Exposure to pesticides and risk of Hodgkin lymphoma in an international consortium of agricultural cohorts (AGRICOH)

Hodgkin lymphoma (HL) is a rare cancer diagnosed in an estimated 83 000 individuals each year worldwide [1‐3]. In most age groups, HL is slightly more common among males, with overall global age-standardized rates of 1.2 and 0.8 per 100,000 among males and females, respectively [2]. Unlike other lymphomas, its incidence has a bimodal age distribution, with one peak among adolescents and young adults (15–35 years) and another at older ages (50 +) [1]. The tumours are characterized by the presence of malignant Reed-Sternberg cells (1%) of B-cell origin but are otherwise dominated by non-malignant inflammatory and accessory cells [1]. The histological subtypes of HL fall into two main groups, classical and nodular lymphocyte predominant; the vast majority are classical (~ 95%) and therefore most epidemiological data reflects this dominant type [1]. Histologic subtype and Epstein-Barr virus infection (present in 40% of tumours) define epidemiologically and etiologically distinct forms of HL [1, 4]. Known risk factors of EBV-positive classical HL include: family history, genetic polymorphisms in human leukocyte antigen complexes, immune deficiency and smoking, but the causes of EBV-negative classical HL and the much rarer nodular lymphocyte predominant HL are largely unknown [1].

Occupational exposure to pesticides has been suggested as a possible risk factor for HL, with oxidative stress and immunotoxicity suggested as potential mechanisms [5]. However, few studies have evaluated the risk of HL associated with exposure to specific pesticide active ingredients, in part due to the challenges of having sufficient statistical power to study this rare outcome, as well as a lack of data on exposure to specific active ingredients. In a pooled analysis of case–control studies from the USA and Canada, ever use of the organophosphate insecticide terbufos was associated with higher risk of HL overall (odds ratio, OR = 2.58, 95% CI 1.06–6.25), and in age-stratified analyses, additional associations were observed for HL at younger ages (≤ 40 years of age) with the organophosphates dimethoate (OR_age≤40 = 3.43, 95% CI 1.04–11.34) and malathion (OR_age≤40 = 1.91, 95% CI 1.07–3.43)[6]. In the Canadian case–control study alone, elevated risks had been reported in association with the phenoxy herbicide dichlorprop (OR = 6.35, 95% CI 1.56–25.92)[7], which was not assessed when pooled with the US studies, as well as with the organophosphate insecticide chlorpyrifos (OR = 5.26, 95% CI 1.56–17.79)[8], which was diminished after pooling (OR = 1.83, 95% CI 0.69–4.89).

To explore associations in a prospective study, including active ingredients not previously investigated, we examined ever vs. never occupational use of 13 pesticide chemical groups and 22 active ingredients in relation to HL incidence in three large agricultural cohorts from France, Norway and the USA participating in the AGRICOH consortium (https://​agricoh.​iarc.​fr/​).

Details on the study design and exposure assessment were published previously [9, 10] and are described briefly below.

Across the three cohorts, a total of 316 270 farmers (127 282 AGRICAN, 137 821 CNAP and 51 167 AHS; Supplementary Fig. 1) contributed 3 574 815 person-years. The median follow-up duration was 16 years overall, ranging from 4 years in AGRICAN to 19 years in CNAP (Table 1). Most participants were male (75%), but the proportion of males varied between cohorts from 56% in AGRICAN to 84% in CNAP and 97% in AHS. The median age at the start of follow-up was 46 years in AHS, 51 years in CNAP and 67 years in AGRICAN. The proportion of current or former smokers was 35% in AGRICAN and 47% in AHS; information on smoking status was not available in CNAP. We estimated that fewer than half of the farm owners in the CNAP cohort had ever used one of the selected active ingredients or chemical groups (45%), while a greater proportion of AGRICAN farmers (67%) and almost all AHS private applicators (99%) were ever users (Table 1). Additional details on the characteristics of the study participants [10] and their exposure to pesticides [9] have been published.

Among the various chemical classes evaluated, organophosphate insecticides were the most prevalent (used by 59% of farmers), followed by carbamate and organochlorine insecticides (53% and 52% of farmers, respectively; Table 1). The most prevalent active ingredients were: the organophosphate insecticide malathion (used by 46% of farmers), the phenoxy herbicide 2,4-D (45%) and glyphosate (45%). Phenoxy herbicides were used the longest (e.g. 2,4-D was used for a median of 24 years, range: 1 to 56 years) whereas newer pesticides such as pyrethroids were used for a shorter duration, with less variability estimated between farmers (e.g. deltamethrin was used for a median of 9 years, range: 1 to 31 years). The estimated exposure prevalence and duration of use of each active ingredient and chemical group can be found in Supplementary Table 1 and in more detail in Brouwer et al. 2016 Supplementary Table S5 [9].

In the combined population, a total of 91 incident HLs were observed, of which 80 (88%) were classical and the remaining 11 (12%) were nodular lymphocyte predominant. The median age at diagnosis was 58 years (range: 26 to 88 years), though this varied from 43 years in AHS to 72 years in AGRICAN, reflecting differences in median age at the start of follow-up in each cohort (Table 1). Only 14 HL cases occurred before the age of 40 years.

We did not observe any statistically significant associations between the 22 active ingredients or 13 chemical groups examined and the risk of HL (Table 2). The highest risks of HL overall were observed among ever users of the pyrethroid insecticides deltamethrin and esfenvalerate, with meta-HRs (and 95% CIs) of 1.86 (0.76–4.52) and 1.86 (0.78–4.43), respectively. Inverse associations of similar magnitude were observed for the organophosphate insecticide parathion (0.53, 0.17–1.66) and the broad-spectrum herbicide glyphosate (0.58, 0.29–1.18). In general, compared to the active ingredients, the meta-HRs for the 13 chemical groups were closer to the null, with point estimates ranging from 1.01 to 1.29 for positive associations and from 0.64 to 0.98 for inverse associations (Table 2). For most meta-estimates, we did not observe evidence of heterogeneity, with a few exceptions (malathion, chloroacetanilide and dinitroaniline herbicides, I² = 41–49%).

In secondary analyses, the risk of HL diagnosed at ≥ 40 years of age was two-fold in association with dicamba (meta-HR_age≥40 = 2.04, 95% CI 0.93–4.50) and inversely associated with glyphosate (0.46, 0.20–1.07); all confidence intervals crossed the null. There were too few exposed HL cases younger than 40 to report HRs for this outcome in relation to any of the active ingredients or chemical groups examined. Among the few instances in which there was a sufficient number of exposed cases in each category of exposure duration (< or ≥ 16 years), no associations or linear trends were observed (Supplementary Table 2).

In this exploratory prospective analysis of three agricultural cohorts, we did not observe statistically significant associations between any of the 22 active ingredients or 13 chemical groups and the risk of HL. We observed some slightly elevated and some slightly diminished hazard ratios with wide confidence intervals that crossed the null. The highest risks of HL overall were observed for the pyrethroids deltamethrin and esfenvalerate, and inverse associations of similar magnitude were observed for parathion and glyphosate. Farmers who had ever used dicamba had approximately two-fold higher risk of developing HL at ≥ 40 years of age. While mechanistic evidence for these pesticides as potential carcinogens is moderate (dicamba) [18‐25] to strong (pyrethroids) [26], there have been few epidemiological investigations, and none have been conclusive [6, 27]. To our knowledge, ours is the first epidemiological study investigating associations between synthetic pyrethroids (permethrin, deltamethrin, and esfenvalerate) and the risk of HL.

The low incidence of HL and prevalence of specific active ingredients contributed to the low precision of our estimates, posing challenges for reporting and comparing results. We could not examine previously reported positive findings in the North American pooled case–control studies due to not having assessed exposure to the certain active ingredients (dichlorprop and dimethoate) or not having a sufficient number of cases exposed to terbufos or cases under 40 years exposed to malathion [6, 7]. A hospital-based French case–control study had reported positive associations between HL and use of chemical groups we did not assess (pyrethrin insecticides, triazole fungicides, and phenoline, picoline and amide herbicides), as well as with groups not associated with HL in our analysis (organochlorine insecticides, carbamate fungicides, and urea herbicides) [28]. However, most of these previously reported elevated risks were based on relatively few exposed cases (4 to 8), with the exception of malathion, which was associated with HL at ≤ 40 years in the North American Pooled Project based on 26 exposed cases [6].

Differences in the age distribution between studies and underlying etiological differences for HL by age may explain some inconsistencies between our findings and the extant literature, therefore reporting age-stratified results may facilitate comparability between studies. Examining risk factors for HL by histological subtype and tumour EBV status is preferable, but the rare nature of this cancer and lack of EBV status information have hindered such analyses in our study and in previous studies. However, some studies used age as a proxy, since the proportion of EBV + tumours is slightly higher among older adults than among younger adults [4].

In this exploratory analysis, we estimated a large number of associations and therefore cannot rule out that some of the suggestive positive or inverse findings occurred simply by chance; thus, our results should be interpreted with caution. Since semi-quantitative exposure information (e.g. probability, frequency) was available in only one of the three cohorts (AHS), we reported results by ever vs. never and duration of use. However, ever vs. never represents a meaningful exposure contrast among farmers, since they are exposed at higher levels than the general population and tend to use a particular active ingredient for several years. Follow-up times varied between cohorts as well as across specific pesticides, since their use has changed over time (for example, due to pesticide bans and replacements). The prevalence of pesticide use overall and of specific active ingredients varied between cohorts (Supplementary Table 1) due to the different recruitment strategies and predominant crops of each country. The different recruitment strategies combined with the gendered nature of farm ownership and farming tasks also led to different proportions of females in each cohort. Despite these differences, we did not find much evidence of statistical heterogeneity between cohorts in the meta-estimates; however, like other indices of heterogeneity, the I² statistic is biased when the number of meta-analysed studies is small [29]. Furthermore, non-differential exposure misclassification, particularly from the use of crop-exposure matrices, may have biased our estimates towards the null [9]. Refinement of exposure assessment is ongoing, including the addition of probability and frequency estimates and consideration of exposure through tasks other than pesticide application (e.g. crop picking). Since men are more likely to apply pesticides than female farmers, who tend to be exposed to pesticides through contact with recently-treated crops, this will reduce misclassification and account for differences in exposure patterns between male and female farmers.

Despite these limitations, this analysis represents the best available data assessing the relationship between exposure to specific pesticides and the risk of HL. The prospective nature of the data avoids recall bias, which may have affected previous findings from case–control studies. Since the analysis is restricted to farmers, it overcomes bias from the lower mortality and cancer incidence commonly observed among farmers compared to the general population, often attributed to their lower smoking rates and possibly higher levels of physical activity [30]. We also made efforts to adjust for exposure to other pesticides, by controlling for individual active ingredients as well as animal production, which likely involves pesticide use. However, we cannot rule out residual confounding due to unmeasured potential confounders, such as EBV infection and genetic predisposition, though there is no reason to believe these would be correlated with occupational use of specific pesticides. Furthermore, most HL cases occurred at older ages, thus we could not explore associations with adolescent or young adult HL. Future work using larger databases with even longer years of follow-up is needed to investigate these associations further, with more refined exposure assessment methods, and, if possible, ascertainment of tumour EBV status, histological subtype, and greater numbers of younger HL cases.