Many pesticides are carcinogenic to animals, and some are considered carcinogenic to humans with varied degree of evidence. For example, the U.S. Environmental Protection Agency has classified chlordane, heptachlor, tetrachlorvinphos, carbaryl, and propoxur as probable or likely human carcinogens, and lindane, dichlorvos, phosmet, and permethrin as suggestive or possible carcinogens (U.S. Environmental Protection Agency 2003
). Maternal and cord blood levels of some pesticides are similar, demonstrating that they are readily transferred from mother to fetus during pregnancy (Whyatt et al. 2003
). Parental exposures may act before the child’s conception, during gestation, or after birth to increase the risk of cancer. Before conception, exposures may cause mutations or epigenetic alterations in gene expression, such as genomic imprinting or DNA methylation, in the sperm or egg (Anderson et al. 2000
). Exposure after conception (i.e., during the pregnancy or after birth) may cause somatic cell mutations or alterations in hormonal or immunologic function (Daniels et al. 1997
) that affect cancer risk (Anderson et al. 2000
However, potential effects of pesticides on risk for childhood cancers are not clearly understood. Several epidemiologic studies reported risk of childhood brain cancer associated with residential pesticide use with mixed results. Pesticide exposure during pregnancy was not a risk factor for childhood brain cancer in some studies (Bunin et al. 1994
; Kuijten et al. 1990
; Leiss and Savitz 1995
), whereas others reported at least one type of pesticide associated with an increased risk (Pogoda and Preston-Martin 1997
; Preston-Martin et al. 1982
; Wilkins and Bunn 1997
). In one study, more case mothers reported using pesticides, but case and control mothers were similar with respect to whether their homes were treated by an exterminator at any time during pregnancy (Preston-Martin et al. 1982
). In another study, bombs or no-pest strips used for nuisance pests during pregnancy were associated with a significant increase in childhood brain cancer risk, but insecticide or herbicide use was not (Wilkins and Bunn 1997
The results of this study must be viewed in light of several possible limitations. First, selection bias can make the interpretation of case–control studies difficult. This is especially true in hospital-based case–control studies, in which it is difficult to identify the source population from which the cases were derived. In our population-based study, this issue appears less of concern because cases were identified by statewide cancer registries and controls were selected from the general population by matching the state of residence accordingly. Nevertheless, control selection on the basis of RDD can lead to selection bias because of potential incomplete phone coverage, residences with multiple phone lines, and nonresponse (Wacholder et al. 1992
). It is also possible that this study may have excluded families with high exposures to pesticides, such as migrant farm workers, because of their language barriers. If there truly was an association between pesticides and childhood brain cancer, this would have excluded cases in greater proportion than controls, resulting in selection bias. However, in our study the proportion of potential study participants who were excluded because of language barrier (any language) was relatively small: about 1% of the located cases and about 2% of the potential controls identified through working residential telephone numbers.
Furthermore, recall bias is always of concern in population-based case–control studies of childhood cancer in which parents of case children are more likely to accurately report (or overreport) specific exposures potentially associated with disease compared with parents of healthy children. Moreover, they may report more detailed job histories than do parents of control children (van Wijngaarden et al. 2003
). Nevertheless, we primarily relied on standard job and industry titles in our exposure assessment, and we believe differential reporting is unlikely. Additionally, misclassification of exposure certainly occurred, which most likely yielded a conservative bias in the ORs. However, in our analysis we excluded 208 fathers (104 pairs) because either their job information was missing or their matched case or control counterparts had missing job information. This exclusion might have introduced bias, the direction of which is difficult to predict.
Interpretation of our findings of significant risk associated with the father’s herbicide application for lawns and gardens requires caution. In examining the risk by who applied the pesticides, we could not analyze the data after excluding those households that reported the applications by both professionals and parents because the numbers were small. Pesticides applied by professionals may have been more toxic (e.g., pesticides requiring “restricted use”) than those used by parents. Further, we relied on maternal report of residential pesticide use, and the mother’s recall on the father’s application could have been inaccurate. However, although data were limited, we found that ORs were higher among fathers who never or occasionally washed immediately afterward or wore protective clothing, compared with those who always or usually took such precautions.
Two raters evaluated all jobs mothers and fathers held during the 2 years before birth, and the interrater agreement was fair to moderate (κ = 0.3–0.6) for fathers and poor to fair (κ = 0.02–0.3) for mothers, according to a previously proposed categorization of κ coefficients (Sim and Wright 2005
). The interrater agreement for fathers is similar to estimates reported elsewhere (van Wijngaarden et al. 2003
), although in the present study paternal work history was provided by mothers and subject to possible inaccuracies. The low inter-rater agreement for mother’s job exposures may reflect the lack of research focused on occupational pesticide exposure among jobs more commonly held by women. The main source of discrepancy between raters was the large number of women who were considered minimally exposed by one rater and unexposed by the other rater, especially for clerical and retail jobs. Given the limited work history data available for mothers in this data set, it is unclear which assignment is more accurate, although a recent report of cases of pesticide poisoning in the retail industry suggests that pesticide exposure in these environments may be likely (Calvert et al. 2007
In conclusion, these data provide some evidence for an association between brain cancer risk in children and paternal exposure to pesticides during the 2 years before birth, in particular for astrocytoma and herbicide exposure. Our findings are consistent with those reported by van Wijngaarden et al. (2003)
and Kuijten et al. (1992)
, although they appear to contradict results published by Bunin et al. (1994)
and Kristensen et al. (1996)
, which showed stronger associations of pesticides with PNET rather than with astrocytoma. Although several suggestions regarding potential biological mechanisms have been made (Olshan and van Wijngaarden 2003
), it is currently unclear whether they are more relevant to astrocytoma or PNET.
Future epidemiologic studies investigating environmental risk factors of childhood brain cancer could benefit from close collaboration with other scientific disciplines. Developing biomarkers—both of exposure and of early health effects—that can be measured reliably should help future studies. In addition, future studies should consider examining potential gene–environment interactions, as the candidate genes involved in the chemical metabolism become known, because the metabolism of environmental chemicals may vary between individuals because of genetic polymorphisms.