After a systematic retrieval and screening of the literature on the relationships between parental occupational pesticide exposure and childhood leukemia, we evaluated the overall evidence using a quantitative meta-analytic approach. Childhood leukemia was not associated with any paternal occupational pesticide exposure in our analyses of all relevant studies or in subgroups of studies with median or higher total quality or bias control scores, well-defined or reasonably inferable preconceptual exposure windows, exposure information collected before offspring leukemia diagnosis, farm-related exposure, data for the major leukemia subtypes, population-based case–control or cohort design, or a publication date of 1990 or later. Childhood leukemia was associated with paternal occupational exposure to insecticides and herbicides, but none of the few relevant studies assessed exposure–risk gradients.
Childhood leukemia was associated with prenatal maternal occupational pesticide exposure with no evidence of statistically significant heterogeneity or publication bias. The association was somewhat stronger among studies with higher exposure measurement quality scores and those with farm-related pesticide exposure. Summary ORs were similar for studies of ALL and AML and for pre-1990 or more recent studies. There were moderately strong associations between childhood leukemia and prenatal maternal occupational exposure to insecticides or herbicides based on the few available studies. All of the eligible studies of prenatal maternal occupational pesticide exposure were based on information collected after offspring leukemia diagnosis. There were too few relevant studies for meaningful analyses of maternal occupational exposure to fungicides or for exposure of either parent to individual pesticides.
Interpretation of our meta-analyses is constrained by limitations in the original studies, particularly exposure assessment and potential sources of bias. We attempted to address these issues by conducting a comprehensive literature search (to reduce publication bias) and independent data extraction and study quality assessment by two persons. We also conducted meta-analyses stratified by parent exposed, study quality scores (total and major components), exposure window definition, leukemia subtype, exposure index, farm versus other workplace exposure, study design, publication period, and broad pesticide class. We assessed study quality with a modified Downs and Black tool (
Downs and Black 1998) [see Supplemental Material, Appendix 3 (doi:10.1289/ehp.0900582.S1)]. The main limitations incurred during quality assessment were incomplete descriptions of study methods and findings in reports of original research, lack of a direct method to assess recall bias, and the largely unknown etiology of childhood leukemia, reducing our ability to assess the control of potential confounders. Studies with median or greater quality scores generally had better exposure assessment and control of potential sources of bias compared with lower ranking studies. Few eligible studies collected exposure information for specific or toxicologically related pesticides. Only three studies collected and assessed exposure frequency or intensity information, and little is known about the etiology of childhood leukemia, apart from ionizing radiation. Accordingly, our results should be interpreted cautiously.
Potential sources of bias in observational epidemiologic studies are well described elsewhere (
Rothman and Greenland 1998). Although case–control studies dependent on parental recall of potentially hazardous exposures may be subject to recall bias, nondifferential misclassification of exposure status may be a bigger problem (
Infante-Rivard and Jacques 2000). The latter source of bias tends to reduce the chance of detecting a true association between a potential causal factor and an adverse health outcome. In a large case–control study, pesticide exposure was the only self-reported paternal occupational exposure associated with childhood AML; this association persisted when paternal occupational pesticide exposure was inferred from a job–exposure matrix (
Buckley et al. 1989). Such findings argue against a major bias arising from self-reported occupational pesticide exposure information. In controlled biomoni-toring field studies of farmers, self-reported pesticide exposure information was a fairly good predictor of body burden, if subjects noncompliant for urine collection or reporting incomplete or inconsistent pesticide use information were removed from analysis (
Scher et al. 2008). However, other controlled field studies of farm children and farmers revealed poor correlations between biomonitoring and self-reported pesticide exposure data (
Arbuckle et al. 2004;
Perry et al. 2006).
Previous reviewers concluded that there were fairly consistent associations between childhood leukemia and parental occupational or residential pesticide exposure (
Daniels et al. 1997;
Zahm and Ward 1998). Recent reviewers noted that associations were strongest for parental pesticide exposure before and during pregnancy and for childhood exposure to household insecticides (
Buffler et al. 2005;
Infante-Rivard and Weichenthal 2007) and that prenatal maternal pesticide exposure may be more important than paternal exposure (
Brown 2006). A recent meta-analysis of seven case–control studies of adult leukemia and occupational pesticide exposure published during 1990–2005 showed a summary OR of 1.35 (95% CI, 0.91–2.0) (
Merhi et al. 2007). A meta-analysis of 17 studies of adult myeloid leukemia and occupational pesticide exposure published during 1979–2005 revealed a slightly elevated risk (summary OR = 1.21; 95% CI, 0.99–1.48) (
Van Maele-Fabry et al. 2007); their subgroup analyses showed stronger associations in the five studies of pesticide applicators (summary OR = 2.14; 95% CI, 1.39–3.31) and the two studies of manufacturing workers (summary OR = 6.32; 95% CI, 1.90–21.0). Although these studies suggest a role for pesticides in adult leukemia, their relevance to childhood leukemia is not clear because the mechanisms may differ.
Childhood leukemia is associated with genetic polymorphisms in genes encoding enzymes or other proteins involved in DNA repair, membrane transport, cell cycle regulation, and phase I and II metabolism of chemical toxicants (
Kim et al. 2006). As noted in a recent review (
Anderson 2008), associations between childhood hematopoietic cancers and genetic polymorphisms in genes encoding phase I and II enzymes are consistent with potential chemical causes of these cancers. For instance, a large Quebec case-only analysis reported relatively large interaction ORs between childhood leukemia and
CYP1A1m1 and
CYP1a1m2 variants and prenatal maternal or childhood pesticide exposure (
Infante-Rivard et al. 1999).
Most childhood leukemia cases have gross chromosomal abnormalities, including translocations caused by faulty repair of double-strand DNA breaks. Double-stranded DNA breaks may be caused directly by ionizing radiation and certain mutagenic chemicals or indirectly by modulation of type II topoisomerase enzymes. Analysis of routinely collected neonatal blood samples revealed leukemia clones with specific chromosomal translocations in children who later developed ALL, suggesting that many such cases originate
in utero (
Gale et al. 1997). About half of all childhood leukemia cases occur by 3 years of age, and most cases probably have a clonal origin, developing from a single abnormal precursor cell over a period of several months (
Ford et al. 1998;
Ma et al. 1999;
Mori et al. 2002;
Taub and Ge 2004). In a small study of infants born in an agricultural region with high pesticide use in the Philippines, the prevalence of the t(8;21) translocation in cord blood samples was 20.5% among those with detectable meconium levels of the methylcarbamate insecticide propoxur, compared with 10% among infants with undetectable levels (crude OR = 2.32; 95% CI, 0.30–57.4; calculated from data given in the report) (
Lafiura et al. 2007). It appears that preleukemic clones can persist during childhood and that only a minority progress to leukemia, suggesting that postnatal exposures could influence progression (
Maia et al. 2004).
The biological plausibility of potential causal relationships between cancer and pesticide exposure is supported by reviews of available evidence, mainly from animal studies. The U.S. Environmental Protection Agency (EPA) and other national and international bodies have identified about 165 pesticidal active ingredients as known, probable, or possible human carcinogens, some of which have been banned or restricted (
Goldman 1998). The 15 most intensely used pesticides in the United States during 2001, based on the amount of active ingredient sold (
U.S. EPA 2004), include three probable human carcinogens (alachlor, metam sodium, and chlorothalonil) and five possible human carcinogens (acetochlor, malathion, metolachlor, pendimethalin, and trifluralin) (
U.S. EPA 2007). Among 60 pesticides still used in Canada but banned in one or more Organisation for Economic Co-operation and Development member countries because of health and environmental concerns (
David Suzuki Foundation 2006), the insecticides carbaryl and propoxur and the fungicides captan, mancozeb, maneb, and metiram are recognized as probable human carcinogens (
U.S. EPA 2007).
In experimental animals, exposure of pregnant females to carcinogens can produce cancer in offspring (
Autrup 1993). Lymphomas in mice were induced by transplacental exposure to the fungicides carbendazim or dodecylquanidine acetate together with sodium nitrite (
Borzsonyi et al. 1976,
1978). Among men, lymphocyte or sperm DNA damage detectable using the comet assay has been associated with background exposure to chlorpyrifos or carbaryl (
Meeker et al. 2004), with occupational exposure to carbofuran (
Zeljezic et al. 2007) or multiple pesticides (
Liu et al. 2006), and with occupations in pesticide production (
Bhalli et al. 2006) and farming (
Naravaneni and Jamil 2007). Male mice preconceptually exposed to ionizing radiation had increased sperm DNA strand breaks, and their offspring demonstrated an increased risk of hematopoietic cancers (
Hoyes et al. 2001). These studies suggest potential mechanisms for relationships between childhood hematopoietic cancers and prenatal maternal or preconceptual paternal pesticide exposures.
This article has been 
