthermore, the U.S. population-based case-control study that reported higher median serum concentration of organochlorine compounds among pancreatic cancer cases when compared with controls,10 also found that the association with DDE was attenuated after controlling for polychlorinated biphenyls (PCBs), suggesting possible confounding by PCBs. The hospital-based case-control study conducted in Spain that observed an exposure-response association between serum concentrations of DDT and pancreatic cancer risk was only statistically significant for those cases with a K-ras mutation.9 These previous findings suggest that organochlorines may be associated with pancreatic cancer, but only among certain populations or conditions. It is unlikely that pendimethalin or EPTC are surrogates for DDT as neither is highly correlated with DDT in the AHS (pendimethalin r = 0.14, EPTC r = 0.03).
Several strengths of this study should be mentioned. Selection bias should be minimal because of the high cohort recruitment (82% of applicators, 75% of spouses) and the low loss to follow-up (less than 2%). Misclassification of pancreatic cancer diagnosis is unlikely given the data linkage to population-based state cancer registries. Pesticide use data, although self-reported, has been shown to be reliable,25 and because of the prospective study design, any misclassification of pesticide use should be nondifferential and therefore unlikely to create false-positive associations. Also, we were able to control for several important risk factors including, age, cigarette smoking, and diabetes.
A few limitations should also be noted. We were unable to evaluate certain pesticides and did not have sufficient statistical power to estimate statistical interactions due to the limited number of exposed cases, however, it is important to note that these cases were drawn from one of the largest prospective cohorts of pesticide exposed subjects, thus no other study would have a larger sample size. Although the response rate to the take-home questionnaire was 40%, an evaluation of the potential biases related to this response rate concluded that there was little meaningful opportunity for confounding.11 Our findings may have limited generalizability due to the predominantly white male study population. Because we examined several pesticides with biological effects in humans that are unclear, and are the first to report associations for pendimethalin and EPTC with pancreatic cancer, these findings should be considered hypothesis generating and in need of confirmation.
In conclusion, we found statistically significant exposure-response associations for pendimethalin and EPTC on pancreatic cancer risk among pesticide applicators in the AHS cohort. Pendimethalin and EPTC are able to form N-nitroso-compounds, thus, our findings are consistent with evidence suggesting a carcinogenic effect of nitrosamines on the pancreas. Future studies with a larger number of pancreatic cancer cases are needed to confirm our findings and to evaluate the effect of other nitrosamine carrying pesticides.
We thank the Agricultural Health Study participants for their participation, and the field station and coordinating centre staff for their efforts.