Agricultural activities and risk of central nervous system tumors among French farm managers: Results from the TRACTOR project

Abstract The etiology of central nervous system (CNS) tumors is complex and involves many suspected risk factors. Scientific evidence remains insufficient, in particular in the agricultural field. The goal of our study was to investigate associations between agricultural activities and CNS tumors in the entire French farm manager workforce using data from the TRACTOR project. The TRACTOR project hold a large administrative health database covering the entire French agricultural workforce, over the period 2002‐2016, on the whole French metropolitan territory. Associations were estimated for 26 activities and CNS tumors using Cox proportional hazards model, with time to first CNS tumor insurance declaration as the underlying timescale, adjusting for sex, age and geographical area. There were 1017 cases among 1 036 069 farm managers, including 317 meningiomas and 479 gliomas. Associations varied with tumor types, sex and types of crop and animal farming. Analyses showed several increased risks of CNS tumors, in particular for animal farming. The main increases in risk were observed for meningioma in mixed dairy and cow farming (hazard ratio [HR] = 1.75, 95% confidence interval [CI]: 1.09‐2.81) and glioma in pig farming (HR = 2.28, 95% CI: 1.37‐3.80). Our study brings new insights on the association of a wide range of agricultural activities and CNS tumor and subtype‐specific risks in farm managers. Although these findings need to be corroborated in further studies and should be interpreted cautiously, they could have implications for enhancing CNS tumor surveillance in agriculture.

cancer etiology, however, remains uncertain. Here, among farm workers in France, 26 agricultural activities were investigated for potential associations with risk of CNS tumors, including type-specific CNS cancers. CNS cancer risk varied in association with sex and types of farming.
In particular, farming of specific animals increased risk of several CNS cancers, including meningioma and glioma. While further research is needed, the findings may have implications for CNS tumor surveillance in agriculture.

| INTRODUCTION
The incidence of central nervous system (CNS) cancers has increased worldwide from 17% in the last two decades. 1 The etiology of the CNS tumors remains mostly unknown and few risk factors have been identified. A synthesis of 40 years of epidemiologic studies of farming and brain cancer concluded that farming (as a large entity), livestock farming and "documented exposure to pesticides" were associated with increased risks of CNS tumors. 2 However, results about the involvement of pesticides in the occurrence of CNS tumors are still scarce, inconsistent and insufficient. [3][4][5] There are limited cohort studies investigated work-related cancer in agriculture. 6 In France, the AGRICAN cohort includes about 180 000 people living in 11 French metropolitan counties beneficiating from cancer registries. [7][8][9] AGRICAN includes only 7% of all active French agricultural workers covered by the National Health Insurance Fund for Agricultural Workers and Farmers (MSA) data (retired persons excluded). Other French studies are also limited both geographically and in scope, and pertained only to a small proportion of the agricultural workforce. [10][11][12] It is therefore paramount to consolidate existing and recent evidences by studying the entire French agricultural workforce on the whole metropolitan territory.
Data on the entire French agricultural workforce are available to the TRACTOR project. 13 The goal of our study was to investigate the associations between agricultural activities and CNS tumors in the entire French farm manager workforce over the period 2002-2016 on the whole French metropolitan territory.

| Population
We selected in our analysis all farm managers, including farm or company managers, owners and self-employed persons, from 2002 to 2016 within the TRACTOR project. The study population have been described previously. 13 Briefly, yearly routinely collected insurance data on contributor' demographic characteristics and health are available for the TRACTOR project. Demographic characteristics (eg, occupation, age, sex, farm surface) are collected by MSA from forms that are filled by farm managers during their yearly insurance affiliation.

| CNS cancer identification and statistical analysis
CNS tumor cases were identified using ICD-10 codes (10th revision of the International Statistical Classification of Diseases and Related Health Problems). Information on CNS tumor cases came from administrative insurance health data (MSA), where each disease is coded by MSA insurance physicians with a 3-digit long ICD-10 code based on patient medical reports. 13 Table 1 presents all ICD-10 codes and grouping of ICD-10 codes considered in this work.
To assess CNS tumor risk related to agricultural activities, hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional hazards model, with time to first CNS tumor insurance declaration as the underlying timescale. The reference group included farm managers who did not carry out the activity of interest.
For instance, for pig farmers, the reference group included every farm managers that did not farm pigs between 2002 and 2016 and could therefore include individuals who may be exposed to pesticides or other risk factors. CNS tumors risks (overall and by types) were estimated according to each of the 26 activities when the number of exposed cases exceeded or equaled 3. Only the main agricultural activity in terms of effective working time was known. 13 All analyses were adjusted for age (<40, 40-49, 50-59, ≥60) and sex. We also conducted analyses stratified by sex to identify potential gender specific CNS tumor risks that may come from differences in occupational exposure and tasks between women and men. Several potential confounders (covariates) were considered ( Table 2). The selection of covariates for the Cox proportional hazards model was based on the variance inflation factor (VIF). 14 Collinear covariates, with a VIF > 2.5, were not included in the models. All analyses were also adjusted for the 13 metropolitan French administrative geographical regions where the farm is located to account for a potential confounding effect related to possible unmeasured and unequally distributed geographically risk factors. Some administrative geographic regions could be correlated with agricultural activities, which may mask associations with exposures. To address this matter, we applied a restrictive variable selection based on the VIF (≤2.5). Therefore, only administrative geographic regions poorly or not collinear with agricultural activities have been included in each model. Hence, depending on the model considered, different administrative geographic regions could be taken into account and, for some models, it could sometimes happened that no administrative geographic regions were considered if they were all found to be collinear (VIF > 2.5) with the activity of interest. No methods to handle missing data was needed because data originated from compulsory agricultural insurance fund, which was complete for all variables of interest available to the TRACTOR project. All statistical analyses were performed using R software 4.1.2 (R Core Team, Vienna, Austria) for Windows 10©.

| Population characteristics
Baseline characteristics of the study population are presented in Table 3. Around one third of all farm managers were crop farmers (29.5%), while 15.3% and 11.4% performed dairy farming and viticulture activities, respectively ( Figure S1). and a higher number of comorbidities (20% vs 15%).
Over 58% of the CNS cases were malignant neoplasms (Table 4).
Most CNS tumors were gliomas (47%). Gliomas affected more men (57%) than women (32%). A total of 32 (3.1%) individuals (13 men and 19 women) were declared with several types of ICD-10 codes for CNS cancers. The percentages of CNS tumors varied depending on the agricultural practice/activity and sex ( Figure S2).
a Hazard ratios were estimated by Cox models with time to first CNS tumor insurance declaration as the underlying timescale, when the number of exposed cases was sufficient (m The percentages in brackets refer to the ratio of exposed cases in the study population and the total number of cases in the overall population. c The percentages in brackets refer to the ratio of exposed cases in the study population and the total number of cases in the study population. were, respectively, associated with increased risks and negative trends of benign meningiomas in women. A detailed comparison between results from the French cohort AGRICAN and our study is presented in Figure 1. There were some differences between AGRICAN and TRACTOR.      Figure 1). The only difference was observed for pig farming for which we found an increased risk, in particular in men, while AGRI-CAN found a positive trend.

| Glioma
Eight studies have found an increased risk of glioma in overall agriculture. 2  Regarding crop farming, we found an increased risk while AGRI-CAN found a positive trend, possibly due to a larger number of exposed cases in TRACTOR (116 vs 79). Regarding animal farming, we found an increased risk in pig farming, in particular in men, while AGRICAN found no trend. We also observed differences of risk between women and men, in particular for dairy farming and for poultry and rabbit farming, suggesting potential gender specific tasks/exposures.

| Meningioma
One study reported a decreased risk of meningioma for open field farming in France (OR = 3.58 [1.20-10.7]). 10 Regarding crop farming, AGRICAN found a risk similar to ours (Figure 1). We found a positive trend for viticulture while AGRICAN reported no trend. Regarding animal farming, we found similar results than AGRICAN for most animal farming, with the exception of mixed dairy and cow farming for which we found an increased risk contrary to AGRICAN. This risk was higher for men than women, suggesting potential gender specific tasks/exposures. Regarding pig farming, AGRICAN reported an increased risk, which we did not find for overall meningioma, but that we observed for uncertain meningioma.

| Risk factors
The etiology of CNS tumors is complex and involves many risk factors that could act differently according to subtype and that could play a role in the positive and negative associations that we found. Ionizing Farmers are exposed to several physical, biological and chemical agents that can be potentially harmful. A recent review synthesizing 40 years of epidemiologic studies, including 20 cohorts, supports an increased risk of CNS cancer from farming related to potential pesticide exposure. 2 Several studies found, for pesticide users and different pesticide classes, an increased risk of CNS tumors, 8,9,22 gliomas 9,23 or meningioma. 8,9,24 By contrast, one study found a decreased risk of gliomas for phenoxys exposure in the United States 25 and a French case-control study reported a decreased risk of gliomas and meningiomas for indirect pesticide exposures. 10 We found no studies reporting either a decreased or an increased risk of CNS tumors other than gliomas or meningiomas related to pesticide exposures.
Biological risk factors are also of upmost interest. The use of pharmaceuticals in veterinary medicine, and in particular progestogens, could be a hypothesis to consider as they produce effects similar to those of the natural female sex hormone progesterone in the body and are, sometimes, associated with brain cancers, in particular meningiomas. 26 In animal farming, progestogens are used to facilitate induction of normal estrous cycle activity in animals, in particular in swine/ pig and horse breeding. 27,28 In addition, the role of infectious agents (eg, mycoplasma, viruses and bacteria) in the development of cancers, in particular for CNS tumors, have been considered recently. 29,30 For instance, some neurotropic viruses could lower 31 or promote 32 the risk of CNS tumors. While many farmers are exposed to vector-borne diseases transmitted by animals or insects (eg, mosquitoes or ticks), the role of infectious agents in the occurrence of neoplasms remains controversial.
Although the etiology is unclear, there is suggestive evidence that parental occupational exposures could increase the risk of childhood brain tumors. [33][34][35][36][37] Several studies reported positive associations between maternal prenatal occupational exposure to farm animals (pigs, horses, and poultry). By contrast, a pooled birth cohorts prospectively evaluating exposure to pesticides, animals, and organic dust in relation to childhood CNS tumor risk found no increased risks of CNS tumors related to paternal exposures to pesticides and animals using pooled data of 329 658 participants from birth cohorts in five countries (Australia, Denmark, Israel, Norway, and the United Kingdom). 38

| Strengths and limitations of this work
The most important strength of our study is the large number of exposed cases and completeness of available data. Because of the exhaustiveness of the population studied (entire French farm manager workforce) and because the reference group included only farm managers who did not carry out the activity of interest, the healthy worker effect remained limited. Compared to most studies, our study was restricted to farm managers. Farm managers and employees were not included in the same analysis due to different coding systems and data structure. 13  with registry data requires proper authorizations from the independent administrative authority protecting privacy and personal data (CNIL) and cancer registries, which we do not have.
Another limitation pertained to the ascertainment of occupation and exposure. Only an indirect exposure estimation was possible using activities from administrative databases. An interesting future step will be to study the active ingredient utilized in agricultural activities using a crop-exposure matrix such as Pestimat in order to ascertain more accurately the use of phytosanitary products. 39 The downside of this approach is that information from crop-exposure matrices are not available for each individual, but only at a large collective scale. Therefore, only a probability of pesticide use can be attributed to each farm manager based on available information (activity and location). In addition, the probability of pesticide used would be a rough estimation as the activities and locations available to TRACTOR are not descriptive enough to exploit the full potential of crop-exposure matrices.
Although information on chemical, biological or physical agents encountered/used by farm managers and several potential confounders (eg, smoking and alcohol habits) were not available due to the inherent nature of available data (health insurance), risks were adjusted on important confounders (sex, age, geographical area) and on several covariates after a conservative selection based on the VIF (VIF ≤ 2.5). Confounding factors not available to the administrative health databases from TRACTOR and therefore not considered in this work could represent a bias. The potential impact of this bias on the results is hard to evaluate as these variables were not available. It is possible that their absence could bias the estimated effects and confounds/masks the genuine relationship between agricultural activities and CNS tumors. Findings should therefore be considered carefully.
To refine analysis and address the aforementioned issue, external sources (eg, cohort studies and exposure matrices) could be linked to the TRACTOR project. 13 In our study, age was considered in the models as a category rather than as a continuous variable. This choice was based on statistical consideration. Indeed, age did not follow a normal distribution and was moderately skewed to the right (data not shown). Categorizing continuous variables is a common practice in epidemiology. 40 However, this practice has shortcomings such as loss of information, statistical power and increased probability of false negative findings (Type II error). 41,42 To reduce the loss of information and minimize the amount of residual confounding, we used four age categories. 43 An alternative solution to the categorization of age could be to consider age as continuous in the models by using regression splines, smoothing splines or relax linearity with polynomial effects. 40  There were a few activities that were found with decreased risks of CNS tumors compared to activities that were found with increased risks. This may be explained by the fact that "potential confounders" differed from a model to another due to the variable selection process (based on the VIF) and because the reference group differed from an activity to another.
To lessen the possibility of chance findings, we conducted an analysis only when the number of exposed cases was ≥3. False associations resulting from multiple comparisons might be an issue in our analysis, but approaches used to limit false positive findings (Type I errors) (eg, Benjamini-Hochberg procedure) are too conservative, increase the risk of false negative findings (Type II errors) and are not relevant in the framework of large cohort study with data on multiple illnesses. 46 In our study, we chose the time to first CNS tumor insurance declaration as the underlying timescale. The choice of the time scale is highly discussed in the literature but, to the best of our knowledge, there is no general consensus on which time scale is the most appropriate for a given question or study. According to several studies, using time-on-study models may be preferable since these models perform at least as well as the left truncated age scale model, and also because they are more robust to misspecification of the underlying time scale and have better predictive ability in general. [47][48][49] There has been some differences in the associations found in this work and the ones from literature. Some differences may be explained by the difference in the study design, health data origin and by different temporal and geographical scales. However, cohort studies adjust on more potential confounders and rely usually on more accurate/descriptive exposure ascertainment. Despite these differences, many findings were consistent with existing literature, but with more exposed cases, narrower 95% CIs and information on both sexes and several CNS tumor types that have been rarely studied before. Nevertheless, findings should be considered carefully by taking into account, the number, the direction and the magnitude of all examined risk associations.
In conclusion, the TRACTOR project brings new insights and a wealth of information on the association of a wide range of agricultural activities and CNS tumor and type-specific risks in farm managers, overall and for both sexes. The completeness of data and the large number of exposed cases offered a unique opportunity to study a rare disease such as CNS tumor. Results from our study are comple-

CONFLICT OF INTEREST
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

DATA AVAILABILITY STATEMENT
The data that support the findings of our study are available upon reasonable request to the Mutualité Sociale Agricole (MSA) but restrictions apply to the availability of these data, which were used under MSA approval and were approved by the French independent administrative authority protecting privacy and personal data (CNIL) for the current study, and so are not publicly available. Further information is available from the corresponding author upon request.

ETHICS STATEMENT
The use of MSA data for the TRACTOR project was approved by the