International Archives of Occupational and Environmental Health

, Volume 77, Issue 8, pp 559–570

Prostate cancer among pesticide applicators: a meta-analysis

Authors

    • Department of Public HealthUniversity of Ghent
    • Unité de Toxicologie Industrielle et Médecine du travail, Ecole de Santé PubliqueUniversité Catholique de Louvain
  • J. L. Willems
    • Department of Public HealthUniversity of Ghent
Original Article

DOI: 10.1007/s00420-004-0548-8

Cite this article as:
Van Maele-Fabry, G. & Willems, J.L. Int Arch Occup Environ Health (2004) 77: 559. doi:10.1007/s00420-004-0548-8

Abstract

Objectives: To analyse data from peer-reviewed, case-referent and cohort studies, studying the occurrence of prostate cancer in pesticide applicators and in some other, related, occupational categories, in order to determine a possible relationship of cancer of the prostate with pesticide exposure; to calculate a meta-rate ratio and to compare it with the meta-rate ratios obtained in a previous meta-analysis performed over a shorter time (1995–2001) in a broader exposure category, including many pesticide-related agricultural and non-agricultural occupations. Methods: Medline was searched for the years between 1966 and 2003, and relevant studies were identified from 1986 on. We conducted a meta-analysis of 22 studies complying with the inclusion criteria in order to pool their relative risk (RR) estimates. Studies were summarised and assessed for homogeneity and publication bias. Results: The meta-rate ratio, based on 22 estimates of RR, is 1.24 [95% confidence interval (95% CI) 1.06–1.45]. This pooled risk estimate for the occupational categories selected is higher than the one previously calculated for farmers in general over a shorter period of publication. Substantial heterogeneity of rate ratios exists between the different studies. The major source of heterogeneity identified is geographic location. Increased meta-rate ratios are observed for studies derived from North America as well as from Europe, the meta-rate ratios from Europe being lower than those from North America. There is no obvious indication of publication bias. Conclusion: The increased meta-rate ratio for prostate cancer in agricultural pesticide applications provides additional evidence for a possible relationship between pesticide exposure and prostate cancer. The homogeneity observed between the individual rate ratios, after we had regrouped the data according to geographic location, tends to increase the consistency of the association. However, the data available from the individual studies do not provide sufficient exposure information for firm conclusions to be drawn about pesticide exposure as the cause of prostate cancer, independently from other factors.

Keywords

Meta-analysisPesticidesProstatic neoplasmsPesticide applicatorsRisk

Introduction

Increasing application of pesticides to control weeds, plant disease and insects has stimulated investigations on their chronic health effects and, particularly, on possible carcinogenic effects in occupational groups exposed to those agents. Compared with the general population, as well as with many other occupational categories, farmers in general were reported to have a lower risk of mortality from all causes and a lower occurrence of all neoplasms (Wiklund 1983; Blair et al. 1985; Notkola et al. 1987; Stark et al. 1987; Inskip et al. 1996; Cerhan et al. 1998; Andersen et al. 1999; Sperati et al. 1999). Several epidemiological studies, however, have shown that some specific cancers, including prostate cancer, may show increased incidence and mortality rates in this group even if the relative risk was often low (Ernster et al. 1979; Burmeister 1981; Burmeister et al. 1983; Blair and Zahm 1991; Forastiere et al. 1993; Morrison et al. 1993; Inskip et al. 1996; Cerhan et al. 1998; Dich and Wicklund 1998, Krstev et al. 1998; Band et al. 1999; Buxton et al. 1999; Cantor and Silberman 1999; Fleming et al. 1999a, 1999b; Parker et al. 1999; Sharma-Wagner et al. 2000). Four meta-analyses (Blair et al. 1992; Keller-Byrne et al. 1997; Acquavella et al. 1998; Van Maele-Fabry and Willems 2003) of epidemiological studies calculated a slightly elevated overall risk for prostate cancer in farmers and other agricultural occupation-related pesticide exposure groups. This suggests that a common characteristic constitutes a risk factor for the disease. As agricultural workers are exposed to many, potentially hazardous, physical, chemical and biological agents (including pesticides, ultraviolet radiation, solvents, mineral and organic dusts, viruses and microbes), it has not yet been possible to identify the specific agent(s) that may be responsible for this elevated risk for prostate cancer, but the strongest link, to date, is with pesticides (Alavanja et al. 1999).

In our previous meta-analysis (Van Maele-Fabry and Willems 2003) of studies published between 1995 and 2001 that addressed a broad range of pesticide-related occupational categories, the highest pooled rate ratio was observed in the subset category of pesticide applicators, but this was based on four studies only. As a consequence, we decided to repeat the exercise, focusing on occupational categories with, most likely, a greater exposure to pesticides than farmers in general, and to enlarge the time period covered by the review (1966–2003 instead of 1995–2001). The most obvious groups in which to study the chronic effects of pesticides are those occupational groups that apply pesticides as part of their daily activities, possibly over many years (Moses et al. 1993; Dich et al. 1997). These include agricultural pesticide applicators, farmers licensed to use pesticides, farmers explicitly reported as having been exposed to pesticides, nursery and greenhouse workers, as well as employees of companies with pesticide-spraying activities.

Materials and methods

Study identification and selection

Study identification

We searched Medline (National Library of Medicine, Bethesda, Md., USA) for the period 1966 to 1 May 2003. The search strategy used several combinations of the following key words: prostatic neoplasms [medical subject headings (MeSH)], pesticides (MeSH), occupational exposure (MeSH), cancer, applicators, licensed, sprayers, users. In a second step we checked the lists of references of the studies identified, going down step by step but limiting ourselves to studies published in the open literature.

Study selection

All studies complying with the following inclusion criteria were taken into consideration for a first overall assessment:
  • Surveys published in English, in peer-reviewed journals between 1966 and 2003

  • Case–control or cohort design studies

  • Providing sufficient data to determine an estimate of relative risk (RR) for prostate cancer and its confidence interval

  • Referring to the occupational groups of interest

Studies were excluded from the analysis if they
  • Included subjects already included in another more complete or more recent study of similar design and examining a greater number of subjects or with longer follow-up time

  • Reported fewer than five exposed cases

  • Gave only proportional mortality ratios (PMRs)

  • Did not report original results (reviews, comments, letters, editorials).

The occupational groups of interest, e.g. agricultural pesticide applicators, farmers licensed to use pesticides, nursery and greenhouse workers, and employees of companies involved in spraying activities (provided 50% or more of the workers were classified as sprayers), are more likely to be more exposed to pesticides than the category of farmers in general. As in our previous meta-analysis, studies reporting fewer than five exposed cases were excluded from the main computation, although they were taken into account in the sensitivity analysis.

Data extraction

A structured abstract was derived by one of us (G.V.M.F.) for each study identified. In addition, we both read the reports and independently extracted and tabulated the most relevant RR estimates, with their 95% confidence intervals (95% CIs). We compared the results of this exercise and obtained a consensus before the meta-analysis.

When multiple estimates of RR were given, we retained the data on which the authors had relied for their assessment or the overall data, if the data concerned specifically one of the exposure groups of interest. We did not include data resulting from further stratifications, e.g., by flight hours of application, by grade of potential exposure or duration of exposure, by year of birth or of licence, by age at death, by duration of employment, by licence type (private, commercial, public), by type of crop grown or of land (arable, woodland, mixed). In the other cases we chose data restricted to men believed to be subject to the least amount of misclassification: licensed farmers with more than 10 years of farming, pesticide or insecticide use explicitly mentioned as being part of their occupational activities, orchards and greenhouses as exposure indicators, having sprayed 250 or more acres with herbicides. We retained regional or provincial references instead of national references and other farmers as a comparison group instead of non-farmers.

Data analysis

A detailed description of the data analysis has been published before (Van Maele-Fabry and Willems 2003). In brief, in order to test between-study comparability, we determined the homogeneity among studies, using a chi-squared test. In the case of heterogeneity, we applied stratification in order to identify its source. A source was considered important if stratification for that source markedly increased the P value of the chi-squared statistic for the stratum-specific estimates of effect (e.g. if the P value increased from less than 0.01 to greater than 0.1) (Lipsett and Campleman 1999). In the absence of heterogeneity, we calculated RRs and CIs according to a fixed-effect model (Woolf 1955), which assumes that results across studies differ only by sampling error. When there was heterogeneity, we calculated RRs and CIs according to a random-effect model (DerSimonian and Laird 1986) and we determined potential sources of heterogeneity by subset analysis.

We conducted sensitivity analyses to estimate the importance of individual studies in the combined summary statistic and to determine whether any of these had a disproportionate influence (Olkin 1994).

Two methods were used to assess publication bias. Potential publication bias due to study size was explored by the plotting of the natural logarithm of the estimate of RR (ln RR) versus the inverse of standard error (1/SE). Funnel plot asymmetry was tested by the linear regression method suggested by Egger et al. (1997).

In order to determine whether any positive or negative trend had occurred with time, we plotted the estimates of RR versus publication date.

Results

Among the references retrieved, more than 200 dealt with our subject of interest. After exclusion of the references reporting non-original results (reviews, comments, letters, editorials) or reporting on occupational groups outside our selected groups, 31 studies were selected (Table 1). Although Medline was searched for the years from 1966 onwards, the first study complying with the inclusion criteria dated from 1986. Nine of these 31 studies were excluded from the meta-analysis (see italics in the table, mentioning the reasons for their exclusion). Among the 22 studies finally included in the meta-analyses, seven were case–control studies and 15 were cohort studies.
Table 1

Abstracted risk estimates and study information from the original studies relating agricultural pesticide application and prostate cancer. Studies that were excluded from the meta-analyses are italicised. Prevalence is prostate cancer/all cancer No. of cases number of exposed cases, RR risk ratio, SMR standardised mortality ratio, SIR standardised incidence ratio, OR odds ratio, NP data not presented, MCPA 2 methyl-4-chlorophenoxyacetic acid

Reference

Exposure group

Geographical location

No. of cases

Specified measure

Estimate of RR

95% CI

Cohort studies

Alavanja et al. (2003)

Pesticide applicators

Iowa and North Carolina

566

SIR

1.14

1.05–1.24

Alberghini et al. (1991)d

Farmers licensed to use pesticides

Italy (Emilia Romagna: between Modena, Ferrara and Bologna)

10

SMR

  

 National basis

0.59

0.281.09

 Regional basis

0.61

0.29–1.13

Cantor and Booze (1991)a

Aerial pesticide applicators

USA

5

SMR

1.36

0.443.17

Cantor and Silberman (1999)

Aerial pesticide applicators

USA

21

SMR

1.40

0.87–2.14

Coggon et al. (1986)

Workers exposed to MCPA and other phenoxy acid herbicides (high potential exposure)

UK

5

SMR

1.51

0.49–3.52

Dich and Wiklund (1998)

Pesticide applicators

Sweden

401

SIR

1.13

1.02–1.24

Figa-Talamanca et al. (1993a)d

Rural licensed pesticide users

Italy (Rome)

6

SMR

  

 Provincial basis

0.89

0.33–1.93

 National basis

1.00

0.372.17

Fleming et al. (1999a)

Licensed pesticide applicators

Florida

353

SIR

1.91

1.72–2.13

Fleming et al. (1999b)

Licensed pesticide applicators

Florida

64

SMR

2.38

1.83–3.04

Fleming et al. (2003)

Pesticide-exposed workers (farmers and pesticide applicators)

USA

22

RR

1.3

0.8–2.2

Gambini et al.(1997)

Rice growers (=seasonal herbicide applicators)

Italy (Piedmont: Novara province)

19

SMR

0.96

0.58–1.50

Garry et al. (1994)b

Pesticide appliers

Minnesota

2

Prevalence: SMR: NP

2/25

 

Kogevinas et al. (1997)e

Workers exposed to any phenoxy herbicides or chlorophenols

International

68

SMR

1.10

0.851.39

Kristensen et al. (1996)

Orchards and greenhouses

Norway

33

RR

1.45

1.01–2.09

Morrison et al. (1993)

Farmers believed to be subject to the least amount of misclassification: >250 acres herbicide sprayed

Canada (Manitoba, Saskatchewan, Alberta)

20

RR

2.23

1.30–3.84

Ritter et al. (1990)

Farm operators (spraying herbicides and insecticides)

Canada (Saskatchewan)

441

SMR

0.96

0.87–1.05

Saracci et al. (1991)a, e

Production workers or sprayers

10 Countries

30

SMR

1.11

0.751.58

Settimi et al.1998c

Greenhouse owners

Italy (Santa Marinella)

1

SMR

0.80

0.024.46

Sperati et al. (1999)

Licensed pesticide users

Italy (Lathium: Viterbo)

5

SMR

0.80

0.26–1.86

Swaen et al. (1992)c

Licensed herbicide applicators

The Netherlands

1

SMR

1.31

0.027.31

Torchio et al. (1994)

Licensed pesticide users

Italy (Piedmont: Asti, Alessandra, Cuneo)

66

SMR

0.96

0.74–1.22

Wiklund et al. (1989)a

Licensed pesticide applicators

Sweden

90

SIR

0.99

0.801.22

Zahm (1997)c

ChemLawn employees including applicators

USA/Canada (46 states and 3 Canadian provinces)

2

SMR

2.77

0.3110.00

Zhong and Ransson (1996)e

Licensed pesticide applicators

Iceland

10

SIR

0.70

0.331.29

Case–control studies

Aronson et al. (1996)

Pesticide exposure as occupational exposure

Canada (Montréal)

19

OR

1.60

0.91–2.81

Checkoway et al. (1987)

Pesticide exposure as occupational exposure

North Carolina (Chapel Hill)

5

OR

1.69

0.45–6.24

Ewings and Bowie (1996)

Farming activity: used pesticides

England and Wales (Somerset and east Devon)

15

OR

0.63

0.28–1.42

Fincham et al. (1992)

Farmers associated with insecticide exposure

Canada (Alberta)

? 5 cases or more

OR

0.73

0.54–0.98

Forastiere et al. (1993)d

Licensed farmers

Italy (Lathium: Viterbo)

5

OR:

  

Non-farmers basis

2.68

0.6710.71

Other farmers basis

2.13

0.53–8.50

Settimi et al. (2001)

Farmers mixing and applying pesticides

Italy (5 areas: Emilia Romagna: Imola; Piedmont: Asti; Toscana: Pescia, Pistoia, Grosseto)

51

OR

1.7

1.2–2.6

van der Gulden et al. (1995)

Farming activity: specific exposure to pesticides

The Netherlands

19

OR

0.93

0.41–2.12

Reasons for exclusion were: aRedundant study

bNo risk estimate available

cFewer than five cases reported

dWhen multiple RR estimates were given, we retained regional or provincial references instead of national reference and other farmers as comparison group instead of non-farmers

eRisk estimate available for companies providing fewer than 50% of workers classified as sprayers

The estimates of the RR for the selected pesticide-exposed groups of workers to develop or die from cancer of the prostate varied between 0.61 and 2.38 and included from five to 566 cases. Nine RR estimates reported a negative association between prostate cancer and the occupation, with one presenting a 95% CI that did not include 1. Thirteen RR estimates reported a positive association, seven of them presenting a 95% CI that did not include 1.

Figure 1 shows the 22 RR estimates versus publication date. Visual examination does not reveal any clear positive or negative trend with time. Figure 2 illustrates the funnel plot of ln (RR) versus 1/SE and reveals no systematic relationship between study size and magnitude of risk. The statistical test applied (Egger et al. 1997) does not produce evidence of funnel plot asymmetry (intercept 0.537; 95%CI −1.985 to 3.060) (P>0.25).
Fig. 1

Relationship between the estimate of relative risk and year of publication of studies on agricultural occupational groups involved with pesticide application and prostate cancer (CI confidence interval).

Fig. 2

Epidemiological studies of agricultural occupational groups involved with pesticide application and prostate cancer. Funnel plot of natural logarithms of RR estimates vs the inverse of their standard errors (1/SE) (lnRR for all studies combined =0.214).

Among the studies included in our meta-analysis, 45% (ten studies) are from the USA/Canada (Alavanja et al. 2003; Cantor and Silberman 1999; Fleming et al. 1999a, 1999b, 2003; Morrison et al. 1993; Ritter et al. 1990; Aronson et al. 1996; Checkoway et al. 1987; Fincham et al. 1992) and 55% (12 studies) are from Europe (Alberghini et al. 1991; Coggon et al. 1986; Dich and Wicklund 1998; Figa-Talamanca et al. 1993; Gambini et al. 1997; Kristensen et al. 1996; Sperati et al. 1999; Torchio et al. 1994; Ewings and Bowie 1996; Forastiere et al. 1993; Settimi et al. 2001; van der Gulden et al. 1995); 45% of studies show data on the incidence of cancer (Alavanja et al. 2003; Dich and Wicklund 1998; Fleming et al. 1999a; Kristensen et al. 1996; Aronson et al. 1996; Checkoway et al. 1987; Ewings and Bowie 1996; Fincham et al. 1992; Settimi et al. 2001; van der Gulden et al. 1995), and 55% show mortality rates (Alberghini et al. 1991; Cantor and Silberman 1999; Coggon et al. 1986; Figa-Talamanca et al. 1993; Fleming et al. 1999b, 2003; Gambini et al. 1997; Morrison et al. 1993; Ritter et al. 1990; Sperati et al. 1999; Torchio et al. 1994; Forastiere et al. 1993). Reference populations in cohort studies represented predominantly national, provincial or regional large populations. Control subjects in the case–control studies were chosen among benign prostatic hyperplasia cases, other hospital cancer cases and patients deceased from all causes selected from the general population. In all case–control studies, except one (Forastière et al. 1993), exposure assessment had been performed from questionnaires-based interviews or questionnaire mailings that collected information on work history and occupational exposure to pesticides. In the study by Forastiere et al. the identity of farmers was ascertained from the register of the local farmers’ pension fund and the identity of licensed pesticide users from a list provided by the provincial inspectorate of agriculture.

Table 2 summarises the results of the different meta-analyses performed and includes Woolf’s homogeneity chi-squared statistic and its P value. A strong heterogeneity existed among these 22 RR estimates, therefore further analyses were carried out pooling studies in function of stratification variables. The random effect procedure—applied in case of heterogeneity—on the 22 studies yielded a meta-RR estimate of 1.24 (95% CI 1.06–1.45).
Table 2

χ2 Woolf and P values for homogeneity, pooled estimates of prostate cancer risk and 95% CIs for several groupings of the data concerning agricultural occupational groups involved with pesticide application. The seven studies from Italy were not redundant as they did not concern the same regions of Italy and/or were not of the same design (cohort study or case–control study). Pooled rate ratios are in bold when the 95% CIs do not include 1. Fixed fixed-effects estimates, otherwise random-effects estimates, OR odds ratio

Grouping

No. of studies

Pooled

Homogeneity

Rate ratio

95% CI

χ2 Woolf

P

All studies:

22

1.24

1.06–1.45

158.924

3.4811×10−23

Study design

Cohort

 All studies

15

1.27

1.06–1.52

139.968

7.1993×10−23

  Mortality studies

11

1.21

0.92–1.59

56.731

1.497×10−8

  Incidence studies

4

1.37

1.03–1.81

67.595

1.3966×10−14

Case–control

 OR

7

1.15

0.77–1.72

16.932

0.0095

  Mortality study

1

2.13

0.53–8.50

  Incidence studies

6

1.10

0.72–1.68

15.929

0.00705

Geographic location:

     

Europe

 All studies

12

1.12

1.03–1.22

15.391

0.165 (fixed)

  Italian studies

7

1.05

0.88–1.26

10.474

0.106 (fixed)

  Non-Italian studies

5

1.14

1.04–1.25

4.311

0.366 (fixed)

USA/Canada

 All studies

10

1.40

1.09–1.80

138.239

2.3821×10−25

  Canadian studies

4

1.16

0.80–1.67

15.647

0.0013

  US studies

6

1.59

1.15–2.20

72.383

3.2681×10−14

   North Carolina, Iowa

2

1.14

1.05–1.24

0.343

0.558 (fixed)

   Florida

2

1.97

1.79–2.18

2.452

0.117 (fixed)

   All states

2

1.35

0.97–1.90

0.046

0.830 (fixed)

If study design is used as a stratification variable, the heterogeneity of the case–control studies is markedly reduced, with a pooled rate ratio of 1.15 (95% CI 0.77–1.72). This is not the case for the cohort studies. Combination of the cohort studies gives a pooled rate ratio of 1.27 (95% CI 1.06–1.52).

Stratification of these data by outcome yields no great differences between incidence and mortality rate ratios in cohort studies, and the heterogeneity remains great, although reduced when compared with all cohort studies combined. Both incidence and mortality rate ratios are above 1, but significant only for incidence studies.

Stratification of the data by geographic location (Europe or USA/Canada) reveals homogeneity for all European studies but not for all US/Canadian studies. If the grouping “Europe” is divided into Italian and non-Italian studies, homogeneity is revealed for each subgroup. The ranges of relative risk estimates were 0.61–2.13 and 0.63–1.51 for Italian and non-Italian studies, respectively. If the grouping USA/Canada is divided into US studies and Canadian studies, heterogeneity is reduced more markedly for Canadian studies than for US studies. The ranges of relative risk estimates were 0.73–2.23 and 1.14–2.38 for Canadian and US studies, respectively. The additional sub-grouping of US studies according to specific areas (Iowa, North Carolina, Florida) or covering the USA as a whole, reveals homogeneity in all subgroups. Overall, pooled rate ratios for studies derived from Europe are lower than those for studies derived from the USA/Canada. The pooled rate ratios for Italian studies (1.05, 95% CI 0.88–1.26) as well as for Canadian studies (1.16, 95% CI 0.80–1.67) are not significantly increased. In all others geographical groupings the pooled rate ratios are significantly increased.

Nine of the standardised mortality ratio (SMR) studies included in our meta-analysis also presented estimates of relative risk for all causes of death and for all neoplasms (Alberghini et al. 1991; Cantor and Silberman 1999; Figa-Talamanca et al. 1993; Fleming et al. 1999b, 2003; Gambini et al. 1997; Ritter et al. 1990; Sperati et al. 1999, Torchio et al. 1994). The pooling of the data from these studies reveals a significantly decreased risk of death from all cause (data not shown; pooled relative risk estimate 0.75; 95% CI 0.64–0.86). Among the nine risk estimates extracted from these studies, seven showed significantly reduced estimates of relative risk for all causes, whereas the remaining two were significantly increased. The pooled cancer mortality rate is significantly decreased for all cancers combined (data not shown; pooled relative risk estimate 0.79, 95% CI 0.71–0.88). Significant decreases were observed for seven studies, non-significant changes were observed for the estimate extracted from the cohort of rice growers (Gambini et al. 1997) and a borderline significant increase was observed for the study by Fleming et al. (2003).

Sensitivity analyses does not substantially alter the results of the meta-analysis (Table 3). Exclusion of the study with the smallest (Alberghini et al. 1991) or largest (Fleming et al. 1999b) estimate of RR, as well as exclusion of the studies with the smallest (Forastiere et al. 1993) or largest (Alavanja et al. 2003) precision (1/SE) makes no great difference. If the analysis is re-run to include the studies reporting fewer than five prostate cancer cases, there is little effect on the overall pooled rate ratio.
Table 3

Pooled estimates of prostate cancer risk: sensitivity analyses. Pooled rate ratios are in bold when the 95% CI do not include 1

Grouping

Number of studies

Pooled

Homogeneity

Rate ratio

95% CI

χ2 Woolf

P

All studies

22

1.24

1.06–1.45

158.924

3.4811×10−23

Deletion of studies with extreme estimators of RR values

 Largest value: Fleming et al. (1999b)

21

1.18

1.01–1.38

131.229

2.2913×10−18

 Smallest value: Alberghini et al. (1991)

21

1.27

1.08–1.49

154.954

7.0440×10−23

Deletion of studies with extreme precision values

 Largest value: Alavanja et al. (2003)

21

1.24

1.03–1.49

155.748

4.955×10−23

 Smallest value: Forastière et al. (1993)

21

1.23

1.05–1.44

158.297

1.6002×10−23

Inclusion of studies reporting fewer than five exposed casesa

25

1.24

1.06–1.46

159.881

4.7295×10−23

aThe studies included were Settimi et al. (1998), Swaen et al. (1992) and Zahm (1997)

Discussion

In a previous meta-analysis of 22 epidemiological studies published between 1995 and 2001, the meta-relative risk for prostate cancer in farmers was 1.13 (95% CI 1.04–1.22). Among these studies, four addressed the subgroup of pesticide applicators, and, here, a larger meta-relative risk was obtained (1.64, 95% CI 1.13–2.38) (Van Maele-Fabry and Willems 2003). The aim of the present study was to investigate whether this could be confirmed when we extended the time period of publications and focussed on pesticide applicators and other agricultural occupational groups potentially more exposed to pesticides than farmers in general. Based on 22 epidemiological studies published between 1986 and 2003, a meta-rate ratio of 1.24 (95% CI 1.06–1.45) is now obtained, confirming our previous result and providing additional evidence for an association between occupational groups exposed to pesticides and cancer of the prostate. The pooled rate ratio remains rather modest, indicating that the association is relatively weak.

As publication or related biases are common in meta-analyses, we applied conventional tests, e.g. the funnel plot and a statistical test, the linear regression method (Egger et al. 1997), and did not find an indication for publication bias. In meta-analyses, date of publication may serve as a surrogate for study quality or serve to detect temporal trends in exposure or disease incidence. Visual examination of Fig. 1 does not reveal any positive or negative trend in RR estimates with date of publication.

Whereas most likely more exposed to pesticides than farmers in general, the occupational categories studied remain heterogeneous and include various types of pesticide users: aerial pesticide applicators, licensed pesticide applicators, private and commercial pesticide applicators, seasonal herbicide applicators (rice growers), orchard and greenhouse workers, farming activities specifically mentioning pesticide use, and manufacturing workers of whom more than 50% are involved in pesticide spraying. They may be exposed to pesticides while transporting, mixing, loading or applying chemicals, through cleaning or repairing equipment or from re-entering treated fields (Dosemeci et al. 2002) and according to different scenarios, e.g. multiple compounds or formulations, simultaneously or successively, often with an intermittent frequency of application, and under a great variety of conditions (Maroni and Fait 1993), finally leading to quantitatively quite different exposures. This diversity of exposure scenarios is a major source of heterogeneity but cannot be taken into account because of lack of detailed exposure data in most studies. Two other sources of heterogeneity were, however, explored: study design and geographic location (Van Maele-Fabry and Willems 2003).

Study design

After stratification of the data by study design, heterogeneity is markedly reduced for only the case–control studies, not for the cohort studies (Table 2). In both, the data remain heterogeneous and, as a consequence, the pooled relative risks calculated have to be taken with caution. The increase in the pooled rate ratio is statistically significant only for the cohort studies combined. Stratification of the data by outcome only slightly reduces heterogeneity. Both incidence and mortality meta-rate ratios of cohort studies are increased. One possible explanation of the elevated prostate cancer risk in pesticide applicators is that these relatively healthy working populations are more likely to develop prostate cancer simply because they do not die of other competing causes of mortality so common in the general population, given the extremely high prevalence of prostate cancer in elderly men (Fleming et al. 1999a, 1999b). The meta-rate ratio for incidence cohort studies is significantly increased and is slightly greater than for mortality studies. This could be expected for prostate cancer, due to the slow progression of most prostate cancers, the availability of an easy diagnostic screening method (prostate-specific antigen) that allows the detection of localised prostatic adenocarcinoma, and the treatment available. Prostate cancer incidence must be interpreted in the context of diagnostic intensity and screening behaviour. Increased incidence rates in some countries (the USA as a prime example) reflect the sum of clinical disease and latent disease but in other countries includes only clinical disease (Signorello and Adami 2002). As a consequence, increased incidence rates may reflect both advances in early tumour detection and a true increased incidence (Aronson et al. 1996).

Geographic location

After stratification of the data by geographic location, heterogeneity is strongly reduced in most of the subgroups considered, except for the studies from North America. The pooled risk estimate for the North American studies is higher than those in studies from Europe. These results are in agreement with our previous observation (Van Maele-Fabry and Willems 2003). Nevertheless, it has to be stressed that when the meta-analysis is limited to agricultural occupational groups involved with pesticide application—as in the present study—an increased meta-rate ratio is also observed for studies derived from Europe. This was not the case in our previous meta-analysis for European studies.

Quite large differences in incidence rates of prostate cancer have been reported between different racial/ethnic groups and countries, with the lowest rates for Asians, intermediate rates for Europeans, high rates for US Caucasians and Canadians, and highest rates among US blacks (Hsing and Devesa 2001). The wide variations in the incidence of prostate cancer between ethnic populations and countries are caused by a combination of underlying differences, such as genetic susceptibility, exposure to unknown external risk factors, or artefactual reasons such as cancer registration and differences in health care (Grönberg 2003). The excess risk in US whites compared with Europeans has been reported to be associated with American life style factors—such as fat intake, obesity and sedentary habits—working through the hormonal or insulin-like growth factor pathways to influence the risk of prostate cancer (Hsing and Devesa 2001). Those authors consider that the differences in incidence rates between US whites and Europeans are unlikely to be explained entirely by more aggressive prostate-specific antigen screening, since higher rates in US whites were found long before PSA testing was available. Since the quality of medical care and cancer registration in most of the European countries is quite high, differences in quality of cancer registration and medical care also cannot explain the excess risk in US Caucasians (Hsing and Devesa 2001). As ethnic origin is considered to be a major risk factor for prostate cancer, it could be suggested that the high ethnic diversity that exists in North America is, at least, partially responsible for the difference in meta-risk ratio. However, it has been reported that in the USA, most of the high-risk populations working in field crops (the largest amount of farm acreage in the USA) are people of colour who are generally migrant and seasonal farm workers or non-certified pesticide applicators (Moses et al. 1993). As a consequence, most of those high-risk populations are not included in the US/Canadian studies selected for the present meta-analysis.

Further geographic stratifications (see Table 2) lead to a ranking of pooled rate ratios for prostate cancer among pesticide-exposed groups from low to high: Italian studies < European non-Italian studies < Canadian studies < US studies. This ranking happens to follow the same order as the ranking of incidence rates for prostate cancer in the overall population from those countries (see Grönberg 2003). The significance of this observation remains unclear, but it can be hypothesised that a population basically more susceptible to the disease will react more strongly to an external risk factor (possibly pesticide exposure) than less susceptible populations. An analogous observation is that a family history of prostate cancer—one of the more consistent risk factors for prostate cancer—appears to increase the risk for prostate cancer for those exposed to pesticides (Alavanja et al. 2003).

The significantly reduced death risk from all causes and the pooled cancer mortality for all cancers combined, that were observed after we had pooled the data from the SMR studies included in our meta-analysis, are in agreement with other studies of agricultural occupational groups (see, for discussion and references, Van Maele-Fabry and Willems 2003).

Whether pesticide exposure—a common factor in the different groups included in the present meta-analysis—means, indeed, causality, remains a difficult question to solve on the basis of the epidemiological studies available. Several criteria can be applied to shed some light on this matter (Hill 1965; Susser 1991; Koh and Seow 1999).

Strength of association

Strength of association remains weak.

Consistency from study to study

Conflicting data were reported between studies: nine RR estimates reported a negative association between prostate cancer and occupation, 13 reported a positive association. However in seven of the 13 the 95% CI did not include 1. This was only the case for one among the nine.

Consistency from meta-analysis to meta-analysis

The slightly higher overall risk for prostate cancer observed in the present meta-analysis (agricultural occupational groups potentially more exposed to pesticides than farmers in general), as compared to overall risks reported in the four meta-analyses performed on farmers and broader pesticide exposure categories, strengthens the suggestion that exposure to pesticides may be a causal factor.

Specificity

Current knowledge of the aetiology of prostate cancer is limited and remains speculative. Some factors are known (age, ethnicity and family history) while, for others, inconsistent results have been reported (e.g. dietary factors, sexual history, physical activity), and others remain probably unknown, which results in a web of causation. Prostate cancer is likely to be multifactorial in aetiology, and, thus, there is no one specific cause.

Dose–response relationship

Some of the studies gave some hint of a dose-response relationship. Risk increased among pesticide applicators who had their licenses issued in 1965–1966 compared to those who had their licenses issued later (1967–1976) (Dich and Wicklund, 1998); a statistical trend was found between the number of acres sprayed with herbicides and risk of prostate cancer (Morrison et al. 1993); within one cohort, a slight excess in mortality was observed in a cluster predominantly working on arable land (where the amount of herbicide used is greater) in comparison with a cluster that worked in predominantly woodland areas and mixed land (Torchio et al. 1994); in a case–control study the risk estimate increased with the number of years spent mixing and applying pesticides (Settimi et al. 2001). On the other hand, no exposure–response relationship was detected in aerial pesticide applicators when number of flight hours was uses as an index of exposure (Cantor and Silberman 1999). Paradoxical data have also been reported, e.g. for 2 methyl-4-chlorophenoxyacetic acid, a slight association was observed with grade of potential exposure but not by duration of exposure (Coggon et al. 1986); apparent trends of decreasing risk (prostate cancer incidence and mortality) were seen with calendar years from the earliest to the most recent years, which were no more observed (incidence) or reversed (mortality) when the number of years of licence was considered (Fleming et al. 1999a, 1999b).

Temporal relationship

In most studies start and duration of possible exposure remain uncertain and cannot be compared with the long latency for prostate cancer development.

Biological plausibility

Since most studies do not identify any specific pesticide or class of pesticides, only very general considerations can be presented here: endocrine disruption because of androgen—or oestrogen—like activities of pesticides or their metabolites (Santti et al. 1994; Fleming et al. 1999a; Golden et al. 1998; Rooney and Guillette 2000); stimulation of the progression of prostate cancer cells by different pesticides via activation of protein tyrosine kinase (Tessier and Matsumura 2001). There is increasing evidence that fat consumption and/or chemical contaminants in fat (notably organochlorine pesticides) may increase production and bioavailability of androgenic hormones or mimic their action (Golden et al. 1998).

Some authors tried to link specific pesticides to the development of prostate cancer. Associations have been observed with the organochlorine pesticides aldrin (Alavanja et al. 2003), DDT (Alavanja et al. 2003; Settimi et al. 2003), dicofol (Settimi et al. 2003), heptachlor (Alavanja et al. 2003; Mills and Yang 2003) and lindane (Mills and Yang 2003), the organophosphate pesticide dichlorvos (Mills and Yang 2003), the carbamate pesticide carbofuran (Alavanja et al. 2003), the pyrethroid pesticide permethrin (Alavanja et al. 2003), a fumigant, methyl bromide (Alavanja et al. 2003; Mills and Yang 2003), the triazine herbicides atrazine and simazine (Mills 1998; Mills and Yang 2003) and the phthalimide fungicide captan (Mills 1998).

It clearly appears from our analysis, and from that of others (Acquavella et al. 2003), that future epidemiological studies focusing on cancer and exposure to pesticides should undertake a major effort to assess the identity and the degree of pesticide exposure. In a pilot study, for example, using serum levels as biomarkers of exposure, it was suggested that exposure to PCB 180 and oxychlordane are each associated with an increased risk of prostate cancer (Ritchie et al. 2003). In addition, consideration of multiple exposures is important in the determination of realistic exposure scenarios and in the accurately estimation of specific effects. In most cases, limited study power has hindered investigation of the risk associated with common pesticide combinations (De Roos et al. 2003).

Conclusion

We found an increased meta-rate ratio of 1.24 (95% CI 1.06–1.45) for prostate cancer in agricultural occupational groups involved with pesticide application, based on epidemiological literature published between 1986 and 2003. This meta-rate ratio is in agreement with the significant increase observed for the occupation category of pesticide applicators in a meta-analysis previously performed and based only on four studies published between 1995 and 2001 (Van Maele-Fabry and Willems 2003). The results of the present meta-analysis, suggesting a weak increase in risk in pesticide applicators—who may be even more exposed to pesticides than others—tends to reinforce the evidence for a relationship between pesticide exposure and cancer of the prostate. Homogeneity between studies was reached after stratification of the studies by geographic location. The increased risk appears more marked in North America than in Europe. However, the studies that were reviewed contained insufficient qualitative and quantitative information on exposure for us to draw firm conclusions about pesticide exposure as the cause of prostate cancer. Nevertheless, the results obtained strengthen the need to minimise exposure towards pesticides, first of all for the applicators, by emphasising the use of correct application techniques.

Acknowledgments

G.V.M.F. was supported by a grant from the Ministry of Health.

Copyright information

© Springer-Verlag 2004