Secondary Logo

Journal Logo

Agricultural Pesticide Use and Childhood Cancer in California

Reynolds, Peggy*; Von Behren, Julie*; Gunier, Robert B.*; Goldberg, Debbie E.; Harnly, Martha*; Hertz, Andrew

doi: 10.1097/01.ede.0000147119.32704.5c
Original Article
Free

Background: Household pesticide use has been associated with higher risk for several childhood malignancies, but few studies have evaluated risks associated with residential proximity to agricultural pesticide use. We conducted a population-based case-control study of early childhood cancer (age 0–4 years) among California children born between 1990 and 1997 and mother's residential proximity to agricultural applications of pesticides at the time of the child's birth.

Methods: Included in the study were 2189 case children and 4335 controls matched for birth date and sex. We estimated the in utero exposure potential from specific chemicals and chemical groups used in the 9 months before birth within a half mile of the maternal residence. We computed odds ratios (ORs) using conditional logistic regression.

Results: No striking patterns emerged. There were modestly elevated ORs for leukemias associated with probable and possible carcinogen use, and with nearby agricultural applications of organochlorines and organophosphates during pregnancy. Two commonly used pesticides were associated with higher leukemia risk when comparing the highest and lowest categories: metam sodium (OR = 2.05; 95% confidence interval = 1.01–4.17) and dicofol (1.83; 1.05–3.22).

Conclusions: The few elevated risk associations in this study are consistent with chance, given the large number of comparisons, but they may deserve more careful consideration. Future studies that integrate specific temporal and spatial exposure potential for targeted pesticides will be important in further evaluating risks associated with childhood cancer.

From the *California Department of Health Services (CDHS), Environmental Health Investigations Branch, Oakland, California; the †Public Health Institute (PHI), Oakland, California; and ‡Impact Assessment, Inc., Oakland, California.

Submitted 21 August 2003; final version accepted 21 September 2004.

Funding provided by National Cancer Institute Grant # R01 CA71745.

Correspondence: Peggy Reynolds, California Department of Health Services, Environmental Health Investigations Branch, 1515 Clay Street, Suite 1700, Oakland, CA 94612. E-mail: preynold@dhs.ca.gov.

Many interview-based, case-control studies have found an increased risk of childhood cancer with household pesticide use and parental occupational exposure to pesticides.1,2 Agricultural pesticide use is another potential exposure source for those living near treated fields,3–5 although such risk associations have seldom been evaluated in epidemiologic studies. A large ecologic study in California found that childhood cancer rates did not substantially differ between children living in neighborhoods with high agricultural pesticide use and those living in neighborhoods with little or no use.6 Conversely, a large records-based study in Norway noted that offspring of farm residents engaged in horticulture that involved pesticide use experienced higher rates of cancer at very young ages.7 The Norwegian study focused on exposure potential in a more specific time window and on a smaller geographic scale than the California study, underscoring the importance of these factors in assessing potential risk relationships.

Much of the current childhood cancer literature has highlighted the perinatal period as an etiologically important window for exposure to environmental toxicants. In particular, a Northern California case-control study of childhood leukemia implicated household insecticide use with the highest risk estimates for maternal exposure during pregnancy.8

In our case-control study of childhood cancer, we linked cancer registry, birth certificate, and agricultural pesticide use data to focus on risk associations during the prenatal period for specific pesticides used within a short distance from the maternal residence at birth.

Back to Top | Article Outline

METHODS

Selection of Cases and Controls

We obtained information on childhood cancer cases from California's statewide population-based cancer registry. To characterize exposure potential for the 9-month period before birth, we selected all childhood cancer cases diagnosed under the age of 5 between 1990 and 1997 among children born between October 1990 and December 1997. Using probabilistic record linkage,9 we matched 2216 (85%) of the 2612 cases to their California birth certificate. For each case, we randomly selected 2 control birth certificates, matched on date of birth and sex, from a roster of all concurrent statewide births among mothers who were California residents at the time of birth. We then matched the controls’ birth certificates to the California Birth Cohort files, which contain linked birth and death records for infants in the first year of life. If a control died in the first year of life and the death occurred at an age younger than that of the corresponding case's age at diagnosis, this control was excluded and replaced by the next eligible live birth certificate in the file (again, matched for birth date and sex).

We abstracted mothers’ residential addresses at the time of birth from each birth certificate and used a geographic information system (GIS) to assign each address to a latitude and longitude coordinate and to a U.S. Census block group. A block group is an aggregation of census data below the census tract level intended to capture approximately 1000 persons.10 We successfully geocoded 2189 (98.8%) population-based cases and 4388 (99%) controls. For the final analysis, we excluded an additional 53 controls because their corresponding cases’ addresses were not geocodable, resulting in a total of 4335 controls. Based on the block group of mother's residence at the time of each child's birth, we also assigned neighborhood attributes for median family income and urbanization from the U.S. Census.11

Back to Top | Article Outline

Pesticide Data

Since 1990, California's Department of Pesticide Regulation has maintained the Pesticide Use Report database to track all statewide commercial agricultural pesticide use. For each application, the pesticide data provide detailed information on the active ingredient, quantity applied, acres treated, date applied, and location (in square-mile sections). We used Pesticide Use Report data from 1990 through 1997 for this study.12

Because over 850 different pesticides were used during the study period, we generated a priori assignments of these agents into 4 toxicologic groups (probable carcinogens, possible carcinogens, genotoxic compounds, and reproductive or developmental toxicants) and 4 chemical classes (organochlorines, organophosphates, carbamates, and dithiocarbamates). California banned or severely restricted all pesticides classified as known human carcinogens before our study. Probable and possible carcinogens are determined almost exclusively from laboratory animal studies.13 We chose those genotoxic chemicals (chemicals that directly damage DNA) with at least 2 positive results in genetic toxicity assays.14 We selected reproductive and developmental toxicants for analysis based on studies conducted in laboratory animals.15 We also generated a prioritized list of individual pesticides for analysis based on statewide use, carcinogenic potency, and exposure potential.16 Ultimately, we selected 7 potentially high-risk pesticides for individual evaluation: propargite, methyl bromide, trifluralin, simazine, metam sodium, dicofol, and chlorothalonil. Our procedures for classifying pesticides into these categories are discussed in greater detail elsewhere.16

Back to Top | Article Outline

Assigning Pesticide Exposure

To characterize potential in utero exposure, we defined children's exposure period as the 9 months before birth. The locations of pesticide applications in the Pesticide Use Report data are reported to sections in the Public Land Survey System, a rectangular survey system that typically divides land into 1-square-mile sections. Using a GIS (see Fig. 1), we created a half-mile buffer around each residence and intersected the buffers with the square-mile Public Land Survey System sections. We chose a half-mile radius buffer to represent the distance where maximum exposure is likely to occur based on pesticide drift studies.17–21 Next, we assigned pesticide use (in pounds) to each residence proportional to the percentage area of each section within the buffer. For each residence, we then summed the proportionally weighted pounds used during the gestation period for all mile-square sections within the buffer. Finally, we divided by the buffer area (0.79 square miles) to obtain pounds per square mile for each pesticide group and individual pesticide of interest. For a subset of pesticides, we also estimated exposure based on pesticide use in the single section that encompassed the child's residence at birth.

FIGURE 1.

FIGURE 1.

For each analysis, we defined our reference group as all subjects with less than 1 pound per square mile of pesticide use for that group or individual pesticide within the half-mile buffer. We based our other 2 use categories on the distributions of pesticide-use densities among control subjects with greater than 1 pound per square mile of use density. These categories were low exposure (1st to 49th percentiles) and high exposure (50th percentile and above). Where we observed suggestive associations, and where numbers permitted, we examined the risk for higher exposure levels (90th percentile and above).

Back to Top | Article Outline

Statistical Analysis

For our matched study design, we calculated odds ratios (ORs) and 95% confidence intervals (CIs) using conditional logistic regression, also including terms for race/ethnicity. We performed analyses for all cancer sites combined, as well as for the 2 major childhood cancer types: leukemias and central nervous system (CNS) malignancies as defined by International Classification of Childhood Cancer diagnostic groups I and III.22

Back to Top | Article Outline

RESULTS

We included 2189 childhood cancer cases and 4335 controls in our analyses. Table 1 shows the distribution by race/ethnicity and sex for each cancer group examined. Over one third of the cases were leukemias and 16% were CNS tumors. Notably, nearly half of the total cases (45%) occurred among Hispanic children.

TABLE 1

TABLE 1

The majority of birth addresses (over 90%) had nearby pesticide use of less than 1 pound per square mile for any individual pesticide. Over 70% of the birth addresses had less than 1 pound per square mile of pesticide use for the toxicologic or chemical groups. There was, however, considerable range in exposure opportunity, with a small proportion of children born in areas with very high agricultural pesticide use. Two percent of mothers had class B (probable carcinogen) pesticide-use exposure estimates of over 350 pounds per square mile within a half mile of their residence at the time of the child's birth.

Tables 2, 3, and 4 present the odds ratios estimated for all cancer sites combined, leukemias, and CNS tumors. Few risk associations were evident, although the odds ratios for leukemia were elevated for the highest exposure category for metam sodium, simazine, chlorothalonil, and dicofol. For the toxicologic or chemical groupings studied, there was no clear pattern of elevated risks; the odds ratios were modestly elevated for leukemia in areas with higher class C (possible) carcinogen use and with lower developmental/reproductive toxin and carbamate use.

TABLE 2

TABLE 2

TABLE 3

TABLE 3

TABLE 4

TABLE 4

We also examined pesticide-use relationships for the most common type of leukemia, acute lymphoid leukemia, despite the small numbers of cases with potential exposure. The odds ratios for acute lymphoid leukemia in the high-use areas were suggestively elevated for trifluralin (1.35; CI = 0.71–2.56), dicofol (1.85; 0.98–3.52), and metam sodium (3.28; 1.37–7.86). The odds ratios for acute lymphoid leukemia for the high organochlorine exposure level was 1.51 (0.85–2.69) and 1.32 (1.01–1.73) for class C carcinogens.

For pesticides for which we observed suggestive associations, classification at higher pesticide exposure levels (≥90th percentile) resulted in few subjects in the top analysis group. For metam sodium, only 7 cancer cases were in the ≥90th percentile exposure category, with an odds ratio of 1.42 (CI = 0.54–3.75). The odds ratio for leukemia for this high metam sodium-exposure category was 2.64 (0.59–11.79) and was 1.91 (0.85–4.27) for the 50th to 89th percentile category. For dicofol, based on only 9 cancer cases in the ≥90th percentile exposure category, the odds ratio for all cancer sites combined was 1.23 (0.54–2.83). For leukemia, the odds ratio for dicofol was 2.11 (1.15–3.87) for the 50th to 89th percentile category, but 0.75 (0.14–3.88) for the ≥90th percentile category. For propargite, the odds ratio for leukemia was 1.77 (0.73–4.30) for the ≥90th percentile category.

Using a more restricted pesticide-exposure measure, based only on the children's section of residence, the odds ratios for leukemias were unchanged for chlorothalonil and the class C carcinogens (data not shown). For class B carcinogens, the odds ratios increased for the low-exposure category and decreased for the high-exposure category (data not shown). The odds ratios for dicofol increased for both the low (0.75–2.09) and the high-exposure category (1.83–3.79), but there were few subjects classified as exposed at the highest level.

When we added terms for median family income and urbanization to the multivariable models, the results were unchanged (data not shown).

Back to Top | Article Outline

DISCUSSION

Our analyses revealed no clear risk patterns. For CNS tumors and for all cancer sites combined, there was little evidence of increased risk with high nearby agricultural pesticide use for any of the toxicologic groups or individual chemicals we examined. Unlike our previous ecologic study,6 which used more generalized exposure estimates, our current analyses did not show a generally increased leukemia risk for children born in areas of high propargite use, although there was some suggestion of elevated risk for exposure at and above the 90th percentile. However, we did observe elevated risks for leukemia in the highest areas of metam sodium and dicofol use, although there was not a trend of increasing risk with higher use.

We know of no other childhood cancer studies that have systematically examined specific agricultural pesticide use near the maternal place of residence during gestation. An earlier California study used a similar approach to evaluate adverse pregnancy outcomes in high pesticide use areas.23 In that study, the temporal aspects of potential pesticide exposures were important, with the observed risk relationship particularly accentuated during the critical time window of early fetal development. A recently published childhood leukemia case-control study also found that the risk estimates for household pesticide exposure varied with the timing of reported use, with indoor applications during pregnancy having the highest odds ratio.8

Unlike our study, most of the previous literature on childhood cancer and pesticides has assessed exposures based on reported home use or parental occupation.8,24–27 One recent exception is a German case-control study of childhood cancer, which did not rely on the parental occupation of “farmer” as a surrogate for agricultural pesticide exposure.28 Instead, the study asked families if they farmed at all, even part-time (which is common in that country). The study found an elevated odds ratio (1.5) for leukemia with reported use of pesticides on farms. Their data indicated that using the “farmer” occupational designation would have excluded approximately one third of the parents who actually had reported agricultural pesticide use.

Most of the evidence about childhood cancer and pesticide use comes from case-control studies that lack information on specific chemicals, thus making it difficult to pinpoint risks related to any specific pesticides or class of chemicals. Based on the detailed information about applied chemicals in our study, we found suggestive risk associations for leukemia with the insecticide dicofol as well as the fumigant metam sodium. In air, metam sodium breaks down into the active pesticide agent methyl isothiocyanate (MITC). Air-monitoring studies conducted in urban centers of California's Central Valley revealed high MITC levels, suggesting that it drifts beyond half a mile.29 A screening risk assessment using available air-monitoring data from high pesticide-use areas in California found that concentrations of MITC and other fumigants exceeded levels of concern for acute and subchronic health effects even at some distance from metam sodium applications.30 The evidence of considerable MITC drift suggests that the association between metam sodium use and leukemia risk deserves more careful consideration in future studies. Similar air-monitoring data was not available for dicofol.

Pesticide exposure estimates based only on the Public Land Survey System section of residence should be more specific, but less sensitive, than exposure estimates using all Public Land Survey System sections within a half-mile buffer. A recent assessment of potential exposure misclassification error compared the results from a “broad” definition of exposure using Pesticide Use Report data, similar to that used in our analyses, to a more refined measure using 500-meter buffers and crop maps combined with Pesticide Use Report data.31 The authors reported that for the 5 individual pesticides evaluated, the sensitivity of the “broad” definition was 100% but the specificity was only 62% to 94%, resulting in an attenuation of a true risk ratio of 1.5 to 1.1. This is generally consistent with the increased leukemia point estimates for dicofol we observed when we used pesticide exposure estimates based only on children's Public Land Survey System section of residence in our study; however, exposure opportunity for this and other chemicals is likely to be a function of many factors.

Our study has a number of limitations. Most children included in our study lived in areas with no or very low agricultural pesticide use. The odds ratios for our highest exposure categories were based on small numbers. We were limited to evaluating the proximity of agricultural pesticide applications to children's birth addresses. We lacked information on home pesticide use and parental occupational exposure to pesticides. Additionally, we assumed that the mother's address at birth was representative of potential exposure during pregnancy, although as many as 25% of California women may move during pregnancy.32 Control children could have moved out of state after their birth, whereas case children had to remain in the state to be included. Although it is possible that some controls who moved may have developed cancer, the likelihood is small. To estimate exposure, we used reported agricultural pesticide-use data and did not collect environmental or biologic samples to validate actual exposure levels. However, several studies have found that concentrations of pesticides in household dust and levels of pesticide urinary metabolites in individuals increase with increased proximity of their residence to treated agricultural fields.4,33–36 The half-mile buffer we used may not be the ideal distance for estimating potential pesticide exposure; for various agents, it may be either too large or too small. Postapplication drift from treated fields depends on many factors, including the pesticide used, application method, and meteorologic conditions.18,20,21,29,37,38

Despite these limitations, our data sources permitted an assessment of potential health risks associated with proximity to high-volume agricultural pesticide use that was more refined than previously possible. The pesticide data were from mandatory statewide reporting by growers and therefore not subject to recall bias. Recall bias is a concern in case-control studies using self-reported exposure data, as illustrated by overreporting of case parents in a recent analysis of parental occupational exposures and childhood cancer.39 The Pesticide Use Report data also provided us with specific chemicals, amounts, and dates applied. As a result, we were able to summarize pesticide use for groups and individual pesticides of interest during the 9-month window corresponding to our subjects’ approximate gestation periods. California's statewide cancer registry provided virtually complete case ascertainment, and selecting matched controls from birth certificate data reduced problems of selection bias. We included only young children because exposures during the perinatal period may be more relevant for these shorter-latency tumors than for cancers in older children.

Future environmental or biologic monitoring studies are needed to evaluate how well proximity to agricultural pesticide use estimates exposure and to determine other significant predictor variables. Multidisciplinary studies that integrate exposure assessment, epidemiologic methods, and genetic information represent important next steps in evaluating risks to children's health associated with pesticide exposure.

Back to Top | Article Outline

ACKNOWLEDGMENTS

We extend our gratitude to the California Cancer Registry for providing the case data used in this study and to the California Office of Vital Records for providing birth certificate data and accommodating our abstracting efforts. We also thank Theresa Saunders for manuscript preparation.

Back to Top | Article Outline

REFERENCES

1. Daniels JL, Olshan AF, Savitz DA. Pesticides and childhood cancers. Environ Health Perspect. 1997;105:1068–1077.
2. Zahm SH, Ward MH. Pesticides and childhood cancer. Environ Health Perspect. 1998;106:893–908.
3. Whitmore RW, Immerman FW, Camann DE, et al. Non-occupational exposures to pesticides for residents of two US cities. Arch Environ Contam Toxicol. 1994;26:47–59.
4. Simcox NJ, Fenske RA, Wolz SA, et al. Pesticides in household dust and soil: exposure pathways for children of agricultural families. Environ Health Perspect. 1995;103:1126–1134.
5. Fenske RA, Kissel JC, Lu C, et al. Biologically based pesticide dose estimates for children in an agricultural community. Environ Health Perspect. 2000;108:515–520.
6. Reynolds P, Von Behren J, Gunier RB, et al. Childhood cancer and agricultural pesticide use: an ecologic study in California. Environ Health Perspect. 2002;110:319–324.
7. Kristensen P, Andersen A, Irgens LM, et al. Cancer in offspring of parents engaged in agricultural activities in Norway: incidence and risk factors in the farm environment. Int J Cancer. 1996;65:39–50.
8. Ma X, Buffler PA, Gunier RB, et al. Critical windows of exposure to household pesticides and risk of childhood leukemia. Environ Health Perspect. 2002;110:955–960.
9. Jaro MA. Probabilistic linkage of large public health data files. Stat Med. 1995;14:491–498.
10. TIGER Line Files [data file]. Washington, DC: US Bureau of the Census; 1995.
11. Census of Population and Housing, 1990: Summary Tape File 3 (California) [data file]. Washington, DC: US Bureau of the Census; 1992.
12. California Department of Pesticide Regulation, Environmental Monitoring and Pest Management Branch. Pesticide Use Reporting Data 1991–1997 [data file]. Sacramento, CA: California Department of Pesticide Regulation; 1999.
13. US Environmental Protection Agency, Office of Pesticide Programs. Reference Dose Tracking Report. Washington, DC: United States Environmental Protection Agency; 1997. Available at: http://nptn.orst.edu/tracking.htm.
14. Gold LS, Zeiger E. Handbook of Carcinogenic Potency and Genotoxicity Databases. New York: CRC Press; 1997.
15. California Department of Pesticide Regulation, Medical Toxicology Branch. Summaries of Toxicology Data. Sacramento, CA: California Department of Pesticide Regulation; 1997.
16. Gunier RB, Harnly ME, Reynolds P, et al. Agricultural pesticide use in California: pesticide prioritization, use densities, and population distributions for a childhood cancer study. Environ Health Perspect. 2001;109:1071–1078.
17. van den Berg F. Measured and computed concentrations of methyl isothiocyanate in the air around fumigated fields. Atmos Environ. 1993;27A:63–71.
18. Woodrow JE, Seiber JN, Baker LW. Correlation techniques for estimating pesticide volatilization flux and downwind concentrations. Environ Sci Technol. 1997;31:523–529.
19. Tiefenbacher JP. Mapping the pesticide driftscape: theoretical patterns of the drift hazard. Geograph Environ Model. 1998;2:83–101.
20. Woods N, Craig IP, Dorr G, et al. Spray drift of pesticides arising from aerial application in cotton. J Environ Qual. 2001;30:697–701.
21. Teske ME, Bird SL, Esterly DM, et al. AgDRIFT: a model for estimating near-field spray drift from aerial applications. Environ Toxicol Chem. 2002;21:659–671.
22. Kramarova E, Stiller CA. The international classification of childhood cancer. Int J Cancer. 1996;68:759–765.
23. Bell EM, Hertz-Picciotto I, Beaumont JJ. A case-control study of pesticides and fetal death due to congenital anomalies. Epidemiology. 2001;12:148–156.
24. Infante-Rivard C, Labuda D, Krajinovic M, et al. Risk of childhood leukemia associated with exposure to pesticides and with gene polymorphisms. Epidemiology. 1999;10:481–487.
25. Leiss JK, Savitz DA. Home pesticide use and childhood cancer: a case–control study. Am J Public Health. 1995;85:249–252.
26. Buckley JD, Robison LL, Swotinsky R, et al. Occupational exposures of parents of children with acute nonlymphocytic leukemia: a report from the Children's Cancer Study Group. Cancer Res. 1989;49:4030–4037.
27. Lowengart RA, Peters JM, Cicioni C, et al. Childhood leukemia and parents’ occupational and home exposures. J Natl Cancer Inst. 1987;79:39–46.
28. Schuz J, Kaletsch U, Meinert R, et al. Risk of childhood leukemia and parental self-reported occupational exposure to chemicals, dusts, and fumes: results from pooled analyses of German population-based case-control studies. Cancer Epidemiol Biomarkers Prev. 2000;9:835–838.
29. Baker LW, Fitzell DL, Seiber JN, et al. Ambient concentrations of pesticides in California. Environ Sci Technol. 1996;30:1365–1368.
30. Lee S, McLaughlin R, Harnly M, et al. Community exposures to airborne agricultural pesticides in California: ranking of inhalation risks. Environ Health Perspect. 2002;110:1175–1184.
31. Rull RP, Ritz B. Historical pesticide exposure in California using pesticide use reports and land-use surveys: an assessment of misclassification error and bias. Environ Health Perspect. 2003;111:1582–1589.
32. Shaw GM, Malcoe LH. Residential mobility during pregnancy for mothers of infants with or without congenital cardiac anomalies: a reprint. Arch Environ Health. 1992;47:236–238.
33. Loewenherz C, Fenske RA, Simcox NJ, et al. Biological monitoring of organophosphorus pesticide exposure among children of agricultural workers in central Washington State. Environ Health Perspect. 1997;105:1344–1353.
34. Azaroff LS. Biomarkers of exposure to organophosphorous insecticides among farmers’ families in rural El Salvador: factors associated with exposure. Environ Res. 1999;80:138–147.
35. Aprea C, Strambi M, Novelli MT, et al. Biologic monitoring of exposure to organophosphorus pesticides in 195 Italian children. Environ Health Perspect. 2000;108:521–525.
36. Lu C, Fenske RA, Simcox NJ, et al. Pesticide exposure of children in an agricultural community: evidence of household proximity to farmland and take home exposure pathways. Environ Res. 2000;84:290–302.
37. Raupach MR, Briggs PR, Ahmad N, et al. Endosulfan transport: II. Modeling airborne dispersal and deposition by spray and vapor. J Environ Qual. 2001;30:729–740.
38. Richards SM, McClure GY, Lavy TL, et al. Propanil (3,4-dichloropropionanilide) particulate concentrations within and near the residences of families living adjacent to aerially sprayed rice fields. Arch Environ Contam Toxicol. 2001;41:112–116.
39. Schuz J, Spector LG, Ross JA. Bias in studies of parental self-reported occupational exposure and childhood cancer. Am J Epidemiol. 2003;158:710–716.
Figure

Figure

© 2005 Lippincott Williams & Wilkins, Inc.