In this study, we did not observe an association between metribuzin use and overall cancer incidence. While we found a statistically significant association between leukemia and metribuzin in the highest tertile of intensity-weighted days which was accompanied by a significant linear trend test, these statistically significant findings were not consistent across exposure metrics and reference groups. We also found a significant trend test for the lymphohematopoietic cancers across tertiles of intensity-weighted exposure days using the low-exposed referent group, but the test only approached significance using the non-exposed referent. These differences may result from a smaller sample size when the low exposed group is used as the referent or may indicated residual confounding. Significant point estimates and tests for linear trend suggest that an association may exist between metribuzin and the increased risk of leukemia and all lymphohematopoietic cancers. In the absence of an a priori hypothesis or supporting epidemiologic, mutagenic, or genotoxic evidence, these positive results should be interpreted with caution. If these findings are confirmed, the relationship may suggest carcinogenesis through a non-genotoxic mechanism.
Our results indicate that there is not an overall increased risk of cancer in pesticide applicators enrolled in the Agricultural Health Study exposed to metribuzin. This is the first study to our knowledge that has examined the association between metribuzin exposure and the incidence of a variety of cancer sites in a human population.
The first epidemiologic study that examined the possible association between metribuzin and cancer did not find an association between metribuzin and NHL (13
). This analysis was limited to an ever/never use categorization and a single cancer, and the retrospective design was not able to ensure that the exposures preceded the onset of disease. Because this was the only prior study exploring metribuzin exposure and cancer in humans, we included NHL in our analysis although we had a small number of cases. Our results for NHL using the non-exposed referent group were similar to those in the previous study. Point estimates for the medium and high exposure tertiles of both exposure metrics were elevated above the null for NHL using the low-exposed referent group; however, there was a significant inverse relationship in the lowest tertile of exposure compared to the non-exposed. Similar to the previous study which included 20 cases who reported ever being exposed to metribuzin (13
), our results were based on small numbers of exposed cases (n=19), making interpretation more difficult.
The toxicologic evidence for metribuzin carcinogenicity is mixed. There are several negative tests for genotoxicity and mutagenicity (5
); however, other in vitro
genotoxicity and mutagenicity studies have yielded positive results (10
). It has been noted that herbicides in general do not directly interact with DNA, but may be able to induce tumors through other mechanisms which include promotion of spontaneous initiation, cytotoxicity, inhibition of apoptosis, oxidative stress, and formation of activated receptors (19
). The potential for metribuzin to act as an epigenetic carcinogen combined with the presence of positive genotoxicity and mutagenicity tests in vitro
and suggestive positive epidemiologic findings for leukemia warrants further exploration of possible carcinogenic effects of metribuzin exposure.
The Agricultural Health Study is the largest study of pesticide applicators exposed to metribuzin to date. A considerable strength of this study is the prospective cohort design. This design allowed the assessment of all pesticide and covariate exposure information prior to cancer diagnosis, which minimized recall bias in cases (15
). The lifetime days metric provides an improvement over previous ever/never analysis of metribuzin use in humans. Use of this metric allowed us not only to quantify duration and frequency of exposure, but also allowed us to test for exposure-response effects. Additionally, we used a published exposure metric that was able to account for differences in intensity of exposure in addition to duration and frequency (17
). A comparison of the results of our analyses shows that the risk estimates were not meaningfully different when using the intensity-weighted metric as opposed to total lifetime days of exposure.
There are some limitations of this study. First, the investigation of certain cancers was hindered by small numbers of exposed cases. Additional follow-up in the AHS will provide more cases that will allow for more powerful analyses of many cancer sites. Second, pesticide applicators are frequently exposed to a variety of chemical, physical, and biological agents in agricultural work, which may potentially confound an exposure-disease relationship in regards to pesticides and cancer. There is the potential for confounding due to exposures that result from farm-related activities, but investigators have determined that substantial bias in this cohort due to confounding from exposures resulting from performing other farming activities is not likely (20
). Since 47% of the applicators completed the take-home questionnaire, there is the possibility for selection bias. However, participants who completed the take-home questionnaire were generally similar to the full cohort with the exception that responders were slightly older than non-responders, which may lead to higher farm-related exposures in responders (16
). Also, since the full cohort provided information on ever use of metribuzin, we were able to compare the results of these analyses between the full cohort and the sub-group who completed the take-home questionnaire and found the results to be similar. Therefore, we believe selection bias is unlikely in this analysis. Exposure misclassification is another potential limitation of this study although this study provides more detailed information on usage of metribuzin than any other study of this chemical. The intensity-weighting algorithm took into account many factors that may affect the level of exposure, such as the use of personal protective equipment (PPE) and application practices. This information is not specific to metribuzin, but the intensity-weighting algorithm has been shown to be reasonably valid when comparing scores to the concentration of urine metabolite concentrations (18
). In a prospective study such as this, it is almost certain that exposure misclassification is non-differential and would attenuate estimates of association. Finally, we included only men in this analysis so if use of metribuzin affects women’s cancer risk differently, the results may not be generalizable.
This study is the only prospective cohort study to examine the effects of metribuzin exposure and the risk of cancer in humans. Although the interpretation of the results from this study is limited at this time because of the paucity of cases for certain cancers, our results suggest that there is little or no association between metribuzin exposure and the incidence of all cancers combined. The results from this study suggest a potential association between metribuzin use and certain lymphohematopoietic malignancies in men; however, having not been observed previously caution should be used in interpretation. Additional follow-up in the Agricultural Health Study will provide more information on these risks as more cases arise.