Triazine herbicides and ovarian epithelial neoplasms

1989;15:47-53. The possible role of a class of herbicides, triazines, in ovarian carcinogenesis has been evaluated in a population-based case-referent study. The trade names reported by the studysubjects, and thetypeof cultivation in which theyworked, were used for theassessment of exposure. Women previous ly exposed to triazines showed a significant relative risk of 2.7 for ovarian neoplasms. Although none of the doses could be quantified for the study subjects, two risk trends in favor of the plausibility of the association were found: the first by duration and the second by probability of exposure. The population representativity of the study and the comparability of information between the cases and referents sug gest the lack of any major bias in the results. Triazine-related risk remained consistent when the analysis was restricted to farmers and when the exposure to other herbicides and to other types of cultivation were considered. Unexposed farmers had the same risk as unexposed nonfarmers.

Triazines are a class of chemicals widely used as herbicides for the control of germinating weeds in corn (zea ma ys), and for the con trol of grass under young for est trees and vines (I).
A possible role in ovarian car cinogenesis has been suggested for triazines in a hospital-based cas e-referent study car ried out by us in 1984 (2). Although the study was inadequate to assess th e risk fo r a speci fic herbicide , expos ure to triazines was o ften rep o rted by the intervi ewed subj ects. Tr iazin es were also the most used herbicides in the are a under study. In 1970 th e Nat ional Institute of Stat istic s (3) reported th at 51 417 kg of atrazine and 5 652 kg of simazine were sold in Alessan dria Province, and it is kno wn that these figures represent onl y a portion of all sales.
To evaluate the specific role of triazines in ovarian ca rcinogenesis with a different case series at the population level, a new case-re ferent study ha s been co ndu cted in Alessandria (Piedmont, Ital y) Pro vince. Th is investigation was planned to ha ve an 80 % po wer of detecting a threefold risk with a 5 % significance level (4) on the basis of the risk ratios and the prevalence of exposu re found in our pr eviou s wo rk .

Study base
In the Alessandria Province, tri azines are mainl y used as herbicides in corn cult ivation. The study was th erefor e confined to the provincial area where corn is grow n , ie, to 143 neighboring municipalities o f the 190 municipalities fo rming the Ale ssandria P ro vince. The study ba se, ie, the " population experience captured" by the study (5), was defined as all women who were at risk of ovarian cancer (ie, had one or both ovaries), were in the age range of 20-69 years, and were residents in the municipalities under stu d y in th e period I July 1980 t o 30 June 1985. This period followed th e incidence peri od of the pr eviou s study (2).
Women aged 70 years and old er were excl uded becau se of the difficulties in tracing and interviewing elderly peo ple and because the histol ogical verificatio n of ovarian tumors is frequently lacking among elde rly peo ple (6). Subject s under 20 years of age were also excluded since ova rian ep ith elial neoplasm s ra rely occur before that age (6) .

Cases and ref erents
A ll wom en histolo gicall y confirme d to have primar y mal ignant epithelia l tumor s o f th e ov ary [as de fined by categor y I of the World Health Or gan izat ion (common "epithelial" tumors, ie, serous, mu cinous , endom etr ioid , clear cell, Brenner tum ors, mixed epitheli al and uncl assified epit helial) (7)] newly diagnosed du ring the study period (1 Jul y 1980-30 June 1985) were eligi ble as cases. The cases were retrospect ively identified at th e end o f the study period from th e Ale ssan-dria Hospital Cancer Register, which contains reports from eight hospitals in the Province, and from the archives of 10 other hospitals to which residents in the study area, according to an ad hoc survey (8), were likely to be referred for ovarian cancer.
After the case search was completed, two referents with the same age (plus or minus five years) were randomly selected per case from the electoral rolls of the municipalities of the study area. Those referents who revealed in the interview that they had had a bilateral oophorectomy prior to the study period were excluded from the analysis. Table 1 shows the distribution of the cases and referents according to their inclusion in the study. Of the 69 eligible cases, 42 were alive and interviewed personally, and 27 were dead at the time of the interview. For 23 of the decedents a close relative could be traced and was interviewed. The mean interval between the diagnosis and the interview was 30 months.
Of the 150referents selected from the electoral rolls, 13 could'not be interviewed, and 11 were excluded because they had undergone a bilateral oophorectomy prior to the study period. The analysis was eventually carried out with data on 65 cases and 126 referents.
Four nonepithelial cases and two cases that were eventually recognized as nonresidents of the study area were excluded, but their referents were nonetheless retained in the referent group. This occurrence accounts for the discrepancy between the numbers of eligible cases (N = 69) and selected referents (N = 150).

Data collection
A questionnaire was designed to obtain information on reproductive factors, including those reported in the literature (9, 10, 11) as determinants of ovarian cancer, and to establish occupational exposure to tri- azines. There were also questions about the subject's residence and gynecological and other relevant diseases.
A lifelong occupational history of the subject was also requested, and, if any involvement with agriculture was reported, the period and type of cultivation were asked. For each type of cultivation present in the study area, a specific form with a detailed description of tasks and, when applicable, use of herbicides was used.
A list of herbicide trade names was prepared to help subject recall, but it was presented only when a subject was unable to remember the substances she used. Information obtained with the aid of this list was recorded separately.
A final question was asked regarding other nonprofessional instances of exposure to herbicides, ie, any use for domestic purposes, eg, courtyard or garden.
All interviews were held at the subjects' homes. For the deceased cases, the relatives were interviewed. All interviews were conducted by the same person, who was unaware of the disease status prior to, and sometimes even after, the completion of the interview. (In Italy cancer patients are frequently told that they have a benign disease.)

Exposure
Triazine herbicides are stable chemicals (12), and the mechanisms of human absorption are unknown. It seemed thus reasonable to consider both the preparation and distribution of such herbicides, usually occurring once or twice a year, and the work in the fields where they were used as instances of exposure.
The categories (i) definitely exposed, (ii) possibly exposed, and (iii) unexposed were adopted for exposure to triazines. Subjects who were involved in the preparation or use of triazine herbicides or who worked in corn cultivation with reported use of herbicides were considered definitely exposed. According to local agricultural experts, triazines are in fact used in all herbicide-treated corn cultivation. The possibly exposed subjects were those who acknowledged personal exposure to herbicides, or work in some job possibly involving herbicide exposure, but for whom the substances could not be more preciselyidentified and those who denied personal usage of herbicides but worked in corn cultivation after 1964, when the use of triazine herbicides on corn crops became extensive in the Alessandria Province, as reported by the National Institute of Statistics (3). The unexposed subjects were those for whom professional or domestic exposure to herbicides could be reasonably ruled out.
Exposure to triazines was established blindly by two of us independently after all the interviews were completed. Only exposure prior to the study period was considered. The length of exposure was defined as the duration, in years, of the occupations involving exposure. Any subject judged as definitely and possibly exposed in different periods was classified as definitely exposed. In such cases the length of exposure was computed for the defi nitely exposed period s only. A summary o f herb icide expos ure of the definitely exposed subjec ts is given in the appendix.

A nalysis
Unconditional logistic analysis yielded estim ates (13) of the risk rat io (R R), and the associ ated 90 % co nfidence inte rval (90 070 C I) was obtained (4). All var iables bu t age, consi dered as suc h, were included in the logistic regression as categorical. T he categories used are tho se reported in th e ta bles. For triazine-relat ed risks, non exposur e was always used as the reference category. We tested the trend of the RR estimates among the categor ies of exposure by considering the sign ificance of the regression coefficient (4).

Reproductive facto rs
Th e RR estimates for the reproduct ive fact ors are reported in ta ble 2. Parity appeared as a protective factor , an d the magnitude of such protection depe nde d on the number of live bir ths, th e RR estimate being 0.4 (90 070 CI 0.2-0.8) fo r the women who had had one to three childre n and 0.3 (90 % CI 0. 1-1 . 1) for th ose who ha d ha d fo ur or more children . T he onetailed P-valu e for trend was 0.0 1 when the preceding cat egor ies were used.
The users, versus nonusers, of or al con traceptives had an RR estimate o f 0.6, (90 % C I 0.2-1.5) when age an d pa rity were adj usted for.
Women with menarche prior to the age of 13 years had a risk o f 1.1 (90 % C I 0.6-2.0). For those who reported difficult ies in conceiv ing, the risk ra tio was 1.7 (90 % C I 0.7-3.9). Those who miscarried ha d a risk rat io of 1.4 (90 0J0 CI 0.7-2.8), and for those who had abortions the risk ratio was 0.8 (90 % CI 0. 1-5.5).
The significant risk for single wom en disappeared a fter adjustment for age, number of live births, and usage of oral contraceptives.
Th e mean age at menopause, among the subjects who were already in menopause before the stu dy period, was 47.6 years for the cases and 48.3 years for the referents.
Age at first birth did not seem to play any particular role. Am ong parou s women, the adj usted risk fo r those who had their first child at ages of 25 years or more was 1.0 (90 % CI 0.9-1.1).

Triazine-related risks
Tab le 3 shows the ovarian cancer risks for the adopted categories of exposure, adjusted for age, number of live births, and use of oral cont raceptives. A significant RR estimate of 2.7(90 % CI 1.0-6.9) was fou nd for th e subjects definitely exposed to tr iazines. Those po ssibly exposed had a lower risk (RR 1.8, 90 % CI 0.9-3 .5). Among both the definitely exposed and possibly exposed subjec ts, the risks were related to the du-  a Subjects in the " exposed" ca tegory; as a whole 65 cases and 126 referents were included in the analysis. b 90 % confidence intervals in parentheses.
C Al l models include age, parity, and ora l contraceptives. d Among parous women , ie, 43 cases and 104 referen ts .
4 Table 3. Distribution of the cases and referents by triazine exposure and the related relative risk ratios for ovarian cancer.

Risk ratto>
Cases' Referents' Crude < 10 years of exposure Adjusted "" 10 years < 10 years and of exposure "" 10 years of exposure combined All subjects Definitely exposed Possibly exposed Unexposed" Agricultural workers Definitely exposed  ration of the exposure. The logistic estimate of the regression coefficient for the definitely exposed was 1.7 (P = 0.05); for the possibly exposed it was 1.4 (P=0.08). The risk for the possibly exposed group, being intermediate between the definitely exposed and unexposed groups, is likely to reflect the prevalence of the exposure in the possibly exposed category. In fact, the possibly exposed category was formed from an unknown proportion of triazine-exposed subjects for whom no precise estimate can be made with the use of subject recall.
To evaluate the latency between the exposure and the diagnosis of the disease, we first considered the occupational histories of the definitely exposed subjects. For almost all the subjects with 10 years or more of exposure, triazines began to be used in the early 1960s and were still in use in the study period. This study, therefore, does not allow an evaluation of the latency dependence of the observed association.

Discussion
The study has been designed with the aim of minimizing selection bias by including all incident cases and a representative sample of the study base as referents, controlling any confounding by known risk factors, and avoiding information bias by prior specification for the criteria of gathering and interpreting the information. The extent to which these objectives have been attained are examined in the following discussion.

Population representativity
We used incidence data from the nearest cancer registry (Lombardy Cancer Registry) (6) to estimate the expected number of cases in the area during the study period, age and age-specific ovarian cancer mortality in the Alessandria province being adjusted for. Seventy-three cases (69 epithelial and 4 nonepithelial) were collected in the study, as against an estimate of 78.19 histologically verified cases. Thus it would seem that only a minority of the cases, if any, might have been lost.
Samples taken from electoral rolls are representative of the source population. Our sampling was, however, carried out at the end of the study period, and interim deaths and migrations from the area were not represented in the file. The expected number of deaths and migrations in the referent group, computed from data of the National Institute of Health, were 5.7 and 0.7, respectively. Thus, live subjects sampled from the electoral rolls gave a 96 % representativity of the study base.

Reproductive factors
The estimates for reproductive factors were close to those reported in the literature. This agreement gives indirect evidence in favor of a lack of major biases in the study, although the study size and the prevalence of certain factors do not permit reliable estimates for some determinants.
The age of a woman at the time of her first birth has been proposed as a major etiologic factor for epithelial cancer of the ovary (14). However, in this study, as in other studies (15,16), no such association was found.

Data collection
The data were examined for any difference in the accuracy of the interviews of the cases and referents which might indicate the existence of an information bias. The mean number of recorded periods in the professional history was 2.46 (SD 1.24) for the cases and 2.51 (SD 1.24) for the referents. The mean duration of the interview was 25 min for both the cases and the referents . For 15 cases and 29 referents another person participated to help subject recall. For two cases and two referents the prepared list of substances was used to help them assess their triazine exposure.
In two instances there was disagreement between the two evaluators in attributing exposure. Both were due to an error in applying the formal exposure criteria, and the correction was straightforward.
Thirty-five percent of the cases were dead at the time of the data collection. Since information obtained from relatives may be less complete and precise, the triazinerelated risk was evaluated separately for the living and dead cases. The crude RR estimate for the deceased subjects was 1.9 (90 % CI 0.5-7.6) for the definitely exposed and 2.0 (90 % CI 0. 8-4.8) for the possibly exposed. The corresponding estimates for the living cases were 2.4 (90 % CI 0.9-6.8) and 1.4 (90 % CI 0.5-4.1), respectively . Thus the risk showed lower values than the overall estimate for the definitely exposed decedents and higher values for the possibly exposed decedents. A possible interpretation could be that a slightly higher proportion of definitely exposed subjects was classified as possibly exposed when relatives were interviewed.

Specificity of association and confounding
To evaluate whether the observed risk for triazine exposure reflected a nonspecific agricultural risk only, we restricted the analysis to those who reported having worked in agriculture, henceforth called " farmers." As shown in table 3, the estimates were slighty higher, and the trends were retained in relation to the duration of exposure among the definitely exposed group, and to the proportion of exposed .
The risk for the unexposed farmers was about the same as that for the unexposed non farmers (RR 0.9, 90 OJo CI 0.5-1 .9) when adjusted for age, number of live births, and use of oral contraceptives.
The criteria used to establish exposure to triazines were based both on the reported herbicide trade names and on work in corn cultivation. This practice is probably acceptable because, in our study, all trade names reported as herbicides used in corn cultivation were in fact triazines.
A number of farmers used herbicides for crops other than.corn, and especially for wheat. Table 4 gives the details of exposure to other herbicides by categories of exposure to triazines. As farmers who use triazines tend also to use other herbicides, exposure to different kinds of herbicides tends to be associated . The data, however, do not suggest any association between exposure to other herbicides and ovarian cancer.
Cultivation of other crops might be an indicator of farmers' exposure to chemicals not considered in the study. Adjustment for other types of cultivation in the area , ie, wheat, vegetables, forage, and vines, produced only minor changes in the risk estimate.
Information on many reproductive factors was collected in this study. Taking into account age, number of live births, and use of oral contraceptives slightly increased the triazine-related risk estimates. When all the reproductive factors listed in table 2 were included in the model, there was virtually no change in the estimates. Confounding by reproductive factors thus does not seem a possible explanation for the estimated risks .
When establishing the study base, as a precaution, we also excluded an area with asbestos pollution, since  thi s substa nce has been reported as a risk factor fo r ovarian cancer (17).

Concluding remarks
The main result of thi s study was that women with previou s exposure to triazine herbicid es showed a twoto threefold risk of epith elial ovarian cancer as compared to unexposed women. This findin g is consistent with the results o f o ur previou s study (2) and alternative explanatio ns seem unlik ely.
Although no do ses could be quantified amo ng th e study subjects, two risk trends in favor of the plau sibility o f the associa tio n are noti ceabl e in the dat a , ie, one for the duration of expos ure and the other for the probability of expo sur e.
Limited experimental evidence of th e carcinogenic effect o f tri azines on animals (commercial products, given subcutaneously or intraperitoneally to Swiss albino fema le mice at 2 ppm) has been reported (18), with a suggestive specific action on mesodermal-derived tissues. As ovarian surface epithelial neoplasms are, in fact , tumors of th e ovari an mesoth elium (19), we might speculate th at this neoplasm is a model for th e carcinogenic action of tri azines on the mesod erm .
Th e study result s, and possibly the presented arguments, are also supported by the risk of non-Hodgkin's lymphomas associated with triazine exposure, as was recently found by Ho ar et al (20) in an epidemiologic stu dy on fa rmers in the state of Kan sas in th e Unit ed States.