Cancer Causes & Control

, Volume 17, Issue 4, pp 353–373

Review and Meta-analysis of Risk Estimates for Prostate Cancer in Pesticide Manufacturing Workers

Authors

    • Unité de Toxicologie Industrielle et Médecine du travail, Ecole de Santé PubliqueUniversité Catholique de Louvain
  • Valérie Libotte
    • Unité de Toxicologie Industrielle et Médecine du travail, Ecole de Santé PubliqueUniversité Catholique de Louvain
    • Fonds des Maladies Professionnelles
  • Jan Willems
    • Department of Public HealthGhent University
  • Dominique Lison
    • Unité de Toxicologie Industrielle et Médecine du travail, Ecole de Santé PubliqueUniversité Catholique de Louvain
Review Article

DOI: 10.1007/s10552-005-0443-y

Cite this article as:
Van Maele-Fabry, G., Libotte, V., Willems, J. et al. Cancer Causes Control (2006) 17: 353. doi:10.1007/s10552-005-0443-y

Abstract

Purpose

The purpose of the present paper is to review cohort studies that examined the occurrence of prostate cancer in pesticide manufacturing workers in order to undertake a qualitative and quantitative evaluation of the risk as well as to assess the level of epidemiological evidence for each class of chemical compounds.

Methods

Following a systematic literature search, relative risk (RR) estimates for prostate cancer were extracted from 18 studies published between 1984 and 2004. All studies were summarised and evaluated for homogeneity and publication bias. As no significant heterogeneity was detected, combined RR estimators were calculated using a fixed effect model. Meta-analyses were performed both on the whole set of data and for each chemical class separately.

Results

The meta-rate ratio estimate for all studies was 1.28 [95% confidence interval (CI) 1.05–1.58]. After stratification by specific chemical class, consistent increases in the risk of prostate cancer were found in all groups but statistical significance was found only for accidental or non-accidental exposure to phenoxy herbicides contaminated with dioxins and furans. There was no obvious indication of publication bias.

Conclusion

The overall meta-analysis provides additional quantitative evidence consistent with prior reviews focusing on other groups exposed to pesticides (farmers, pesticide applicators). The results again point to occupational exposure to pesticides as a possible risk factor for prostate cancer but the question of causality remains unanswered. Epidemiological evidence did not allow identifying a specific pesticide or chemical class that would be responsible for the increased risk but the strongest evidence comes from workers exposed to phenoxy herbicides possibly in relation with dioxin and/or furan contamination.

Keywords

Prostatic neoplasmPesticidesMeta-analysisManufacturingOccupational exposureReview

Background

The increased incidence of cancer in recent years, coupled with the widespread use of pesticides has raised increasing concerns for public health. Human exposure to pesticides occurs during the manufacturing, in end-use applications (e.g. farmers and applicators) and in non-occupational settings via environmental release and groundwater supplies. Several epidemiological investigations have examined whether occupational exposure to pesticides is associated with prostate cancer. Reviews and meta-analyses focussing on farming, pesticide application and related occupational categories suggest an increased risk of prostate cancer [16] although others have found reduced or non-excess risks [79]. The most important etiologic factors for prostate cancer include age, race/ethnicity and family history [10]. Among others, pesticides have been suggested as a potential risk factor for prostate cancer but their role has been a subject of controversy [11]. In addition, there are large numbers of different pesticides and most of the studies were unable to identify the contribution of specific compounds.

Exposure to any particular pesticide occurs relatively infrequently in farmers and pesticide applicators, posing a challenge for identifying specific risk factors. It is therefore difficult to attribute any adverse effect observed in these occupational groups to a particular pesticide or group of pesticides.

The principal advantage of studying workers employed in the manufacture of pesticides rather than end-product users (farmers or pesticide applicators) stems from the fact that manufacturing workers may have a less diverse occupational environment and are likely to have been more frequently and potentially more intensely exposed, especially during the early years of production when industrial hygiene controls were not very sophisticated [12, 13]. Manufacturing workers may have had relatively high (cumulative) exposure, although no data are available to confirm these assumptions [14, 15]. The manufacturing environment also offers a different set of potential confounding exposures than those existing on the farm, thus providing a useful contrast to test the consistency of the observations [13]. Beside pesticides, employees engaged in production-related activities are potentially exposed to raw materials, intermediate products and end products [16].

The purpose of this paper is to review cohort studies that have examined the occurrence of prostate cancer in workers ever employed in pesticide manufacturing. We postulated that stratifying the studies according to chemical classes would allow to better target their effect on prostate cancer. It was also considered that a class-specific approach to evaluate pesticides as risk factors for prostate cancer should facilitate the interpretation of epidemiological studies for regulatory purposes.

Materials and methods

Study identification and selection

Study identification

A search on MEDLINE (National Library of Medicine, Bethesda, MD) was conducted for the period 1966 to 1 August 2004. The search strategy used several combinations of the following key words: prostatic neoplasms (MeSH), pesticides (MeSH), industry (MeSH), cancer, manufacturing workers, employees, triazines, organochlorines, halogenated hydrocarbon nematocides, phenoxy herbicides, chloroacetanilides. Recent articles in occupational medicine and epidemiology journals were also scanned for relevant publications. Finally, the reference lists of the relevant publications identified were checked for additional studies, limiting the search to cohort studies published in the open literature. Cohort studies were selected as they are the best observational design in order to assess causality of a disease. Published studies were used as they are likely to be more reliable than unpublished reports.

Study selection

All studies meeting the following inclusion criteria were taken into consideration for a first overall evaluation:
  • surveys published in English, in peer-reviewed journals between 1966 and 2004,

  • with a cohort design,

  • providing sufficient data to determine an estimator of relative risk (RR) for prostate cancer and its confidence interval,

  • referring to the occupational group of interest (workers engaged in the manufacture of pesticides).

Studies were excluded from the analysis if they

  • included subjects already included in another more complete or more recent study examining a greater number of subjects or with longer follow-up duration,

  • were case–control (due to the concern about quality of exposure information and biases in case or control selection and as case–control studies are less useful to explore causation) or proportional mortality ratios (PMR) studies (mainly due to ambiguities in interpreting results),

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

  • investigated women cohorts,

  • clearly examined a specific cancer type other than prostate cancer.

Data extraction

A structured abstract was derived for each study identified. Two authors read the reports and independently extracted and tabulated the most relevant RR estimators, with their 95% CIs. The results of this exercise were compared between the two authors and consensus was obtained before the meta-analysis.

If more than one follow-up analysis had been published for the same population, we used the most recently published report because it provided the longest follow-up and the most up-to-date information. Generally, when multiple estimates of RR were given, we retained the data on which the authors had relied for their assessment or the overall data for the total cohort and for the total follow-up period. Exceptions were papers reporting data for individual plants having specific exposures justifying their inclusion in different meta-analyses as detailed in the section ‘stratifying studies in the meta-analyses’. We did not include data resulting from further stratification, e.g., by latency period, by grade of potential exposure, by duration of potential exposure, by years since/of first exposure, by job title/employee group, by ethnicity. In some cases, we calculated a measure of RR and/or 95% CI if it had not been published but was calculable from the data reported [17].

Stratifying studies in the meta-analyses

Given the various pesticide classes for which cohorts were followed and the known contamination of some of them (e.g. phenoxy herbicides) with polychlorinated dibenzo-dioxins and polychlorinated dibenzo-furans (PCDDs/PCDFs), several groupings of the data were performed. An overall meta-analysis including all classes of pesticides was first conducted. Separate meta-analyses were conducted for the classes of pesticides other than phenoxy herbicides and for phenoxy herbicides only. Among the pesticides other than phenoxy herbicides, estimators of relative risk could be extracted for organochlorines, chloroacetanilides, triazines and halogenated hydrocarbon nematocides. Meta-analyses could be conducted for triazines and for halogenated hydrocarbon nematocides. Within the phenoxy herbicides, we tried to distinguish cohorts of workers unlikely to have been contaminated by PCDDs/PCDFs from those contaminated with higher chlorinated PCDDs/PCDFs. In several cohorts, workers were exposed to high concentrations of PCDDs/PCDFs as a result of an accident. Separate meta-analyses were performed for plants where there had been an accident (accident only) and for plants without accident. Other studies included workers exposed to PCDDs/PCDFs as a result of an accident and other exposed workers, and constitute the subgroup ‘including accident’.

Comments regarding some individual cohorts having specific exposures are required to justify the selected estimators of RR and their inclusion in different meta-analyses.

Among the studies dealing with halogenated hydrocarbon nematocides (pesticides others than phenoxy herbicides), Wong et al. [18] conducted a mortality study on workers potentially exposed to organic and inorganic brominated chemicals including chemicals other than pesticides (e.g. Tris(2,3-dibromopropyl) phosphate, polybrominated biphenyls). As a consequence, in our meta-analysis, we did not retain the overall data of the total cohort but only data concerning the subcohort of workers potentially exposed to the pesticide 1,2-dibromo-3-chloropropane (DBCP).

For phenoxy herbicides, there has been a series of publications on several plants from different countries which produced phenoxy herbicides contaminated or not with PCDDs/PCDFs. To prevent multiple appearances of the same cohort, studies were grouped so that specific populations could be traced from the earliest to the most recently published. They were classified according to the stratifying scheme previously described and are reported in Fig. 1. In the overall meta-analysis, all selected studies were included except those by Collins et al. [19], Ramlow et al. [20] and Bodner et al. [21] to avoid redundancy. As a consequence, the meta-rate ratio estimate for the overall meta-analysis was based on 16 studies from the 18 initially selected.
https://static-content.springer.com/image/art%3A10.1007%2Fs10552-005-0443-y/MediaObjects/10552_2005_0443_Fig1.jpg
Fig. 1

Cohort studies of workers manufacturing phenoxy herbicides classified according to their contamination or not with PCDDs/PCDFs and by countries. Note: Studies in bold are those included in the meta-analyses

The cohorts studied by Collins et al. [19] and by Ramlow et al. [20] are subsets of the larger study of United States production workers exposed to PCDDs (National Institute of Occupational Safety and Health—NIOSH-study) published by Fingerhut et al. [22] and updated by Steenland et al. [23]. This last comprehensive update was already included in the overall meta-analysis. The study by Collins et al. [19] was included in the meta-analysis concerning phenoxy herbicides contaminated, ‘accident only’ and that of Ramlow et al. [20] was included in the meta-analysis concerning phenoxy herbicides contaminated ‘without accident’.

Bodner et al. [21] is the last update of the mortality experience of Dow Chemical Company workers (USA) contributing to about 40% of the NIOSH study. The cohort of Bodner et al. [21] was not included in the overall meta-analysis. It was only included in the meta-analysis concerning phenoxy herbicides contaminated ‘without accident’ although a very small subset of this cohort (about 3%) includes workers accidentally exposed [24].

Hooiveld et al. [25] reported mortality rates among all workers exposed to phenoxy herbicides, chlorophenols and contaminants (including workers exposed as a result of the accident) and, separately, mortality rates among workers exposed as a result of the accident, only. These data were never presented together in a single meta-analysis. Mortality data in workers exposed as a result of the accident were included in the meta-analysis dealing with phenoxy herbicides contaminated, ‘accident only’. In the other meta-analyses (overall meta-analysis, phenoxy herbicides, phenoxy herbicides contaminated with higher chlorinated PCDDs/PCDFs), the data of all exposed workers were included.

Becher et al. [26] investigated the relation between exposure to phenoxy herbicides and cancer mortality in four manufacturing plants in Germany. The data of the four plants have been considered separately respecting the a priori decision of the authors that all plants have a specific exposure pattern. One of them (cohort I in the original paper) has been updated by Flesch-Janys et al. [27] and the updated data was included in the subgroup of phenoxy herbicides contaminated ‘without accident’. The data of the second plant (cohort II in the original paper) producing also 2,4,5-T, was not updated and was included in the subgroup of phenoxy herbicides contaminated ‘without accident’. In cohort III, no prostate cancer deaths were reported and the data could not be included in the meta-analyses. Finally, the last plant (cohort IV in the original paper) produced essentially 2,4-D and MCPA and was included in the group of phenoxy herbicides unlikely to have been contaminated by PCDDs/PCDFs.

The mortality in two cohorts of workers exposed to phenoxy herbicides and chlorophenols in The Netherlands was examined by Bueno de Mesquita et al. [28]. In one of the plants (factory A) an accident occurred. The cohort of accidentally exposed workers has been later followed up and updated data were published by Hooiveld et al. [25, 29]. The most recent data [25] have been included in the meta-analysis concerning phenoxy herbicides contaminated, ‘accident only’. In factory B [28] workers were unlikely to have been contaminated and data are included in the corresponding meta-analysis.

Data analysis

A detailed account of the procedure for data analysis has been published before [5]. In brief, homogeneity among studies was evaluated to test between-study comparability. The significance of the between-study variance was evaluated with the ln(RR) statistic test which has a χ2 distribution with degrees of freedom equal to the number of studies pooled minus 1. The applied formula is: χ2wi[ln(RR)i − ln(RR)p]2, for i=1–N, where N is the number of studies combined, RRp is the overall pooled RR estimate, RRi is the RR for the ith study and wi=1/Vi where Vi is the variance of the ln(RR)i. A low p value for this statistic indicates the presence of heterogeneity, which questions the validity of the pooled estimates [30, 31].

As no significant heterogeneity was observed, we calculated meta-rate ratios and CIs according to a fixed effect model [32] which assumes that results across studies differ only by sampling error. The study variance (Vi) was calculated, using the CI given, according to the equation Vi=[(ln(CIupper) − ln(CIlower))/3.92]2. As detailed by Stewart and collaborators [33] and Dennis [34] the maximum likelihood estimate of the pooled RR in the fixed effect model is the exp(ln(RR)p). The pooled ln(RR)p equals Σ [ln(RR)i/Vi]/[Σ(1/Vi)], Vi is the variance for an individual study as described above and ln(RR)i is the log RR estimate for study i. This is a variance-weighted least square mean. The variance of the pooled ln(RR)p, Var(ln(RR)p) or Vp is given by: [SE(ln(RR)p)] = [Σ(1/Vi)]−1 where SE is the standard error. The pooled variance is used to calculate a 95% CI around the pooled RR estimate.

We conducted sensitivity analysis including, in the overall meta-analysis: (a) deletion of studies reporting extreme RR estimators values [28 factory B or 35]; (b) deletion of studies reporting extreme precision (1/SE) values [23 or 35]; (c) deletion of studies reporting data from workers exposed as a result of an accident [23, 25, 36]; (d) deletion of incidence studies [3639]; (e) replacing, in the triazines meta-analysis, the study of MacLennan [16] by the data of Sathiakumar and coworkers reported by MacLennan [38].

Potential publication bias due to study size was explored by plotting the natural logarithm of the estimator of RR (lnRR) versus the inverse of standard error (1/SE) and funnel plot asymmetry was tested using the linear regression method of Egger and collaborators [40].

In order to determine whether any positive or negative trend had occurred with time, a plot of the estimators of RR versus publication date was made.

Results

Review of the literature

Among the retrieved scientific references, more than 100 dealt with manufacturing workers exposed to pesticides. After application of the exclusion criteria, 63 studies were selected and are reported in Table 1. Published data were available for organochlorines, chloroacetanilides, triazines, halogenated hydrocarbon nematicides and phenoxy herbicides. Forty-five additional studies were excluded from the analysis for the reasons mentioned in Table 1, leaving 18 studies that could be introduced in the different meta-analyses or in the sensitivity analyses.
Table 1

Industrial cohorts of pesticide manufacturing workers examining the risk of prostate cancer (PC)

Country

Industrial cohort

Main exposure

Author (ref.) Year

Comments

Class of pesticides

Organochlorines

USA

Plant 1: Illinois

Chlordane

Shindell and Ulrich [41] 1986*

? PC death reported among the 9 different cancer types accounting for 1 case?

 

Plant 2: Tennessee

Heptachlor, endrin

  
 

Plant 3: Denver, Colorado

Aldrin, dieldrin, endrin

Amoateng-Adjepong et al. [42] 1995*

? PC death included in ‘other cancers’? (26 other cancers/104 all cancers)

 

Plant 4: California

DDT

  
 

Plant 1 + Plant 2

DDT

Wang and MacMahon [43] 1979§

1 PC death reported but no expectedcases => no estimators of relative risk could be established

 

Plant 1 + Plant 2 + Plant 3+ Plant 4

DDT

Ditraglia et al. [44] 1981*,║

No PC death reported for plants 1, 2and 4; ? PC included in ‘other and unspecified malignant neoplasms’ for plant 3?

 

Plant 1 + Plant 2 + Plant 3+ Plant 4

DDT

Brown [45] 1992*

Update of Ditriglia et al. 1981; ? PCdeath included in ‘other neoplasms’? (29 other neoplasms/143 all neoplasms)

 

Michigan plant (A)

DDT

Wong et al. [18] 1984

No PC death observed among the 19cases of all cancer deaths

The Netherlands

Pernis cohort—Rotterdam

Aldrin, dieldrin, (telodrin, endrin)

Jager [46] 1970†,║

No PC deaths observed; 1 carcinomaof the stomach.

  

Aldrin, dieldrin, (telodrin, endrin)

Versteeg and Jager [47] 1973†,║

Update of Jager, 1970; no PC deathobserved; no additional cancer death observed

  

Aldrin, dieldrin, (telodrin, endrin)

Van Raalte [48] 1977†,║

Update; no PC observed out of 2 reportedcancer deaths

  

Aldrin, dieldrin, (telodrin, endrin)

Ribbens [49] 1985†,║

Update; no PC observed among the 9cancer deaths

  

Aldrin, dieldrin, (telodrin, endrin)

de Jong [50] 1991

Update; 1 PC observed/1.4 expected (SMR)

  

Aldrin, dieldrin, (telodrin, endrin)

de Jong et al. [51] 1997

Update; 1 PC observed/2.4 expected (SMR)

  

Aldrin, dieldrin, (telodrin, endrin)

Swaen et al. [35] 2002

Update; 1 PC observed/5.78 expected (SMR)

Chloroacetanilides

USA

Monsanto plant: Muscatine (Iowa)

Alachlor

Leet et al. [14] 1996†,§,║

0 observed/0.29 expected (SIR)

  

Alachlor

Acquavella et al. [52] 1996†,§

Update of Leet et al. 1996; 0observed/0.2 expected (SMR); 0 observed/0.7 expected (SIR)

  

Alachlor

Acquavella et al. [37] 2004

Update of Leet et al. 1996 and ofAcquavella et al. 1996; 0 observed/0.6 expected (SMR); 4 observed/3.5 expected (SIR)

Triazines

USA

Plant 1: Alabama

Triazines

Sathiakumar et al. [53] 1992†,║

No PC death observed among the 54 reportedcancer cases

 

Plant 2: Louisiana

Atrazine

MacLennan et al. [38] 2002

11 observed/6.3 expected (SIR)

  

Atrazine

MacLennan et al. [16] 2003

1 observed/0.5 expected (SMR)

 

Plant 1 + Plant 2

Triazines

Sathiakumar et al. [54] 1996†,§,║

0 observed PC/1.18 expected (SMR)

 

Plant 1 + Plant 2

Triazines

Sathiakumar and Delzell [15]1997

Update of Sathiakumar et al. 1996; no dataconcerning PC death were reported in the original paper but 3 observed PC/3.1 expected reported by MacLennan et al. 2002 as to be observed by Sathiakumar et al.

Halogenated hydrocarbon nematocides

USA

Down chemical company,Michigan division

DBCP

Hearn et al. [55] 1984

2 observed PC/0.4 expected (SMR)

  

DBCP

Olsen et al. [56] 1995

Update of Hearn et al. 1984; 2 observed PC/1.0 expected (SMR)

 

Velsicol chemical corporation: 3 manufacturing plants (2 Michigan, 1 Arkansas) + 1 research establishment

DBCP

Wong et al. [18] 1984

1 observed PC/0.71 expected (SMR)

Phenoxy herbicides

International

20 cohorts from 10 countries

Phenoxy herbicidesand/or chlorophenol

Saracci et al. [57] 1991

First follow-up of the IARC study; 30 observed PC/27.02 expected (SMR)

  

Phenoxy herbicides and/or chlorophenol

Kauppinen et al. [58, 59] 1993, 1994*

Exposure data and related information obtainedfor the cohorts participating in the IARC study

 

36 cohorts from 12 countries

Phenoxy herbicides and/or chlorophenol

Kogevinas et al. [60] 1997**

Update of Saracci et al. 1991 + US studies(Fingerhut et al. 1991) + studies from Germany (Becher et al. 1996); 68 observed PC/61.81 expected (SMR)

Denmark

4 factories: 1: Kemisk VaerkKoege, 2: Esbjerg Kemikaliefabrik (3: Cheminova, 4: Danske Gasvaerkers Tjaerekompagni)

Factory 1: 2,4-D; 2,4-DP; MCPA; MCPP; (2,4,5-T very limited quantity) Factory 2: MCPA; 2,4-D (Factories 3 and 4: MCPA)

Lynge [61] 1985

Factories 3 and 4 not considered suitable for inclusion in a cohort analysis; 9 observed PC/10.86 expected (SIR)

 

Factory 1 + factory 2

Factory 1: 2,4-D; 2,4-DP; MCPA; MCPP; (2,4,5-T very limited quantity) Factory 2: MCPA; 2,4-D

Lynge [62] 1987*

Background and design of the cohort

 

Factory 1 + factory 2

Factory 1: 2,4-D; 2,4-DP; MCPA; MCPP; (2,4,5-T very limited quantity) Factory 2: MCPA; 2,4-D

Lynge [63] 1993*

Update of Lynge, 1985, 1987; focus on STS and NHL, no data reported for PC

 

Factory 1 + factory 2

Factory 1: 2,4-D; 2,4-DP; MCPA; MCPP; (2,4,5-T very limited quantity) Factory 2: MCPA; 2,4-D

Lynge [39] 1998

Update of Lynge, 1993; 15 observed PC/14.99expected (SIR)

United Kingdom

6 British companies:Companies 1 and 2, Factories A, B, C, D

   
 

Company 1

MCPA

Coggon et al. [64] 1986

Company which has both manufactured and sprayed MCPA; 18 observed PC/13.51 expected (SMR)

 

Company 2

Dioxins

Bishop and Jones [65] 1981*

Focus only on NHL

 

Factory A

2,4,5-T; 2,4-D; 2,4-DP; 2,4-DB; MCPA; MCPP; MCPB; PCPA; PAA

Coggon et al. [66] 1991*

No data reported for PC (focus on other cancer types).

 

Factory B

2,4,5-T; 2,4-D; 2,4-DP; 2,4-DB;MCPA; MCPP; MCPB;

Coggon et al. [66] 1991*

No data reported for PC (focus on other cancer types).

 

Factory C

2,4,5-T; 2,4-D; MCPB; PBA

Coggon et al. [66] 1991*

No data reported for PC (focus on other cancer types).

 

Factory D

2,4,5-T; 2,4-D; 2,4-DP; MCPA;MCPP; 2,4,6-TCP

Coggon et al. [66] 1991*

No data reported for PC (focus on other cancer types).

The Netherlands

2 factories among 10 eligible companies: A: accident in 1963 B: no accident

   
 

Factory A + Factory B

Factory A: 2,4,5-T; 2,4,5-TCP(PCDDs) Factory B: MCPA; MCPP; 2,4-D

Bueno de Mesquita et al.[28] 1993

A: accident in 1963; 2 observed PC/0.93 expected (SMR); B: no accident, 1 observed PC/0.21 expected (SMR)

 

Factory A

2,4,5-T; 2,4,5-TCP (PCDDs)

Hooiveld et al. [29] 1996*,

Update of Bueno de Mesquita et al. 1993, factory A; no data reported for PC (focus on other cancer types).

 

Factory A

2,4,5-T; 2,4,5-TCP (PCDDs)

Hooiveld et al. [25] 1998

Update of Hooiveld et al. 1996 and Bueno deMesquita et al. 1993, factory A; All workers exposed (549): 4 observed PC/1.81 expected (SMR)

   

Hooiveld et al. [25] 1998

Only workers exposed as a result of the accident(140): 3 observed PC/0.577 expected (SMR)

Germany

4 Plants: Plant I: BoehringerIngelheim (Hamburg); Plant II: Bayer (Uerdingen); Plant III: Bayer (Dormagen); Plant IV: BASF (Ludwigshafen): 2 cohorts: A without accident and B with accident in 1953

Plant I: 2,4,5-T; 2,4,5-TCP; 2,5-DCP; Plant II: 2,4,5-TCP; Plant III: 2,4,5-T; 2,4,5-TP; 2,4-D; 2,4-DP; 2,4-DCP; MCPA; MCPP; Plant IVA: MCPA; MCPP; 2,4-D; 2,4-DP; 2,4-DCP (formulation of purchased 2,4,5-T)

  
 

Plant I + II + III + IV.A.

See above

Becher et al. [67] 1992*

Design and first results; no data on PC reported

 

Plant I

2,4,5-T; 2,4,5-TCP; 2,5-DCP

Manz et al. [68] 1991

7 observed PC/4.9 expected (SMR)

  

2,4,5-T; 2,4,5-TCP; 2,5-DCP

Flesch-Janys et al. [69] 1995*,║

Update of Manz et al. 1991; no data reported forPC (focus on all cancer).

  

2,4,5-T; 2,4,5-TCP; 2,5-DCP

Becher et al. [26] 1996

Update of Manz et al. 1991, Becher et al. 1992,Flesch-Janys et al. 1995; 7 observed PC/4.8 expected (SMR)

  

2,4,5-T; 2,4,5-TCP; 2,5-DCP

Becher et al. [70] 1998*,║

Cohort previously studied by Flesch-Janys et al.1995, Becher et al. 1996 and Manz et al. 1991; no data reported for PC; dose–response analysis for total cancer mortality—quantitative risk assessment

  

2,4,5-T; 2,4,5-TCP; 2,5-DCP

Flesch-Janys et al. [27] 1998

Update of Manz et al. 1991, of Flesch-Janys et al. 1995, of Becher et al. 1996, 1998; 9 observed PC/6.14 expected (SMR)

 

Plant II

2,4,5-TCP

Becher et al. [26] 1996

1 observed PC/0.7 expected (SMR)

 

Plant III

2,4,5-T; 2,4,5-TP; 2,4-D; 2,4-DP; 2,4-DCP; MCPA; MCPP

Becher et al. [26] 1996§

0 observed PC/0.2 expected (SMR)

 

Plant IV.A.

MCPA; MCPP; 2,4-D; 2,4-DP; 2,4-DCP (formulation of purchased 2,4,5-T)

Becher et al. [26] 1996

1 observed PC/1.5 expected (SMR)

 

Plant IV.B. accident

2,4,5-T; 2,4-D; 2,4-DP; MCPA; MCPP; Accidental release of 2,3,7,8-TCDD

Thiess and Frentzel [71] 1977;Thiess et al. [72] 1982

Thiess et al. 1982: update of Thiess and Frentzel 1977

  

2,3,7,8-TCDD

Zober et al. [73] 1990§

Partial update of Thiess et al. 1982; C1: basic cohort 1954; 0 observed PC/0.49 expected (SMR); C2: additional cohort 1983; 0 observed PC/0.39 expected (SMR); C3: additional cohort 1987; 0 observed PC/0.42 expected (SMR)

  

2,3,7,8-TCDD

Zober et al. [74] 1994*

No data reported for PC (focus on other diseases)

  

2,3,7,8-TCDD

Ott and Zober [36] 1996

Update of Zober et al. 1990; 0 observed PC/31malignant neoplasms (SMR); 4 observed PC/3.64 expected (SIR)

   

Zober et al. [75] 1997–1998

Redundant study; same data reported for PC asthose reported by Ott and Zober 1996

USA

NIOSH study 12 USA plantsincluding Nitro plant (about 9% of the cohort) and Midland plant (about 40% of the cohort)

Chemicals known to becontaminated with 2, 3, 7, 8-TCDD (TCP and 2, 4, 5-T and derivatives)

Fingerhut et al. [22] 1991

17 PC observed/13.9 expected (SMR)

  

TCDD-contaminated products

Steenland et al. [23] 1999

Update of Fingerhut et al. 1991; 28 observed PC/23.93 expected (SMR)

 

Nitro Plant (MonsantoCompany) + accident

TCDD; accident

Zack and Suskind [76] 1980

Men + chloracne; 9 malignant neoplasms, 0genitourinary organs

  

TCP, 2,4,5-T, potentialexposure to TCDD

Zack and Gaffey [77] 1983*

Partial update of Zack and Suskind, 1980; totalNitro Plant population; genitourinary organs others than bladder? including PC? 3 observed deaths/2.84 expected

  

TCDD; accident

Collins et al. [19] 1993

Update of Zack and Suskind, 1980; workersbetween accident and last reported acne case; 9 observed PC/5.625 expected (SMR)

 

Midland Plant (Dow ChemicalCompany)

2,4-D (and TCDD forsubcohort exposed between 1945 and 1983)

Bond et al. [78] 1988

1 observed PC/0.96 expected (SMR)

  

2,4-D (and TCDD forsubcohort exposed between 1945 and 1983)

Bloemen et al. [13] 1993*,║

Update of Bond et al. 1988; 2 additionalmalignant neoplasms among which 1 additional neoplasm of the genitourinary system? PC?

  

2,4-D (and TCDD forsubcohort exposed between 1945 and 1983)

Burns et al. [79] 2001

Update of Bloemen et al. 1993; 7 observedPC/5.2 expected (SMR)

  

PCP

Ramlow et al. [20] 1996

Update of a portion of the cohort of Ott et al.1987; 3 observed PC/3.8 expected (SMR)

  

2,3,7,8-TCDD (TCPproduction; accident in 1964)

Cook et al. [24] 1980†,║

3 deaths due to malignant neoplasms, 0 PC

  

2,4,5-T

Ott et al. [80] 1980†,║

1 single malignancy: respiratory malignancy

  

Higher chlorinated phenols andderivatives products (TCP, 2,4,5-T, Silvex, Ronnel, Erbon, chlorophenol)

Cook et al. [81] 1986

Partial update of Ott et al. 1980 and Cook et al.1980; 6 observed PC/3.19 expected (SMR)

  

Higher chlorinated phenols andderivatives products (TCP, 2,4,5-T, Silvex, Ronnel, Erbon, chlorophenol)

Ott et al. [82] 1987

Update of Cook et al. 1986; 8 observed PC/4.2expected (SMR)

  

Higher chlorinated phenols andderivatives products (TCP, 2,4,5-T, Silvex, Ronnel, Erbon, chlorophenol)

Bond et al. [83] 1989*,║

Update of Ott et al. 1987; PC data not shown butexposed workers experienced rates of PC significantly higher than those of unexposed workers

  

Higher chlorinated phenols andderivatives products (TCP, 2,4,5-T, Silvex, Ronnel, Erbon, chlorophenol)

Bodner et al. [21] 2003

Update of Bond et al. 1989; number of PC notreported but must be higher than 8; SMR=1.7 (95% CI=1.0–2.6)

Studies that were finally included in the meta-analyses are in bold as well as subcohorts or individual plants having specific exposures justifying their inclusion in different meta-analyses. Reasons for exclusion of studies were:* existence of prostate cancer could not be deduced (extracted) from the data; no prostate cancer cases were observed; study not included in this form (e.g. data not reported by the author in the original paper but mentioned by other authors; combined data from several plants);§ no estimator of relative risk could be calculated; results updated in later reports/redundant study;** study including also sprayers (applicators)

Abbreviations: PC: prostate cancer; SMR: standardized mortality ratio; SIR: standardized incidence ratio; DDT: dichlorodiphenyltrichloroethane; DBCP: 1,2-dibromo-3-chloropropane; 2,4-D: 2,4-dichlorophenoxyacetic acid; 2,4-DP: 2,4-dichlorophenoxypropionic acid; 2,4-DB: 2,4-dichlorophenoxybutyric acid; MCPA: 4-chloro-2-methylphenoxyacetic acid; MCPP: 4-chloro-2-methylphenoxypropionic acid; MCPB: 4-chloro-2-methylphenoxybutyric acid; 2,4,5-T: 2,4,5-trichlorophenoxyacetic acid; PBA: phenoxybutyric acid; 2,4,6-TCP: 2,4,6-trichlorophenol; 2,4,5-TCP: 2,4,5-trichlorophenol; 2,5-DCP: 2,5-dichlorophenol; 2,4-DCP: 2,4-dichlorophenol; 2,4,5-TP: 2,4,5-trichlorophenoxypropionic acid; 2,3,7,8-TCDD: 2,3,7,8-tetrachlorodibenzo-p-dioxin; PCDDs/PCDFs: polychlorinated dibenzo-dioxins and polychlorinated dibenzo-furans

Meta-analysis

Eighteen studies, contributing a total of 20 RR estimators met the inclusion criteria and were taken into consideration. Details on these studies included are summarized in Table 2. Ten cohorts followed workers from USA and 10 followed European workers.
Table 2

Abstracted risk estimates and study information from the included studies relating to prostate cancer in manufacturing workers

Author (ref.) year

Main typeof compound

Geographiclocation

Timeperiod

N. of cases

Specifiedmeasure

Estimator ofrelative risk

95% CI

Class of compound

Pesticides others than phenoxy herbicides

Organochlorines

Swaen et al. [35] 2002

Dieldrin and aldrin

The Netherlands

1954–2001

1

SMR

0.17

0.002–0.87

Chloroacetanilides

Acquavella et al. [37] 2004

Alachlor

USA (Muscatine, Iowa)

1968–1999

4

SIR

1.15

0.31–2.95

Triazines

MacLennan et al. [38] 2002

Triazines

USA (Louisiana)

1985–1997

11

SIR

1.75

0.87–3.12

MacLennan et al. [16] 2003

Triazines

USA (Louisiana)

1970–1997

1

SMR

2

0.05–11.14*

Halogenated hydrocarbon nematocides

Olsen et al. [56] 1995

DBCP

USA (Michigan)

1957–1989

2

SMR

1.96

0.24–7.08

Wong et al. [18] 1984

DBCP

USA (Michigan)

1935–1976

1

SMR

1.42

0.02–7.88

Phenoxy herbicides

Unlikely to have been contaminated by PCDDs (2,4-D; MCPA)

Lynge [39] 1998

MCPA

Denmark

1947–1993

15

SIR

1.00

0.6–1.7

Coggon et al. [64] 1986

MCPA

United Kingdom

1947–1983

18

SMR

1.33

0.79–2.11

Bueno de Mesquita et al.[28] 1993, factory B

MCPA/2,4-D

The Netherlands

1965–1986

1

SMR

4.76

0.12–26.53

Becher et al. [26] 1996, plant IVA

MCPA/2,4-D

Germany (Ludwigshafen)

1952–1989

1

SMR

0.67

0.02–3.73

Contaminated with higher chlorinated PCDDs

Without accident

Ramlow et al. [20] 1996

PCP (without 2,4,7,8-TCDD)

USA (Michigan)

1940–1989

3

SMR

0.79

0.16–2.31

Burns et al. [79] 2001

2,4-D (and TCDD for subcohortexposed between 1945 and 1983)

USA (Midland, Michigan)

1945–1994

7

SMR

1.34

0.54–2.77

Bodner et al. [21] 2003

TCDD

USA (Midland, Michigan)

1940–1994

? more than 8

SMR

1.7

1.0–2.6

Flesch-Janys et al. [27] 1998

Phenoxy herbicides and dioxins

Germany (Ingelheim)

1952–1992

9

SMR

1.47

0.67–2.78

Becher et al. [26] 1996, plant II

Phenoxy herbicides and dioxins

Germany (Uerdingen)

1952–1992

1

SMR

1.53

0.04–8.53

Accident only

Collins et al. [19] 1993

2,3,7,8-TCDD in a trichlorophenolprocess accident

USA (Nitro, West Virginia)

1949–1987?

9

SMR

1.6

0.7–3.0

Ott and Zober [36] 1996

Workers exposed to 2,3,7,8-TCDDas a result of an accident

Germany (Ludwigshafen)

1953–1992

4

SIR

1.1

0.3–2.8

Hooiveld et al. [25] 1998

Workers exposed to phenoxy herbicides,chlorophenols and contaminants as a result of the accident

The Netherlands

1955–1991

3

SMR

5.2

1.1–15.3

Including accident

Hooiveld et al. [25] 1998

All workers exposed to phenoxy herbicides,chlorophenols and contaminants

The Netherlands

1955–1991

4

SMR

2.2

0.6–5.7

Steenland et al. [23] 1999

TCDD

USA (12 plants)

1942?–1993

28

SMR

1.17

0.78–1.69

* SMR and 95% CI calculated based on the observed and expected prostate cancer deaths [17]

The estimators of RR for the manufacturing workers to develop or die from cancer of the prostate varied between 0.17 and 5.2 and included from 1 up to 28 cases. Sixteen RR estimators reported a positive association between prostate cancer and the occupation, with 2 presenting a 95% CI that did not include 1. Three RR estimators reported a negative association, 1 of them presenting a 95% CI that did not include 1. One study reported no association with a RR of 1. Four were incidence rate ratios and 16 were mortality rate ratios.

Table 3 summarises the results of the meta-analyses performed and includes the Woolf’s homogeneity χ2 statistic and its p value. No evidence of heterogeneity existed among the different combinations of RR estimators performed.
Table 3

Prostate cancer in pesticide manufacturing workers: meta-analysis results according to pesticide class stratification

Groups

N. studies

Pooled

Homogeneity

Rate ratio

95% CI

χ2 Woolf

p-value

(a) Overall MA

16

1.28

1.05–1.58

6.379

0.973

(b) Non-phenoxy

6

1.52

0.92–2.52

2.549

0.769

(c) Triazines

2

1.76

0.95–3.28

0.009

0.925

(d) Halogenated hydrocarbon nematocides

2

1.81

0.42–7.90

0.034

0.854

(e) All phenoxy

10

1.24

0.99–1.55

3.306

0.951

(f) Phenoxy unlikely contaminated

4

1.18

0.83–1.67

1.822

0.610

(g) Phenoxy contaminated (all)

6

1.29

0.96–1.72

1.340

0.931

(h) Phenoxy contaminated, accident only

3

1.80

1.03–3.13

3.343

0.188

(i) Phenoxy contaminated, without accident

5

1.50

1.06–2.11

1.226

0.874

N. studies=number of studies, CI=confidence interval, statistically significant pooled rate ratios are in bold

Studies included in the meta-analyses: (a) Overall MA: [16, 18, 23, 25 all exposed workers, 26 plants II & IVA, 27, 28 factory B, 3539, 56, 64, 79]; (b) non-phenoxy herbicides: [16, 18, 35, 37, 38, 56]; (c) triazines: [16, 38]; (d) halogenated hydrocarbons nematocides: [18, 56]; (e) all phenoxy herbicides: [23, 25 all exposed workers, 26 plants II & IVA, 27, 28 factory B, 36, 39, 64, 79]; (f) phenoxy unlikely contaminated: [26 plant IVA, 28 factory B, 39, 64]; (g) phenoxy contaminated (all): [23, 25 all exposed workers, 26 plant II, 27, 36, 79]; (h) phenoxy contaminated, accident only: [19, 25 workers exposed as a result of the accident, 36]; (i) phenoxy contaminated, without accident: [20, 21, 26 plant II, 27, 79]

The fixed effects procedure applied on the 16 studies on workers ever employed in a pesticide manufacture and potentially exposed to pesticides, included in the overall meta-analysis yielded a meta-rate ratio of 1.28 (95% CI: 1.05–1.58). A forest plot of these 16 studies is reported in Fig. 2. The figure also displays the weights applied in each study result in the overall meta-analysis. The study of Steenland and collaborators [23] contributed 47% of the total weight. Removal of this study from the overall meta-analysis resulted in a meta-rate ratio of 1.33 (95% CI: 1.05–1.69). No other studies contributed more than 30% of the total weight.
https://static-content.springer.com/image/art%3A10.1007%2Fs10552-005-0443-y/MediaObjects/10552_2005_0443_Fig2.gif
Fig 2

Forest plot of studies on prostate cancer among manufacturing workers exposed to pesticides. Note: Estimators of RR (SMR or SIR) and 95% confidence intervals (CIs) of studies included in the overall meta-analysis are presented. Each SMR and SIR was assigned a weight (wi) equal to the inverse square of its standard error (SE): wi=1/(SE)2

Fig. 3 shows the 16 RR estimators included in the overall meta-analysis versus publication date. Visual examination does not reveal any clear positive or negative trend with time. Fig. 4 illustrates the funnel plot of ln(RR) versus 1/SE and does not reveal a systematic association between study size and the magnitude of risk. The statistics did not yield evidence of asymmetry (intercept 0.612; 95% CI: −0.284 to 1.509) (p = 0.20) [40].
https://static-content.springer.com/image/art%3A10.1007%2Fs10552-005-0443-y/MediaObjects/10552_2005_0443_Fig3.gif
Fig. 3

Relation between the estimator of relative risk and year of publication of the 16 studies included in the overall meta-analysis and prostate cancer. Note: CI=confidence interval

https://static-content.springer.com/image/art%3A10.1007%2Fs10552-005-0443-y/MediaObjects/10552_2005_0443_Fig4.jpg
Fig. 4

Epidemiological studies of manufacturing workers exposed to pesticides and prostate cancer: funnel plot of natural logarithms of relative risk (RR) estimates versus the inverse of their standard errors (1/SE) (lnRR for the 16 studies combined=0.251)

After grouping the data into broad classes of pesticides (Table 3), increased pooled rate ratios were observed for each group with a borderline statistical significance for all phenoxy herbicides (pooled RRs: 1.24; 95% CI: 0.99–1.55). Within this last group, statistically significant increases were observed for the workers contaminated with PCDDs/PCDFs as a result of a plant accident (pooled RR: 1.80; 95% CI: 1.03–3.13) as well as for workers who did not experience an accident (pooled RR: 1.50; 95% CI: 1.06–2.11).

Most of the sensitivity analyses did not substantially alter the results of the meta-analysis (Table 4). In the overall meta-analysis, exclusion of the studies with the lowest or highest estimator of RR, exclusion of the studies with the lowest or highest precision (1/SE), deletion of studies concerning workers exposed as a result of an accident as well as deletion of incidence studies did not markedly change the pooled rate ratios. In the triazines meta-analysis, rerunning the analysis by replacing the study of MacLennan et al. [16] by the data of Sathiakumar and coworkers reported by MacLennan et al. [38] did not substantially modify the results.
Table 4

Pooled estimates of prostate cancer risk: sensitivity analyses

Grouping

N. studies

Pooled

Homogeneity

Rate ratio

95% CI

χ2 Woolf

p-value

Overall meta-analysis

16

1.28

1.05–1.58

6.379

0.973

Deleting studies

With extreme estimators of relative risk values

    Highest value

15

1.28

1.04–1.57

5.469

0.978

Bueno de Mesquita et al. [28] 1993 (factory B)

     

    Lowest value

15

1.30

1.06–1.59

4.668

0.990

Swaen et al. [35] 2002

     

With extreme precision values

    Highest value

15

1.33

1.05–1.69

6.066

0.965

Steenland et al. [23] 1999

     

    Lowest value

15

1.30

1.06–1.59

4.668

0.990

Swaen et al. [35] 2002

     

Cohorts of workers exposed as a result of an accident*

15

1.37

1.10–1.70

6.726

0.945

Incidence studies

12

1.32

1.03–1.69

4.444

0.955

Triazines meta-analysis

2

1.76

0.95–3.28

0.009

0.925

Replacing MacLennan et al. [16] 2003 study by the dataof Sathiakumar et al. reported by MacLennan et al. [38] 2002

2

1.57

0.88–2.78

0.618

0.432

N. studies=number of studies; CI=confidence interval; statistically significant pooled rate ratios are in bold

* and : Studies included in the meta-analyses: * Overall meta-analysis less cohorts of workers exposed as a result of an accident: [16, 18, 20, 21, 26 plants II & IVA, 27, 28 factory B, 35, 37, 38, 39, 56, 64, 79]

Overall meta-analysis less incidence studies: [16, 18, 23, 25 all exposed workers, 26 plants II & IVA, 27, 28 factory B, 35, 56, 64, 79]

Discussion

Concern about the potential carcinogenicity of pesticides has been prompted by epidemiologic studies reporting a possible association between farming or pesticide application and certain types of cancers, including prostate cancer. Several epidemiologic studies also examined prostate cancer risk among pesticide manufacturing workers but none of these individual studies allows to draw definitive conclusions. The present study is the first comprehensive meta-analysis of prostate cancer risk in workers from the pesticide chemical industry.

Results of the meta-analysis

The overall meta-RR after pooling 16 studies shows a statistically significant 28% increased risk of prostate cancer as a result of working in a pesticide manufacturing environment. This result is consistent with several prior reviews and meta-analyses dealing with other occupational groups exposed to pesticides (farmers, pesticide applicators), which have reported a slightly increased risk for prostate cancer [16].

As pesticides are an heterogeneous group of chemicals, differentiating the studies according to chemical classes provides an opportunity to estimate the effect of specific pesticides on prostate cancer. This approach may also facilitate the interpretation of the data for regulatory purposes. However, it appeared that only scarce epidemiological data were available for each specific class. The results of the present study did not allow identifying one or several classes that would be responsible for the increased risk of prostate cancer among manufacturing workers. All stratified analyses showed increased risk for prostate cancer but none was significantly increased, except for phenoxy herbicides (borderline statistical significance for all phenoxy). Commercial preparations of phenoxy acids and chlorophenols can be contaminated by dioxins and furans formed during their manufacture. The concomitant exposures to phenoxy herbicide(s), chlorophenol(s) and contaminant(s) present a special problem of potential confounding in the epidemiological literature. The focus of several studies has even been on the likely dioxin contaminants rather than on phenoxy herbicides themselves. The relatively high and statistically significant increases observed for the workers exposed to contaminated phenoxy herbicides as a result of an accident or not, as compared to the slight and non-significant increase observed for the workers exposed to phenoxy herbicides unlikely to have been contaminated with PCDDs/PCDFs, suggest a significant contribution of these contaminants. The non-significantly increased pooled rate ratio (1.29; 95% CI: 0.96–1.72) obtained for the six studies including workers exposed to contaminated phenoxy herbicides (all) may appear in contradiction with significant increases observed separately for workers exposed as a result of an accident (pooled RR: 1.80; 95% CI: 1.03–3.13) or not (pooled RR: 1.50; 95% CI: 1.06–2.11) (Table 3). This is most likely due to the different set of data included in the separate meta-analyses to avoid redundancy and to better target the type of exposure (accidental, occupational without accident and mixed). As an example, the study by Steenland et al. [23] was included in the meta-analysis ‘phenoxy contaminated (all)’ as it represents the largest study of workers exposed to PCDDS (NIOSH study). The studies by Ramlow et al. [20] and Bodner et al. [21] being, respectively, subset and update of this NIOSH study, were not included in the meta-analysis ‘phenoxy contaminated (all)’ but well in the meta-analysis ‘phenoxy contaminated without accident’. Details concerning studies overlaps were given in the material and methods section (stratifying studies in the meta-analyses) and studies included in the different meta-analyses were enumerated in Table 3.

Methodological aspects of the present meta-analysis

Independence of cohorts: Efforts have been made to avoid duplication of cohorts by abstracting results from the most recently updated publication. In addition, cohorts were classified by pesticide exposure and, for the phenoxy herbicides, by type of exposure (following an accident or not) to avoid redundancy. However, the possibility remains that we may have inadvertently included cohorts containing overlapping populations. As an example, we cannot exclude that workers employed in a single plant producing different types of pesticides, may have worked during several years in a section producing one pesticide and during another period in a section producing a pesticide of another type. As a consequence, these workers may appear in different cohorts included in the meta-analysis and thus the outcomes of such populations would be inappropriately over-represented in the results. There is, however, no mean to address this possibility on the basis of available data.

Exposure: Given the complexity of pesticide exposure, characterising and appropriately assigning exposure remains a challenge in epidemiology. Regarding exposure, the definitions of the cohorts (study populations) differ among studies. All studies included workers ever employed in a pesticide factory but with different exposure assessments. An assumption that underlies our analyses is that all individuals included in these cohorts were exposed to pesticides in their workplace. This relative paucity of information on workers exposure only allows to provide a RR of prostate cancer for individuals ever employed in a pesticide industry relative to those never employed in this industry. For causal inference, it would, however, be important to examine the risk according to a gradient of exposure. As reviewed by Alavanja et al. [84] exposure assessment methods have progressed from crude surrogates to improved techniques such as job exposure matrices. Among the studies included in the overall meta-analysis, exposure assessment varied considerably: (a) assessment by job title and/or history or work area [18, 28, 39, 56, 64]; (b) by job exposure matrices [23, 37, 79]; (c) by model based estimates of cumulative dose [36] or (d) by serum, blood, fat and/or urine measurements [16, 25–27, 35, 38]. The diversity of information on worker exposures is likely to have attenuated any association with prostate cancer. After stratification of the data by above-reported types of exposure assessment types [(a), (b), (c), (d); data not shown], increased pooled rate ratios were observed for each group [(a) pooled RR: 1.22, 95% CI: 0.86–1.72; (b) pooled RR: 1.19, 95% CI: 0.86–1.67; (c) SIR: 1.1, 95% CI: 0.3–2.8)] with the highest value observed for the group (d) based on biological samples measurements (pooled RR: 1.59; 95% CI: 1.05–2.41).

In addition, it has to be stressed that the final pesticide product is a combination of active ingredients together with a wide array of compounds of different toxicity. As a consequence, the health effect of a pesticide product may be a consequence of either the active substance, the other ingredients of the formulation or both [84]. Furthermore, workers are potentially exposed during the process of pesticide manufacture to a large array of reactive chemicals and intermediates involved in the chemical synthesis process, that may also contribute to the carcinogenic risk.

Confounding variables: Control of confounding remains a concern as it is for virtually all occupational cohort studies. Current knowledge of the aetiology of prostate cancer is limited and remains speculative. There are very few known risk factors (age, genetic predisposition and ethnic origin) and therefore very few confounders that could be controlled. Age is the most well documented risk factor for prostate cancer and most studies included in the meta-analyses controlled for age and calendar (time) period. Although familial history is difficult to imagine being correlated with exposure and therefore potentially confounding, Alavanja and collaborators [85] reported important family history–pesticide interactions in the Agricultural Health Study. For several pesticides, they showed a significantly increased risk of prostate cancer among study subjects with a family history of prostate cancer but not among those with no family history. Information on genetic predisposition was generally lacking in studies that were included in the meta-analysis. Wide variations in the incidence of prostate cancer between ethnic populations and countries have been reported with the lowest rates for Asians and the highest rates among African Americans in the US [86]. The relevance of this risk factor for European studies is probably limited and information on ethnic origin was available in too few studies to allow an aggregation of results adjusted for this factor.

Detection bias: It has been suggested that prostate screening programs (medical surveillance programs that included the determination of prostate specific antigen in serum) may have confounded the results of cancer incidence studies in some plants [37, 38]. Therefore, we rerun the overall meta-analysis excluding incidence studies (Table 4) [3639]. The meta-RR after pooling the 12 remaining mortality studies was 1.32 (95% CI: 1.03–1.69), reinforcing the observed association.

Publication bias: After applying conventional tests, the association observed in the present overall meta-analysis does not appear to have been significantly influenced by publication bias. The studies that were included in the present meta-analysis did not focus specifically on prostate cancer. Prostate cancer data were reported among other cancer types as part of the results of specific population surveillance. As a consequence, the criticism of the meta-analysis method concerning the limited use of negative findings (less likely to be reported in peer-reviewed sources) is less pertinent for the present meta-analysis. The failure to observe publication bias reflects the lack of evidence for a substantial deficit in small negative studies. However, it should be recognised that some data were omitted from the present analysis: published studies reporting no cases of prostate cancer [18, 24, 26 plant III] or giving too scarce information [45, 66] as well as studies ever conducted but not published for several reasons and reviewed by Greenberg et al. [87].

Causality of the association (s)

Several criteria proposed to assess whether an association may be considered as causal were reviewed, focusing on specific chemical classes when possible [8890].

(1) Strength of association. The meta-rate ratios remain lower than 2 for all conducted meta-analyses.

(2) Consistency between manufacturing workers studies. The great majority of risk estimators (16/20) from cohort studies reported an increased risk of prostate cancer in pesticide manufacturing whatever the pesticide class considered. Only two of them, belonging to the contaminated phenoxy herbicide group, presented a 95% CI that did not include 1 [21, 25]. A nested case–control study among manufacturing workers exposed to atrazine reported no evidence for an association between atrazine and prostate cancer [91].

(3) Consistency with other pesticide exposed groups. Consistency from meta-analysis to meta-analysis. The slightly increased overall risk of prostate cancer observed in the present meta-analysis is consistent with the increases observed in meta-analyses of end-product users—farmers or pesticide applicators [16] and strengthens the suggestion that exposure to pesticides may be a causal factor.

Consistency from study to study. Several studies dealing with exposures to specific pesticides or class of pesticides among groups others than manufacturing workers and reporting estimators of RR for prostate cancer are available in the literature. Cohort, PMR, case–control and ecological studies were conducted on groups (farmers/farm workers, pesticide applicators, outdoor workers, general population) exposed to organochlorines [85, 92101], chloroacetanilides [85, 102], triazines [85, 95, 98, 103], halogenated hydrocarbon nematocides [104] and chlorophenoxy herbicides [95, 105112]. Inconsistent results were observed among studies concerning a single pesticide or a single class of pesticides. Significant associations have been observed for organochlorines including DDT, aldrin, heptachlor, oxychlordane and dicofol [85, 99, 100] for the triazine simazine [98] and for phenoxy herbicides [107, 111, 112].

(4) Specificity. Prostate cancer is likely multifactorial in etiology and thus there is most probably not related to a single specific cause. Current knowledge of the etiology of prostate cancer is limited and remains speculative. The only risk factors that can be considered established are age, race/ethnicity and family history. Potential risk factors include diet, anthropometric factors, hormone profiles and concomitant medical conditions; others remain probably unknown resulting in a web of causation. Conversely, pesticide exposure is not uniquely associated with prostate cancer. The literature suggests an association between pesticides and other cancers including non-Hodgkin’s lymphoma, leukaemia, multiple myeloma, soft tissue sarcoma, pancreatic cancer, lung cancer, ovarian cancer [84], although studies have been inconsistent. However, it has to be stressed that the specificity criterion has proved invalid in a number of instances (e.g. smoking and lung cancer) and the lack of specificity should not be taken as evidence against causality.

(5) Dose-response and/or temporal relationship. In most studies, the available data are inadequate to determine whether dose-response and/or temporal trends exist due to limited number of observed prostate cancer cases, lack of environmental and/or personal monitoring data, uncertainty concerning start and duration of possible exposure, ... Only a few studies provide some indication of a possible quantitative relationship or lack thereof but it remains insufficient to drawn clear conclusions. Possible relationships with long potential induction time and long duration of employment have been reported for triazines [38]. No associations with dose and duration of potential exposure and/or years of latency have been observed in some studies dealing with phenoxy herbicides [36, 79]. Finally, results that seem paradoxical have also been reported for phenoxy herbicides [64] and chloroacetanilides [37]: analyses by grade (background, low, high exposure) and duration of potential exposure and allowing for a high latency period did not suggest any association but SMR values for prostate cancer increased with grade of potential exposure [64]. A slight but non-significant increase of prostate cancer incidence has been reported among workers with high alachlor exposure but there were no discernible relation between cancer incidence and years of exposure or time since first exposure [37].

(6) Biological plausibility. We could not locate an experimental study involving a pesticide considered in the present study that reported on tumours of the prostate. Only limited considerations with regard to consistency with existing knowledge can be presented because of the lack of fundamental understanding of the basic biology of human prostate cancer. Although the mechanism is unclear, it is known that hormones (both androgens and estrogens) likely play a significant role in the etiology or promotion of prostate cancer. As a result, it is plausible that chemicals able to modulate steroid sex hormones as agonists, antagonists or as a mixed agonist-antagonist may contribute to the development of prostate cancer through hormone-mediated effects. Several pesticides may interfere with sexual hormones through direct action on receptors but also through indirect non-receptorial mechanisms (e.g. organochlorines: for review, see [99, 113]; triazines: for review, see [114123]; for 2,4-D: [124126]. For other pesticides, it is reported that mutagenicity and alterations of other cellular processes including protein metabolism and cell proliferation may contribute to their clastogenic and tumorigenic activity (chloroacetanilides: [127]; halogenated hydrocarbon nematocide—DBCP: [128132]. In addition, as DBCP is also known to damage human male reproductive organs, it is plausible that it could contribute to the development of prostate cancer (for review, see [56]). The scientific evidence in humans and animals relevant to cancer risk of chlorophenoxy herbicides has been reviewed by several authors (e.g., [133136]). IARC evaluation of the carcinogenicity of chlorophenoxy herbicides concluded that there is inadequate evidence for carcinogenicity in animals and limited evidence that occupational exposures to chlorophenoxy herbicides are carcinogenic to humans (group 2B: possibly carcinogenic to humans) [133]. No scientific consensus about the potential human carcinogenicity of contaminants (dioxin-like compounds) has still been reached [137141].

Taken together, the analysis of the above criteria does not provide sufficient support for an unequivocal causal association between pesticide exposure and prostate cancer. The consistency between manufacturing workers studies and with other pesticide exposed groups strengthens the evidence that pesticide exposure may be a causal factor for prostate cancer. However, the studies reviewed contain insufficient information on exposure in order to distinguish the possible influence of pesticides from other factors and to demonstrate clear dose–response and/or temporal relationship. The lack of a clear dose-response and/or temporal relationship is most probably the reflection of the scarcity of adequate data in studies conducted to date rather than the non-existence of such a relationship. Although some existing scientific knowledge fits well with a possible association between pesticide exposure and prostate cancer, the scarcity of biological explanation of mechanisms inducing prostate cancer do not allow to reasonably draw conclusions concerning biological plausibility.

No specific chemical class of pesticide can be highlighted that would be responsible for a causal association between prostate cancer and working in a pesticide manufacture but the strongest evidence presented is for those workers exposed to phenoxy herbicides probably with a significant contribution of the contaminants (PCDDs/PCDFs).

Conclusion

The present evaluation indicates that none of the individual studies reviewed allows drawing definitive conclusions on the risk of prostate cancer in pesticide manufacturing workers. The overall meta-analysis provides quantitative evidence of a moderately increased risk (meta-rate ratio of 1.28, 95% CI: 1.05–1.58), in agreement with prior reviews focussing on other occupational groups exposed to pesticides (farmers, pesticide applicators). The homogeneity observed between the individual rate ratios tends to increase the consistency of the association. These results reinforce the evidence for a possible relationship between pesticide exposure and prostate cancer. 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. Meta-analyses conducted for specific chemical classes of pesticides did not allow to highlight a class that would be responsible for a causal association between prostate cancer and working in a pesticide manufacture but emphasized a more marked increased risk for the manufacture of phenoxy herbicides possibly related to PCDD/PCDF contaminants.

Acknowledgement

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

Copyright information

© Springer 2006