American Journal of Epidemiology
ª The Author 2011. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: email@example.com.
Vol. 173, No. 11
Advance Access publication:
March 28, 2011
Prostate Cancer and Ambient Pesticide Exposure in Agriculturally Intensive
Areas in California
Myles Cockburn*, Paul Mills, Xinbo Zhang, John Zadnick, Dan Goldberg, and Beate Ritz
* Correspondence to Dr. Myles Cockburn, Department of Preventive Medicine, Keck School of Medicine, University of Southern
California, 1441 Eastlake Avenue, MC 9175, Los Angeles, CA 90089-9175 (e-mail: firstname.lastname@example.org).
Initially submitted July 28, 2010; accepted for publication January 6, 2011.
In a population-based case-control study in California’s intensely agricultural Central Valley (2005–2006), the
authors investigated relations between environmental pesticide/fungicide exposure and prostate cancer. Cases
(n ¼ 173) were obtained from a population-based cancer registry, and controls (n ¼ 162) were obtained from
Medicare listings and tax assessor mailings. Past ambient exposures to pesticides/fungicides were derived from
residential history and independently recorded pesticide and land-use data, using a novel geographic information
systems approach. In comparison with unexposed persons, increased risks of prostate cancer were observed
among persons exposed to compounds which may have prostate-specific biologic effects (methyl bromide
(odds ratio ¼ 1.62, 95% confidence interval: 1.02, 2.59) and a group of organochlorines (odds ratio ¼ 1.64,
95% confidence interval: 1.02, 2.63)) but not among those exposed to other compounds that were included as
controls (simazine, maneb, and paraquat dichloride). The authors assessed the possibility of selection bias due to
less-than-100% enrollment of eligible cases and controls (a critical methodological concern in studies of this kind)
and determined that there was little evidence of bias affecting the estimated effect size. This study provides
evidence of an association between prostate cancer and ambient pesticide exposures in and around homes
in intensely agricultural areas. The associations appear specific to compounds with a plausible biologic role in
fungicides, industrial; hydrocarbons, brominated; pesticides; prostatic neoplasms; selection bias
Abbreviations: CI, confidence interval; DDE, dichlorodiphenyldichloroethane; GIS, geographic information systems; OR, odds
ratio; PLSS, Public Land Survey System; PUR, Pesticide Use Reports.
There are few established risk factors for prostate cancer,
and the search for plausible environmental causes is under
way. Prostate cancer incidence rates after age 45 years
increase at a rate approaching the ninth power of age (1),
which is compatible with an accumulated environmental
exposure, such as long-term exposure to occupational or en-
vironment toxins interfering with normal hormone function
(2), and with genetic differences in susceptibility (3). Vari-
ations in hormone levels affect prostate cancer risk, since
balance of androgen (sex hormone) levels (4). A variety of
pesticides have the ability to affect hormone functioning by
mimicking hormones, affecting enzyme systems involved in
hormone metabolism, or directly affecting the brain regions
involved in hormone functioning (5, 6). Furthermore, certain
pesticides may affect androgenic stimulation of the prostate,
potentially leading to cell proliferation and cancer (7).
The pesticide methyl bromide exhibits genotoxic poten-
tial (8). Studies in rats and mice (9, 10) indicate that methyl
bromide is an alkylating agent in which the methyl group
covalently binds to DNA, creating DNA adducts, O6- and
N7-methylguanine (9, 11). O6-methylguanine is a directly
miscoding lesion capable of pairing with both cytosine (the
correct nucleotide) and thymine (the incorrect nucleotide)
during DNA replication, resulting in a G:C to A:T transition
mutation (12, 13). These gene mutations may represent the
early steps in prostate carcinogenesis. Some organochlorine
pesticides induce overexpression of an oncogene implicated
1280Am J Epidemiol. 2011;173(11):1280–1288
in prostate cancer (erbB-2), which in turn produces cell
proliferation in prostate cancer cells (14).
Various farming occupations have high rates of prostate
cancer (15). In a recent case-control study carried out in
the United Farm Workers of America cohort, Mills and
Yang (16) derived pesticide exposure information by
cross-linking county-, crop-, month-, and year-specific work
histories from union records with similar county-, crop-,
month-, and year-specific Pesticide Use Reports (PUR)
from the California Department of Pesticide Regulation.
Associations were found between prostate cancer and a
group of organochlorine pesticides, and there was a weak
association with methyl bromide (16). The fungicide cap-
tan has also been shown to be related to prostate cancer
incidence in California farm workers (17).
(7) reported a weak summary relative risk of 1.1 for the
relation between farming and prostate cancer, a majority
of studies with a sufficiently large number of subjects have
shown excess relative risks of prostate cancer ranging from
1.06 to 5.0 among farmers, farm laborers, pesticide manu-
facturers, and pesticide applicators (5, 18), and a strong
case can be made for the biologic plausibility of an effect
of specific pesticides (such as some organochlorines and
methyl bromide) on prostate cancer, especially in agri-
culturally intense areas such as California’s Central Valley.
The majority of studies conducted to date have relied on
self-reports of pesticide exposure or on exposure data based
broadly on occupation; as a result, many investigators’ fail-
ure to find associations may have been due to exposure
misclassification, biasing effects towards the null (19).
Much current evidence regarding the role of pesticide
exposure in prostate cancer focuses on occupational expo-
sures not representative of those experienced by the general
population (15, 20–22). In this report, we focus on objective
estimates of residential exposure to pesticides in the envi-
ronment (‘‘ambient exposure’’) from drift or contaminated
soil/dust (19, 23?25) and their impact on prostate cancer
in a population-based case-control study. We studied the
impact of 3 select compounds with a biologically plausible
link to prostate cancer (i.e., methyl bromide, a group of
organochlorines (dicofol, dieldrin, dienochlor, endosulfan,
heptachlor, lindane, methoxychlor, and toxaphene), and
captan) and 3 compounds that are commonly used in the
Central Valley but have no specific link to prostate cancer
(simazine, maneb, and paraquat dichloride), to help rule out
the possibility that some factor associated in general with
pesticide exposures was responsible for increased risk.
Because previous studies may have been limited by their
potential biases by 1) improving exposure assessment meth-
ods and 2) assessing the potential impact of selection bias on
risk estimates for pesticide exposure in a novel manner.
MATERIALS AND METHODS
Selection of cases and controls
histologically confirmed prostate cancer between August
2005 and July 2006 in Tulare, Fresno, and Kern counties
(California) from the records of the California Cancer
Registry. Only non-Latino whites and Latino whites were
included; prostate cancer in other racial/ethnic groups was
too rare for inclusion (either because of the low rate of pros-
tate cancer in most other groups or, as in the case of African
Americans, because they represented small populations in
the Central Valley). The Registry provided cases’ addresses
and telephone numbers, which were used for recruitment,
and whenever these were unusable, we traced patients using
marketing company services to obtain updated contact
Control subject recruitment, described elsewhere (26),
was from a study of Parkinson’s disease being conducted
in the same study area. Controls aged 65 years or more were
identified from Medicare lists in 2001 and were mailed
recruitment materials, but because of implementation of
the Health Insurance Portability and Accountability Act,
which prohibited the continued use of Medicare enrollees,
additional controls were recruited from randomly selected
tax assessor residential units (parcels) in each of the 3
counties. Controls were recruited between 2004 and 2006.
We mailed recruitment materials to a random selection
of residential units and also attempted to identify head-of-
household names and telephone numbers for these parcels
using the services of marketing companies and Internet
searches. Eligibility criteria for the Parkinson’s disease
study controls were: 1) not having Parkinson’s disease
or prostate cancer and 2) being at least 60 years of age.
Additional eligibility criteria for both cases and controls
were: 1) currently residing primarily in one of the 3 study
counties and 2) having lived in California for at least 5 years
prior to the study.
Source of exposure data
Recruited cases and controls were mailed an informed
consent document, a timeline of important world events,
a blank 1900–2005 calendar (to aid in recalling residential
history), and a job history questionnaire, all to be completed
prior to interview. A comprehensive risk factor interview
was conducted by telephone and obtained demographic in-
formation (age (in years), race/ethnicity (non-Latino white
or Latino white), and birthplace (city and state in the United
States or city and country outside the United States)). We
obtained detailed information on residential history, includ-
ing dates of residence, address information, and local land-
marks, for the purpose of determining ambient pesticide
Assessment of ambient pesticide exposure
Age-specific ambient pesticide exposures and exposures
occurring between 1974 and 1999 were determined using
PUR data regarding the application of pesticides were com-
bined with data from land-use surveys (based on California’s
Public Land Survey System (PLSS), specifying the exact
location of specific crops on which those pesticides were
most likely used) to determine the geography of pesticide
Prostate Cancer and Ambient Pesticide Exposure1281
Am J Epidemiol. 2011;173(11):1280–1288
use between 1974 and 1999. Historical residential locations
between 1974 and 1999 were then used to determine poten-
place by summing data on pesticide use within a 500-m
buffer around the dwelling.
The California PUR form documents, for each pesticide
application, the name of the pesticide’s activeingredient, the
number of pounds of pesticide applied, the crop and acreage
of the field, the application method, and the date and loca-
tion of the application. We calculated annual application
rates (total pounds applied per acre in a PLSS section) for
all polygons linked to the PUR data, using an algorithm
developed and validated previously (19). When a PUR
matched exactly to land-use polygons in a PLSS section
by crop type, both records were directly linked. If pesticide
use was reported on a crop that did not match any of the
crops listed in the land-use survey in a PLSS section, yet the
section contained other field, vineyard, or orchard crops, we
assumed that these crop locations were possible sites where
the reported crop had been grown. Finally, if a PUR matched
a PLSS section but, according to the land-use survey, no
field, vineyard, or orchard crops were present in the section,
we assumed that any area within the entire section could
have been treated (19).
All historical residential addresses supplied by cases and
controls were geocoded to obtain a latitude and longitude.
Exposure misclassification due to the exclusion of locations
that were not geocodeable was a concern (28). We used man-
ual resolution methods to pinpoint locations that could not be
geocoded automatically, as detailed elsewhere (29). Briefly,
for every address, we obtained information on landmarks
and surrounding geographic features to aid in the manual
resolution of nonstandard addresses to a latitude/longitude.
We recorded the ‘‘accuracy’’ of geocoding in every case.
Sensitivity analyses specific to various levels of geocoding
accuracy did not alter any of the results presented herein.
Measurement of other pesticide exposures
While we collected detailed information on home use of
pesticides and occupational pesticide use, for the purposes
of this analysis we considered self-reports of home pesticide
use as representing ‘‘ever’’ or ‘‘never’’ use. Foroccupational
exposures, which were assessed by means of a job exposure
matrix approach according to job titles and tasks, we used
the method outlined in the article by Young et al. (30) to
define 3 exposure levels: ‘‘probably exposed to pesticides’’
(‘‘intensity’’ ? 0.3 in the paper by Young et al. (30)); ‘‘pos-
sibly exposed’’ (0 < ‘‘intensity’’ < 0.3); and ‘‘not exposed’’
(‘‘intensity’’ ¼ 0). The ‘‘intensity’’ of exposure reflects
the likelihood of exposure to pesticides/herbicides based
on self-reported occupation, weighted by usual exposures
experienced by persons in those occupations (30).
Selection of pesticides for analysis
We focused in this study on those chemicals thought most
likely to be related to prostate carcinogenesis (methyl bro-
mide, the group of organochlorines (dicofol, dieldrin, dien-
ochlor, endosulfan, heptachlor, lindane, methoxychlor, and
toxaphene), and captan). As control exposures, we selected
3 chemicals that are widely used in the 3-county study area
and have similar geographic distributions as each of the 3
potential prostate carcinogens yet are unlikely to produce
prostate-specific carcinogenic effects (paraquat dichloride,
simazine, and maneb).
We calculated odds ratios and 95% confidence intervals to
assess associations between specific pesticide exposures and
prostate cancer using unconditional logistic regression in
SAS 9.1.3 (SAS Institute, Inc., Cary, North Carolina). We
considered residential exposure for the period 1974–1999,
which is the time frame represented by complete PUR and
land-use data in our geographic information systems (GIS)
model, by summing all exposure values weighted by the
number of years of exposure. To assess the potential impact
of a lag between exposure and the development of disease,
we separately calculated exposures accumulated until 15
years before diagnosis and until 10 years before diagnosis.
In all cases, for years with missing exposures we imputed
exposure using the time-weighted average approach, which
imputes missing data with the average of the data from
nonmissing years for the same individual (31). Sources
of missing exposure data included missing residential in-
formation, unusable residential information, and residence
outside the 3-county study area.
Odds ratios were calculated comparing ‘‘any exposure’’
with ‘‘no exposure,’’ and also by comparing ‘‘low’’ and
‘‘high’’ exposures with ‘‘no exposure’’ to address possible
exposure-response. ‘‘Low’’ and ‘‘high’’ exposure cutpoints
were based on the median of the distribution of exposures to
each pesticide (pesticide group, in the case of organochlo-
rines) in control subjects. We adjusted these odds ratios for
age (continuous), race/ethnicity (Latino white, non-Latino
white), self-reported home pesticide use (ever/never), and
occupational pesticide exposures derived from the job ex-
posure matrix (not exposed, possibly exposed, or probably
exposed, as detailed above). No other variables from our
extensive risk factor questionnaire (e.g., body mass index
and crude food frequency items) were implicated in univar-
iate analyses (data not shown), so none were included in the
Assessment of selection bias in cases and controls
We estimated the impact of sample selection/participation
bias on pesticide exposures by calculating pesticide expo-
sures for all originally selected cases (n ¼ 670) and all orig-
inally selected controls (n ¼ 1,212) and comparing those
exposures to exposures among participating cases (n ¼
173) and controls (n ¼ 162). Littlewas known about prostate
cancer risk factors among the cases who did not respond to
the study invitation (other than the details provided in
Table 1), and nothing was known about nonresponding con-
trols. However, residential locations at recruitment (diagno-
sis address for Registry cases and residential tax assessor
parcel for all selected control locations, including those se-
lected from Medicare address listings) were known.
1282 Cockburn et al.
Am J Epidemiol. 2011;173(11):1280–1288
Therefore, pesticide exposures were generated for the 1974–
1999 time period (using the methods described above) for all
originally selected cases (n ¼ 670) solely on the basis of their
diagnosis address and for all tax assessor parcel centroids for
the population of potential control subjects (n ¼ 1,212). The
odds ratios for the selected cases and controls could be con-
sidered population-based estimates, whichare not affectedby
respondent participation bias. Comparable odds ratios were
also calculated for those cases and controls participating in
the study based on their residential diagnosis/contact address
we have presented here, because the bias analysis results are
based on 1974–1999 PUR data referenced only to a single
address at diagnosis/contact rather than a true residential his-
with which to adjust these estimates. Finally, we compared
participating cases with all those reported to the Registry dur-
and race/ethnicity—all obtained from Registry records.
The institutional review boards of the participating insti-
tutions approved this study.
Participating cases and controls
We attempted to contact 563 of the 640 eligible cases
before the study ended. The response rate among cases was
30.7%, and after removing 112 cases who had no valid con-
tact information, 37.9%. The only notable difference be-
tween the originally selected and analyzed cases was that
73% of the analyzed cases were non-Hispanic white as
compared with only 60% of the underlying Registry cases
(Table 1). All subsequent analyses were controlled for
race/ethnicity. We conducted analyses stratified by stage of
disease (localized vs. regional, distant, and unknown com-
bined),andthe results were similar tothosepresented below.
For the Parkinson’s disease study (26), the source of con-
trol subjects, we successfully contacted 1,212 potential con-
trols by mail and/or phone for eligibility screening. A total
of 457 controlswere ineligible: 409were tooyoung,44 were
terminally ill, and 4 primarily resided outside of the study
area. Of the 755 eligible population controls, 409 (54%)
declined participation, were too ill to honor an appointment,
or moved out of the area prior to interview; 346 (46%) were
enrolled, and 162 of these persons were males aged 60–74
years and were included in this analysis (Table 2).
Exposure to chemicals with a potential role in prostate
We provide results only for the most accurate exposure
metric, based on residential locations between 1974 and
1999, when PUR and PLSS data were both available
(Table 3). Results for exposures incurred prior to 15 years
before diagnosis, exposures incurred prior to 10 years before
diagnosis, and residence at diagnosis are presented in the
text below only where they differed from the 1974–1999
exposure results. For 23 cases, there were no available ex-
posure data because no usable residential history for any
year of the period 1974–1999 could be obtained. These 23
cases did not differ from the remaining 150 according to any
of the variables presented in Table 1, and their diagnosis-
address pesticide exposures appeared to be similar to those
occurring among the 150 included cases. Only 12 partici-
pants had 1 or more years of missing address information
Cases Selected From the Population-based California Cancer
Registry, California’s Central Valley, 2005–2006
Characteristics of Surveyed Prostate Cancer Cases and
Selected CasesSurveyed Cases
Age at diagnosis, years
60–64151 26.851 29.5
65–69215 38.266 38.2
70–74197 35.056 32.4
or missing data
10318.3 28 16.2
United States 230 40.97945.7
Other country 70 12.425 14.5
Missing data263 46.76939.9
Non-Latino white 33759.9 12672.8
Latino white 22640.147 27.2
Total563 100 173 100
Subjects, California’s Central Valley, 2005–2006
Characteristics of Prostate Cancer Cases and Control
(n 5 173)
(n 5 162)
Age group, years
60–6451 2946 28
65–69 6638 3723
70–7456 3279 49
Non-Latino white126 73124 77
Latino white 4727 3823
Not exposed 9756 10565
Home pesticide usea
Not exposed138 159
Prostate Cancer and Ambient Pesticide Exposure1283
Am J Epidemiol. 2011;173(11):1280–1288
during the 1974–1999 period (representing 0.5% of the per-
son-years of exposure data, because most were missing only
1 year); 98 participants spent 1 or more years out-of-county
during the 1974–1999 period (1.96% of the person-years of
exposure data, again because most were out-of-county for
only 1 year), and these were the participant-years for which
exposures were imputed. Excluding any of these observa-
tions did not affect the results presented below.
Exposure to methyl bromide was associated with an in-
creased risk of prostate cancer (odds ratio (OR) ¼ 1.62, 95%
California’s Central Valley, 2005–2006b
Associations Between Prostate Cancer Risk and Exposureato Selected Pesticides,
(High vs. Low)
Unexposed 63 85 1.00
Exposed87 701.62 1.02, 2.59
Low exposure4535 1.81 1.03, 3.180.10
4235 1.45 0.82, 2.57
Unexposed 55 78 1.00
Exposed9577 1.64 1.02, 2.63
Low exposure35 38 1.25 0.75, 2.080.037
High exposure6039 2.031.17, 3.52
Missing data 237
Exposed58 511.20 0.74, 1.96
Low exposure1725 0.68 0.34, 1.360.04
High exposure 41261.74 1.01, 3.13
Exposed 68 68 0.960.60, 1.53
Low exposure30340.84 0.47, 1.520.79
High exposure 38 34 1.07 0.60, 1.89
Missing data 237
Exposed32 34 0.850.48, 1.51
Low exposure 14170.71 0.32, 1.57 0.70
High exposure 1817 1.000.47, 2.09
Missing data 237
Exposed 103 93 1.420.87, 2.31
Low exposure 4946 1.370.78, 2.41 0.37
High exposure 54 47 1.470.82, 2.60
aBased on historical exposures at reported residences between 1974 and 1999.
bResults are based on California Pesticide Use Reports and address data for 1974–1999.
cAdjusted for age, race/ethnicity, home pesticide use (yes/no), and occupational pesticide
exposure (none, possible, or probable); see text for details.
dThere were no exposure data for missing observations for the period 1974–1999 (n ¼ 23).
eDicofol, dieldrin, dienochlor, endosulfan, heptachlor, lindane, methoxychlor, and toxaphene.
1284 Cockburn et al.
Am J Epidemiol. 2011;173(11):1280–1288
confidence interval (CI): 1.02, 2.59), but we did not observe
evidence for exposure-response (Table 3). Methyl bromide
exposures at the diagnosis address were associated with
a higher risk of prostate cancer (OR ¼ 3.60, 95% CI:
1.62, 8.20) than any of the other periodic exposures, and
there was evidence of exposure-response (for ‘‘low’’
exposure, OR ¼ 2.75; for ‘‘high’’ exposure, OR ¼ 4.01;
P ¼ 0.009 for the difference). Exposure to the group of
organochlorines was associated with an overall increase in
risk (OR ¼ 1.64, 95% CI: 1.02, 2.63; Table 3), and therewas
substantial evidence for exposure-response, with a stronger
risk increase in the ‘‘high’’ exposure category (OR ¼ 2.03,
95% CI: 1.17, 3.52). Exposure to captan showed little dif-
ference in the comparison of unexposed and exposed groups
(OR ¼ 1.20, 95% CI: 0.74, 1.96) but elevated risk at high
levels of exposure (OR ¼ 1.74, 95% CI: 1.01, 3.13), with
a statistically significant exposure-response trend (P-trend ¼
0.04). None of the estimates for captan, the organochlorines,
or methyl bromide restricted to the periods 10 and 15 years
prior to diagnosis varied significantly from those based on
the 1974–1999 exposure period reported above.
Exposure to chemicals considered unlikely to be
related to prostate carcinogenesis
We observed a slightly increased risk of prostate cancer at
the highest level of simazine exposure, but the 95% confi-
dence interval included the null value. Exposure to paraquat
showed some evidence of a small increase in risk, but that
increase was compatible with chance (Table 3). Any expo-
sure to paraquat (based on residence of diagnosis only) in-
creasedtheriskofprostate cancer(OR¼ 2.04,95%CI:1.10,
3.78), but the exposure-response pattern was inconsistent:
Higher risk occurred at low levels of exposure (low ex-
posure: OR ¼ 3.78; high exposure: OR ¼ 1.84). Exposure
to maneb did not appear to be related to increased risk of
prostate cancer for any of the exposure metrics.
Assessment of selection bias
Using the limited exposure data available on all selected
cases and controls (regardless of participation status), we
foundthatpesticide exposures in participating studysubjects
were likely to be slightly underestimated (Table 4). Some
odds ratios appeared to be affected by sample selection bias,
but the direction and magnitude of the bias depended on the
agent: For the group of organochlorines, the odds ratio in
the case-control sample was 2.05, whereas the odds ratio in
the underlying population was only 1.50, an estimate closer
to the value of 1.64 based on address history in our sample
(Table 3). The odds ratio for captan changed in the opposite
direction, and the methyl bromide estimate appeared to
be relatively unaffected by selection bias (Table 4).
In this population-based case-control study, we found ev-
idence of a strong association of ambient exposure to methyl
bromide and a group of organochlorines with prostate can-
cer risk. We tested only selected pesticides based on a priori
hypotheses, and we found some evidence that the effects
were limited to those compounds or chemicals most likely
to be involved in prostate cancer initiation and promotion
(methyl bromide, captan, and the group of organochlorines),
with little evidence of a consistent effect for a selection of
Valley As Compared With the Study Sample, Using Only Diagnosis Address (Cases) and Recruitment Address (Controls)a, 2005–2006
Associations Between Prostate Cancer Risk and Exposure to Selected Pesticides in the Underlying Population in California’s Central
Exposure and SampleTotal No. No. Exposed % ExposedData SourceOdds Ratio
Population controls 1,21265954.3
Sample controls162 7345.1
Population cases 670 40660.6 Population 1.471.22, 1.78
17392 53.1Sample 1.44 0.93, 2.22
Sample controls162 8451.9
Population cases670 45968.5 Population1.50 1.23, 1.83
Sample cases 173 11767.6Sample2.051.31, 3.21
Population controls 1,212 39232.3
Sample controls 16251 31.5
Population cases670 322 48.1Population2.12 1.75, 2.57
Sample cases 17368 39.3Sample 1.45 0.92, 2.28
aExposures were determined for all selected cases and controls, only from the diagnosis address in cases and the address used for recruitment
for potential controls; see text for details.
bDicofol, dieldrin, dienochlor, endosulfan, heptachlor, lindane, methoxychlor, and toxaphene.
Prostate Cancer and Ambient Pesticide Exposure 1285
Am J Epidemiol. 2011;173(11):1280–1288
compounds that are widely used but not thought to have
a biologically plausible role in prostate cancer (paraquat
dichloride, simazine, and maneb). Some of the associations
observed were quite substantial, with the group of organo-
chlorines showing over a 2-fold increase in risk of prostate
cancer at higher exposure, albeit this estimate had relatively
wide confidence intervals (OR ¼ 2.03, 95% CI: 1.17, 3.52).
To our knowledge, no previous studies of prostate cancer
and nonoccupational exposure to pesticides have been con-
ducted in areas of such high potential exposure. California is
the most productive agricultural state in the nation, with an
annual production each year of more than 20 billion dollars’
worth of different crops and commodities. Each year, the
use of pesticides for agricultural purposes in California
exceeds 250 million pounds of active ingredients, about
one-quarter of all pesticides used in the United States.
Fresno, Kern, and Tulare counties are ranked as the top
agricultural counties in California by value of production
(32), increasing the possibility of residential (‘‘ambient’’)
exposures’ occurring via aerial drift of pesticides into neigh-
borhoods and contamination of drinking water (33, 34).
While the prevalence of residential exposure may be very
ably relatively low; however, they are likely to accumulate
substantially over a lifetime (35).
The most important recent data supporting a role of
agricultural exposures in prostate cancer come from the
Agricultural Health Study, in which Alavanja et al. (15)
reported statistically significant relative risk estimates of
approximately 1.3 for any pesticide exposure, but that study
did not assess agricultural exposuresother than occupational
ones (such as residential exposure among persons living
in agricultural areas) and may have provided poorer esti-
mates of longer-term exposure (relying on accurate recall).
Recently, Greenburg et al. (36) did not find any association
between captan and prostate cancer in the Agricultural
Health Study (odds ratios across quartiles were 1.00, 1.13,
0.82, and 1.02; none were significant), but it is difficult to
compare that finding with our results, which deal with res-
idential exposure in the general population rather than in
specific occupational groups. While we might expect that
any true associations between pesticide exposures and
disease outcome would be especially high in occupationally
exposed persons, selection bias (the healthy worker effect),
exposure misclassification (e.g., recall bias), and real differ-
ences in exposure levels (e.g., if pesticide workers use
effective exposure remediation methods) could all produce
substantial differences in observed odds ratios between
occupational and nonoccupational studies.
In this study, we considered exposures occurring in the
general population, not a specific occupationally exposed
cohort. This had the advantage that results were not biased
by the healthy worker effect and could address the impact
of ubiquitous pesticide exposures occurring in the general
population. Estimation of exposure did not rely on recall
of pesticide use. The validity of our PUR plus land-use
GIS model in estimating potential residential exposures
from pesticide applications to nearby agricultural crops
(37) was previously examined: Using lipid-adjusted dichlor-
odiphenyldichloroethane (DDE) serum levels as the ‘‘gold
standard’’ for pesticide exposure, GIS-based organochlorine
estimates had high specificity (87%), and together with body
mass index, age, gender, mixing and loading pesticides
by hand, and using pesticides in the home, they explained
47% of the variance in serum DDE levels (37). Adjustment
(albeit somewhat crude) in the current study included occu-
pational exposure to pesticides and home pesticide use,
along with other potential confounders of the association
between pesticide exposures and prostate cancer.
We also assessed the potential impact of selection bias
for exposures in this case-control study. Ordinarily, this is
not possible because the exposures in nonresponding survey
or study subjects are unknown. There was some suggestion
of bias in our pesticide exposure estimates; however, the
direction of this bias was unpredictable, and the overall
effects of specific pesticides on prostate cancer remained.
Underestimation of pesticide exposures among control sub-
jects in the worst case resulted in overestimation of the
effect of any organochlorine exposure by 50% (in the par-
ticipating subjects, OR ¼ 2.05, 95% CI: 1.31, 3.21; in the
underlying population, OR ¼ 1.50, 95% CI: 1.23, 1.83), but
it also resulted in underestimation of exposures for cases,
such that for captan, the odds ratio in the selected cases
and controls was 2.12 (95% CI: 1.75, 2.57). While the re-
sponse rate in our study was relatively low, it is comparable
to the rates commonly found in population-based case-
control studies, and the cases included in this study were
a relatively unbiased sample of the population-based series
from which they were drawn with respect to demographic
factors and the distribution of potentially confounding
factors such as age.
Methyl bromide use has recently substantially declined in
California because of international recognition that it is an
ozone-depleting substance. The Environmental Protection
Agency implemented a 100% phase-out plan scheduled
for completion by 2005, but specially permitted use (where
no practical alternative exists) continues (38). The grouping
of organochlorines that we considered here consists of
chemicals whose use is regulated (and must be reported)
but is not limited in any manner that has affected their
expanding use over time.
sample size and a more detailed assessment of the relative
importance of combinations of pesticides and of highly
correlated pesticide exposures. While further improvements
in exposure assessment and study design are undoubtedly
warranted, we have demonstrated a method of pesticide
exposure assessment that appears to minimize selection bias
and exposure misclassification, resulting in fairly compel-
ling evidence of measurable associations between prostate
cancer and those pesticides with a biologically plausible
mechanism in prostate carcinogenesis.
Author affiliations: Department of Preventive Medicine,
Keck School of Medicine, University of Southern California,
Los Angeles, California (Myles Cockburn, Xinbo Zhang, John
1286Cockburn et al.
Am J Epidemiol. 2011;173(11):1280–1288
Zadnick); Spatial Sciences Institute, University of Southern
California, Los Angeles, California (Myles Cockburn, Dan
Goldberg); Department of Medicine, University of California,
San Francisco, Fresno, California (Paul Mills); and Depart-
ment of Epidemiology and Center for Occupational and
Environmental Health, School of Public Health, University
of California, Los Angeles, Los Angeles, California (Beate
This work was supported by the National Institute of En-
vironmental Health Sciences (grants ES10544, U54ES12078,
and P30 ES07048), the National Cancer Institute (grant
CA110846), the National Institute of Neurological Disorders
Prostate Cancer Research Program (grant 051037); in addi-
tion, initial pilot funding was provided by the American
Parkinson’s Disease Association.
Conflict of interest: none declared.
1. Ross RK, Schottenfeld D. Prostate Cancer. New York, NY:
Oxford University Press; 1996.
2. Chen C, Weiss NS, Stanczyk FZ, et al. Endogenous sex
hormones and prostate cancer risk: a case-control study nested
within the Carotene and Retinol Efficacy Trial. Cancer
Epidemiol Biomarkers Prev. 2003;12(12):1410–1416.
3. Ntais C, Polycarpou A, Tsatsoulis A. Molecular epidemiology
of prostate cancer: androgens and polymorphisms in andro-
gen-related genes. Eur J Endocrinol. 2003;149(6):469–477.
4. Hsing AW. Hormones and prostate cancer: what’s next?
Epidemiol Rev. 2001;23(1):42–58.
5. Keller-Byrne JE, Khuder SA, Schaub EA. Meta-analyses of
prostate cancer and farming. Am J Ind Med. 1997;31(5):
6. Janssens JP, Van Hecke E, Geys H, et al. Pesticides and
mortality from hormone-dependent cancers. Eur J Cancer
7. Parent ME, Siemiatycki J. Occupation and prostate cancer.
Epidemiol Rev. 2001;23(1):138–143.
8. Garrett NE, Stack HF, Waters MD. Evaluation of the genetic
activity profiles of 65 pesticides. Mutat Res. 1986;168(3):
9. Djalali-Behzad G, Hussain S, Osterman-Golkar S, et al.
Estimation of genetic risks of alkylating agents. VI. Exposure
of mice and bacteria to methyl bromide. Mutat Res. 1981;
10. Pletsa V, Steenwinkel MJ, van Delft JH, et al. Methyl bromide
causes DNA methylation in rats and mice but fails to induce
somatic mutations in klacZ transgenic mice. Cancer Lett.
11. Gansewendt B, Foest U, Xu D, et al. Formation of DNA
adducts in F-344 rats after oral administration or inhalation
of [14C]methyl bromide. Food Chem Toxicol. 1991;29(8):
12. Coulondre C, Miller JH. Genetic studies of the lac repressor.
IV. Mutagenic specificity in the lacI gene of Escherichia coli.
J Mol Biol. 1977;117(3):577–606.
13. Loechler EL, Green CL, Essigmann JM. In vivo mutagenesis
by O6-methylguanine built into a unique site in aviral genome.
Proc Natl Acad Sci U S A. 1984;81(20):6271–6275.
14. Tessier DM, Matsumura F. Increased ErbB-2 tyrosine kinase
activity, MAPK phosphorylation, and cell proliferation in the
prostate cancer cell line LNCaP following treatment by select
pesticides. Toxicol Sci. 2001;60(1):38–43.
15. Alavanja MC, Samanic C, Dosemeci M, et al. Use of agri-
cultural pesticides and prostate cancer risk in the Agricultural
Health Study cohort. Am J Epidemiol. 2003;157(9):800–814.
16. Mills PK, Yang R. Prostate cancer risk in California farm
workers. J Occup Environ Med. 2003;45(3):249–258.
17. Mills PK. Correlation analysis of pesticide use data and cancer
incidence rates in California counties. Arch Environ Health.
18. Dich J, Wiklund K. Prostate cancer in pesticide applicators in
Swedish agriculture. Prostate. 1998;34(2):100–112.
19. Rull RP, Ritz B. Historical pesticide exposure in California
using pesticide use reports and land-use surveys: an assess-
ment of misclassification error and bias. Environ Health
20. Boers D, Zeegers MP, Swaen GM, et al. The influence of
occupational exposure to pesticides, polycyclic aromatic
hydrocarbons, diesel exhaust, metal dust, metal fumes, and
mineral oil on prostate cancer: a prospective cohort study.
Occup Environ Med. 2005;62(8):531–537.
21. Zeegers MP, Friesema IH, Goldbohm RA, et al. A prospective
study of occupation and prostate cancer risk. J Occup Environ
22. Van Maele-Fabry G, Willems JL. Occupation related pesticide
exposure and cancer of the prostate: a meta-analysis. Occup
Environ Med. 2003;60(9):634–642.
23. Ritz B, Yu F, Chapa G, et al. Effect of air pollution on preterm
birth among children born in Southern California between
1989 and 1993. Epidemiology. 2000;11(5):502–511.
24. Ritz B, Yu F, Fruin S, et al. Ambient air pollution and risk of
birth defects in Southern California. Am J Epidemiol. 2002;
25. Rull RP, Ritz B, Shaw GM. Neural tube defects and maternal
residential proximity to agricultural pesticide applications. Am
J Epidemiol. 2006;163(8):743–753.
26. Costello S, Cockburn M, Bronstein J, et al. Parkinson’s disease
and residential exposure to maneb and paraquat from agri-
cultural applications in the Central Valley of California. Am J
27. Marusek JC, Cockburn MG, Mills PK, et al. Control selection
and pesticide exposure assessment via GIS in prostate cancer
studies. Am J Prev Med. 2006;30(suppl 2):S109–S116.
28. Bonner MR, Han D, Nie J, et al. Positional accuracy of
geocoded addresses in epidemiologic research. Epidemiology.
29. Goldberg DW, Wilson JP, Knoblock CA, et al. An effective and
efficient approach for manually improving geocoded data. Int
J Health Geogr. 2008;7:60. (doi: 10.1186/1476-072X-7-60).
30. Young HA, Mills PK, Riordan D, et al. Use of a crop and job
specific exposure matrix for estimating cumulativeexposure to
triazine herbicides among females in a case-control study in
the Central Valley of California. Occup Environ Med.
31. Weinberg CR, Moledor ES, Umbach DM, et al. Imputation for
exposure histories with gaps, under an excess relative risk
model. Epidemiology. 1996;7(5):490–497.
32. California Department of Pesticide Regulation. Pesticide Use
Reporting—2008 Summary Data. Sacramento, CA: California
Department of Pesticide Regulation; 2010. (http://www.
October 10, 2010).
33. Brouwer DH, Brouwer EJ, van Hemmen JJ. Estimation of
long-term exposure to pesticides. Am J Ind Med. 1994;25(4):
Prostate Cancer and Ambient Pesticide Exposure1287
Am J Epidemiol. 2011;173(11):1280–1288
34. Savitz DA, Arbuckle T, Kaczor D, et al. Male pesticide
exposure and pregnancy outcome. Am J Epidemiol. 1997;
35. Garcı ´a AM. Occupational exposure to pesticides and
congenital malformations: a review of mechanisms, methods,
and results. Am J Ind Med. 1998;33(3):232–240.
36. Greenburg DL, Rusiecki J, Koutros S, et al. Cancer
incidence among pesticide applicators exposed to captan in
the Agricultural Health Study. Cancer Causes Control.
37. Ritz B, Costello S. Geographic model and biomarker-derived
measures of pesticide exposure and Parkinson’s disease. Ann N
Y Acad Sci. 2006;1076:378–387.
38. Environmental Protection Agency. The Phaseout of Methyl Bro-
mide. Washington, DC: Environmental Protection Agency; 2010.
(http://www.epa.gov/ozone/mbr/). (Accessed October 10, 2010).
1288Cockburn et al.
Am J Epidemiol. 2011;173(11):1280–1288