Agricultural Pesticides and Shingles Risk in a Prospective Cohort of Licensed Pesticide Applicators

Abstract

Background:

Some pesticides are immunotoxic and have been associated with an increased risk of immune-mediated diseases. The risk of shingles, the clinical reactivation of varicella-zoster virus, increases with aging and immunosuppression; little is known about its associations with pesticides.

Objective:

We examined the use of agricultural pesticides in relation to incident shingles in a prospective cohort of licensed pesticide applicators.

Methods:

The study sample included 12,820 (97% male) farmers (enrolled in 1993–1997 in North Carolina and Iowa), who were followed for a median of 12 y (interquartile range: 11–13). Shingles was self-reported at enrollment and at follow-up. We evaluated ever-use of 48 agricultural pesticides reported at study enrollment in relation to shingles risk and considered exposure–response for intensity-weighted lifetime days (IWLDs) of use. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated using Cox proportional hazard models, adjusting for state, and allowing estimates to vary by median attained age (60 y).

Results:

Incident shingles was reported by 590 participants. Associations were positive ($\text{HRs}>1.2$) for ever- vs. never-use of eight insecticides, three fumigants, two fungicides, and five herbicides, and exposure–response trends were seen across increasing quartiles (Q3 and $\mathrm{Q}4>\mathrm{Q}1$) or tertiles (T3 and $\mathrm{T}2>\mathrm{T}1$) of IWLDs for four insecticides [permethrin (crops), coumaphos, malathion, and lindane], two fumigants (carbon tetrachloride/carbon disulfide and methyl bromide), and three herbicides [alachlor, trifluralin ($<60\text{\hspace{0.17em}years of age}$) and 2,4-dichlorophenoxyacetic acid]. Shingles was not associated with total years or days per year mixed or applied any pesticides, but in older participants, shingles was associated with a history of a high pesticide exposure event [$\text{HR}=1.89$ (95% CI: 1.45, 2.45)].

Conclusions:

Several specific pesticides were associated with increased risk of shingles in farmers, especially at higher levels of cumulative use. These novel findings, if replicated in other populations, could have broader implications for the potential effects of pesticides on vaccine efficacy and susceptibility to other infections. https://doi.org/10.1289/EHP7797

Introduction

Incidence of herpes zoster (commonly known as shingles), the symptomatic loss of latency of varicella-zoster virus (VZV), is estimated at 3–5 cases/1,000-person years in the United States, with unexplained increasing rates across several decades (Hales et al. 2013; Kawai et al. 2016; Wolfson et al. 2020). Aging is an established risk factor for shingles; rates rise to 6–8 cases/1,000 at age 60 y and continue to increase with age owing to diminished cellular immunity or immunosenescence, with a lifetime risk of up to 30% in the absence of VZV vaccination (Yawn et al. 2007). In addition to age, shingles risk varies by sex and race/ethnicity. Women have an elevated risk of shingles compared with men, whereas African Americans have lower risk compared with Whites (Kawai and Yawn 2017). Shingles risk has also been associated with a family history of shingles, a personal history of diagnosis with certain diseases (autoimmune diseases, asthma, diabetes, and some solid and lymphoid cancers), and the use of immunosuppressive medications (Forbes et al. 2014; Gershon et al. 2015).

Exposures that impact the immune system, such as smoking, psychosocial stress, ultraviolet (UV) light, and sleep disturbance have also been evaluated as potential risk factors for shingles (Kawai and Yawn 2017, 2020; Marin et al. 2016; Schmidt et al. 2018). Research on these and other modifiable risk factors is relatively sparse, especially for environmental chemicals. In one study, residents near a North Carolina Superfund site (a pesticide dump site with elevated levels of organochlorine pesticides, volatile organic compounds, and metals) were more likely to report shingles at a younger age than residents of a nearby community (Arndt et al. 1999); residents closer to the site also had lower mitogen-lymphoproliferative activity than those living farther away (Vine et al. 2000), suggesting the possibility of immunosuppression due to pesticide or other chemical exposures. To our knowledge, other research on shingles and pesticides is limited to one case–control study that reported no association between shingles and general pesticide use (Marin et al. 2016).

Individual pesticides with different mechanisms of action may contribute to the loss of VZV latency through a variety of pathways impacting immunity (Corsini et al. 2013; Mokarizadeh et al. 2015; Ryu et al. 2018). Research on aging and zoster has established the importance of (T) cell-mediated immunity to VZV (Gershon et al. 2015; Levin et al. 2003; Qi et al. 2016); pesticides may contribute to zoster risk through direct immunotoxicity by, for example, effects on cell-mediated immunity and also indirectly through endocrine or neuroimmune pathways impacting the immune response (Bansal et al. 2018; Galloway and Handy 2003; Tarkowski et al. 2004; Weinberg et al. 2019). In the Agricultural Health Study (AHS), a prospective cohort study of licensed pesticide applicators (mostly farmers) and their spouses from North Carolina and Iowa, past use of specific pesticides has been associated with risk of immune-mediated diseases, such as rheumatoid arthritis (RA) and lymphoid malignancies (Alavanja et al. 2014; Meyer et al. 2017; Parks et al. 2016). Here we extend this research to examine ever and cumulative use of pesticides at enrollment in relation to shingles risk during follow-up among AHS private applicators.

Methods

Population and Sample

The AHS has been previously described (Alavanja et al. 1996). Between 1993 and 1997, 52,394 private pesticide applicators (mostly White male farmers) applying for their pesticide application license in North Carolina or Iowa enrolled in the study by completing a questionnaire at the licensing site. Enrolled participants were given a second questionnaire to complete at home (returned by 44%). Shingles diagnoses were queried on the baseline take-home questionnaire and in the second follow-up survey in 2005–2010 (completed by 46% of enrolled applicators). Pesticide and covariate data used in the present analysis were collected on the baseline enrollment and take-home questionnaires.

Shingles Case Ascertainment and Health Covariates

At enrollment (1993–1997), participants were asked if a doctor had ever told them they had shingles and, if so, their age at first diagnosis ($<20$, 20–39, 40–59, $\ge 60\mathrm{y}$). At follow-up (2005–2010), participants were asked if they had been diagnosed with shingles in the past 10 y and the age of their most recent diagnosis. A total of 13,014 AHS participants had data on shingles from both the enrollment take-home and follow-up (2005–2010) questionnaires; 709 reported shingles at enrollment and 615 reported shingles for the first time at follow-up (Figure S1). Of the cases reported at enrollment, 95 also reported an episode of shingles during follow-up. These potentially recurrent cases were not included in our analysis of incident shingles. Of the remaining cases first reported at follow-up, we excluded 2 participants who reported an age at diagnosis before enrollment and 23 who were missing diagnosis age. The primary analysis sample thus included a total of 590 incident cases and 11,690 noncases (total 12,280). Restricted samples were analyzed in sensitivity analyses a) excluding those with a prevalent diagnosis of autoimmune disease or leukemia/lymphoma (532 cases and 10,045 noncases) and b) limited to incident shingles during the first 5 y of follow-up (149 cases and 12,131 noncases).

Baseline demographic characteristics were identified based on enrollment questionnaires, including self-reported age, state, race/ethnicity, and education. We identified prevalent autoimmune diagnoses based on enrollment, take-home, and follow-up questions on RA, lupus, Sjögren’s, scleroderma or sarcoidosis, and multiple sclerosis, with prevalence defined as a diagnosis age prior to enrollment or incident shingles diagnosis age, and self-reported or registry-identified leukemia or lymphoma (i.e., based on state cancer registry data previously linked to the cohort; https://aghealth.nih.gov/collaboration/outcomes.html).

Pesticide Use

History of general pesticide use at AHS enrollment was based on participant responses to the question: “During your lifetime, have you ever personally mixed or applied any pesticides?” This was followed by questions specifying years participants had personally mixed or applied (categorical responses $\le 1$, 2–5, 11–20, 21–30, and $>30\mathrm{y}$) and, during those years, days per year they had personally mixed or applied pesticides ($<5$, 5–9, 10–19, 20–39, 40–59, 60–150, $>150\mathrm{d}$). On the enrollment questionnaire, participants were also asked: as a result of using pesticides, how many times they had seen a doctor or been hospitalized (options included never, once, twice, three or more times); positive response categories were combined as “ever” seen a doctor or been hospitalized due to pesticide use. In the take-home questionnaire, participants were asked if they had ever had an experience or incident while using any type of pesticide that caused them an unusually high personal exposure, and they were also asked if a doctor had ever told them they had pesticide poisoning.

At enrollment, applicators were asked questions about their lifetime use of any agricultural pesticides, including the total number of years and days per year, and ever-use of 50 specific pesticides. For 22 pesticides, data were collected on the duration (years) and frequency (days per year) of use. Details on the remaining 28 pesticides were collected in the take-home questionnaire. Cumulative lifetime days of specific pesticide use was calculated by multiplying the midpoint of categories of duration and frequency; the product term was multiplied by an intensity score to account for factors that influence personal exposures—including repairing pesticide application equipment, application methods, and the use of personal protective equipment—to derive the variable for cumulative intensity-weighted lifetime days (IWLDs) (Coble et al. 2011; Dosemeci et al. 2002).

For sensitivity analyses, we constructed a variable to focus on potential chronic effects based on past pesticide use by identifying those with the preenrollment use only separately from those reporting use in the past year. At enrollment, participants reporting ever-use were asked to mark if they had used the pesticide in the past year [except for the organochlorine insecticides and herbicides 2,4,5-trichlorophenoxyacetic acid (2,4,5-T) and 2,4,5-trichlorophenoxy) propionic acid (2,4,5-TP)], which were no longer approved for use in the United States). Using this information, we subdivided individuals into preenrollment users (who reported ever-use, but not in the past year at enrollment), and enrollment users (who reported using the pesticides in the past year regardless of whether they also used the product in prior years).

Statistical Analyses

Analyses were conducted using AHS data files releases: P1REL201701.00, P2REL201701.00, P3REL201809.00, and AHSREL201706.00 in SAS (version 9.4; SAS Institute Inc.).

We first evaluated frequencies of sociodemographic covariates and state, using logistic regression to calculate age-adjusted odds ratios (ORs) and 95% confidence intervals (CIs). Cox proportional hazard regression models were used to calculate hazard ratios (HRs) and 95% CIs, with age as the time scale, to estimate pesticide associations. The analysis sample included 590 incident cases with complete data on age, including diagnosis age, sex, and state; 6 cases reported a shingles diagnosis age that was the same as their enrollment age and were assigned a value of 0.5 y follow-up to be retained in the models. Time at risk was accrued until shingles diagnosis, death, and loss or end of follow-up. Risk was allowed to vary by median attained age when the assumption of proportional hazards was not met, resulting in age-stratified HRs for some models. The proportional hazards assumption was tested using an interaction term between attained age and each specific exposure separately ($p\le 0.10$).

A priori, we determined that our pesticide association models would include state (North Carolina or Iowa), due to differences in agricultural crops and practices. Given the small number of females and non-Whites, these covariates were not evaluated as confounders for pesticide associations. For each pesticide association, we evaluated education as a potential confounder, allowing it to enter the strata statement as needed. We then derived correlation coefficients between all pairs of individual pesticides (ever-use) and evaluated potential confounding by correlated pesticides (coefficient $>0.30$), assessed one at a time (Table S1). Variables (i.e., education or correlated pesticides) were retained in the model when we saw either a $>10\%$ change in main pesticide effect estimate or if the proportional hazard assumption was violated (or no longer violated). If multiple factors met these criteria, then covariates were run together in a single model. Thus, models used to estimate associations of shingles with specific pesticides were sometimes adjusted for education or one or more correlated pesticides. We performed a complete case analysis, so 420 noncases and 23 incident cases were excluded from models adjusted for education due to missing data for education. Numbers were also reduced for some models due to missing data on correlated pesticides being included as covariates (range: 2–9 cases and 57–116 noncases). Final covariate adjustments by education or correlated pesticides are noted in table footnotes.

Associations were estimated for pesticides with at least 10 exposed cases (e.g., $\ge 2\%$ of 590). For pesticides with at least 20 exposed cases, we considered exposure–response for shingles risk across categories of IWLD [by median split, tertiles (T), or quartiles (Q)], to achieve a minimum of 10 exposed cases per level), calculating HRs across increasing categories, including the unexposed. Findings from the main analyses are highlighted in the text when HRs were elevated ($>1.2$) or inverse ($<0.80$). Potential exposure response for specific pesticides were noted when IWLD Q3 and $\mathrm{Q}4>\mathrm{Q}1$; or T3 and $\mathrm{T}2>\mathrm{T}1$) or when one or more elevated HRs from the upper quartiles/tertiles were statistically significant ($p<0.05$).

We conducted two sensitivity analyses. First, for analyses of incident shingles in association with general pesticide exposures and ever-/never-exposure to specific pesticides, we excluded participants who reported prevalent autoimmune diseases or a history of leukemia/lymphoma before shingles diagnosis (645 noncases, 58 cases) because these conditions may increase the risk of shingles and could also influence pesticide use/reporting or be on the causal pathway. We also examined ever-use associations in the first 5 y of follow-up, stratified by median attained age of 53 y when proportional hazards were violated, also looking at the timing of pesticide use relative to enrollment with a specific focus on preenrollment use, separately from use reported in the year prior to enrollment. Other cases identified later in follow-up remained part of the risk data set.

Results

In our primary analysis sample of 12,280 study participants at risk of shingles (median follow-up of 12 y; IQR: 11–13), we identified 590 incident cases with a known diagnosis age (median diagnosis at 60 years of age (IQR: 55–70). Sample characteristics are shown Table 1. Odds of developing shingles were higher for older individuals, and after adjusting for age, ORs were elevated for women and those with a history of leukemia or lymphoma or autoimmune diseases. An inverse association was seen for non-Whites (mostly African Americans).

Table 1.

Enrollment characteristics of AHS private applicators with incident shingles.

Characteristics	Noncases ($N=\mathrm{11,690}$)	Cases ($N=590$)	OR (95% CI)a
	$n$ (%)	$n$ (%)
Age (y)
$<45$	4,759 (41)	123 (21)	1.00 (Ref)
45–54	3,068 (26)	147 (25)	1.85 (1.45, 2.37)
55–64	2,649 (23)	228 (39)	3.33 (2.66, 4.17)
$\ge 65$	1,214 (10)	92 (15)	2.93 (2.22, 3.87)
State
Iowa	8,037 (69)	383 (65)	1.00 (Ref)
North Carolina	3,653 (31)	207 (35)	1.03 (0.86, 1.22)
Gender
Male	11,393 (97)	571 (97)	1.00 (Ref)
Female	297 (3)	19 (3)	1.29 (0.80, 2.08)
Race/ethnicityb
White	11,287 (97)	576 (98)	1.00 (Ref)
African American/other	230 (2)	7 (1)	0.55 (0.26, 1.17)
Educationb
$\le \text{High school}$	6,008 (52)	334 (57)	1.00 (Ref)
$>\text{High school}$	5,262 (45)	233 (39)	0.97 (0.82, 1.16)
Health conditionsc
Autoimmune disease	631 (5)	54 (9)	1.33 (0.99, 1.79)
Leukemia or lymphoma	20 (0)	4 (1)	3.54 (1.19, 10.56)

Note: AHS, Agricultural Health Study; CI, confidence interval; OR, odds ratio; Ref, reference.

^aLogistic regression models adjusted by continuous age except for age categories.

^bComplete enrollment age available on all participants. Most non-Whites were African American (103 noncases, 3 cases). Missing data on race for 173 noncases and 7 cases; education for 420 non cases and 23 cases, smoking for 47 noncases and 1 case, and alcohol for 465 noncases and 26 cases; percentages may not sum to 100 owing to missing values.

^cHealth conditions self-reported at AHS enrollment or prior to age at shingles diagnosis; autoimmune diseases included lupus, rheumatoid arthritis, multiple sclerosis, Sjögren’s syndrome, scleroderma, and sarcoidosis.

Incident shingles was not associated with the number of lifetime years or days per year that participants mixed or applied any pesticides (Table 2) but was associated ($\text{HRs}>1.2$) with having been seen by a doctor or hospitalized due to pesticide use and with a self-reported diagnosis of pesticide poisoning. In addition, shingles was positively associated with a self-reported history of a high pesticide exposure event in participants who were $>60\text{\hspace{0.17em}years of age}$. These findings were consistent after excluding those with a history of lymphoid malignancies or autoimmune disease at enrollment or prior to shingles enrollment.

Table 2.

Lifetime days of pesticide use and high pesticide exposures.

Enrollment history	Age (y)	Total sample			Without baseline autoimmune/lymphoid malignanciesa
		Noncases ($N=\mathrm{11,690}$) %	Cases ($N=590$) %	HR (95% CI)	Noncases ($N=\mathrm{11,045}$) %	Cases ($N=532$) %	HR (95% CI)
Mixed or applied pesticides
Total years
$\le 5$	—	12	8	1.00 (Ref)	12	9	1.00 (Ref)
6–10	—	14	11	1.06 (0.72, 1.55)	14	11	0.96 (0.64, 1.45)
11–20	—	33	29	1.15 (0.83, 1.60)	33	29	1.08 (0.76, 1.51)
21–30	—	26	29	1.06 (0.76, 1.48)	26	28	0.97 (0.68, 1.37)
$>30$	—	15	23	1.03 (0.73, 1.46)	14	22	0.97 (0.68, 1.40)
Average days per year
$<5$	—	19	20	1.00 (Ref)	18	20	1.00 (Ref)
5–9	—	25	28	1.11 (0.86, 1.42)	26	29	1.05 (0.82, 1.39)
10–19	—	32	32	1.09 (0.86, 1.39)	32	32	1.00 (0.77, 1.29)
$\ge 20$	—	25	20	0.99 (0.76, 1.29)	25	20	0.93 (0.70, 1.24)
As a result of pesticides
Seen a doctor	—	7	9	1.26 (0.94, 1.69)	7	8	1.23 (0.90, 1.68)
Been hospitalized	—	1	2	1.79 (0.99, 3.26)	1	1	—
Diagnosed with pesticide poisoning	—	2	4	1.53 (0.99, 2.37)	2	3	1.48 (0.90, 2.43)
High pesticide exposure event	$\le 60$	18	18	1.12 (0.80, 1.56)	15	20	1.57 (1.26, 1.95)
	$>60$	13	21	1.89 (1.45, 2.45)	—	—	—

Note: Estimates were derived using Cox proportional hazard models with age as the time scale, limited to at least 10 exposed cases. When proportional hazards were not met, HRs were stratified by median attained age (60 y). All models were adjusted for state, with additional adjustment for education or correlated pesticides as indicated for specific exposures. —, Not applicable; CI, confidence interval; HR, hazard ratio; Ref, reference.

^aSensitivity analysis excluded study participants with prevalent autoimmune diseases or leukemia/lymphoma (645 controls and 58 cases).

Ever-use of 18 specific pesticides reported at AHS enrollment was associated with higher incidence of shingles, with $\text{HRs}>1.2$ (Table 3). These included 8 insecticides [permethrin use on crops, coumaphos, fonofos (among those with an attained age of $>60\mathrm{y}$), malathion, parathion, phorate (at $\le 60\text{\hspace{0.17em}years}$ of age), chlordane (at $>60\text{\hspace{0.17em}years}$ of age), and lindane]. Associations were also seen for 3 fumigants (carbon tetrachloride/disulfide, ethylene dibromide, and methyl bromide), 2 fungicides [benomyl and maneb/mancozeb (at $\le 60\text{\hspace{0.17em}years}$ of age), and 5 herbicides [2,4-dichlorophenoxyacetic acid (2,4-D) and 2,4,5-TP, cyanazine, glyphosate, paraquat (at $>60\text{\hspace{0.17em}years}$ of age)]. We did not hypothesize a decreased risk of shingles associated with specific pesticides but noted two inverse associations ($\text{HR}<0.80$), including chlordane (at $\le 60\text{\hspace{0.17em}years}$ of age) and chlorpyrifos (at $>60\text{\hspace{0.17em}years}$ of age). In general, associations were similar after excluding participants with leukemia or lymphoma (20 noncases, 4 cases) or autoimmune diseases (631 noncases, 54 cases) at enrollment or prior to shingles diagnosis, with the most notable differences resulting from age stratification in the main model but not the restricted model (e.g., no association with chlorpyrifos in the population as a whole) or vice versa [a positive association with terbufos (at $\le 60\text{\hspace{0.17em}years}$ of age)].

Table 3.

Association between ever-use of pesticides and incident shingles.

Specific pesticides	Total sample				Without baseline autoimmune/lymphoid malignanciesa
	Age (y)	Noncases	Cases	HR (95% CI)b	Age (y)	Noncases	Cases	HR (95% CI)b
		$N=\mathrm{11,690}$ %	($N=590$) %			($N=\mathrm{10,045}$) %	($N=532$) %
Insecticides
Pyrethroids
Permethrin (crops)	—	13	14	1.25 (0.97, 1.60)	—	13	14	1.24 (0.95, 1.60)
Permethrin (animals)	—	15	13	1.02 (0.79, 1.32)	—	15	13	1.05 (0.80, 1.37)
Organophosphates
Coumaphos	—	10	14	1.35 (1.06, 1.73)	—	10	14	1.40 (1.08, 1.81)
Dichlorvos	—	13	14	0.97 (0.76, 1.25)	—	13	14	1.02 (0.79, 1.32)
Chlorpyrifos	$\le 60$	46	49	1.18 (0.91, 1.53)	—	43	39	0.95 (0.80, 1.14)
	$>60$	40	33	0.78 (0.62, 0.98)	—	—	—	—
Diazinon	—	37	44	1.15 (0.97, 1.36)	—	38	43	1.12 (0.94, 1.34)
Fonofos	$\le 60$	22	22	0.91 (0.65, 1.26)	$\le 60$	23	21	0.88 (0.63, 1.23)
	$>60$	24	27	1.24 (0.95, 1.62)	$>60$	24	28	1.27 (0.96, 1.67)
Malathion	—	78	84	1.36 (1.09, 1.71)	—	78	84	1.32 (1.05, 1.67)
Parathion	—	17	24	1.30 (1.07, 1.59)	—	17	24	1.36 (1.10, 1.67)
Phorateb	$\le 60$	38	44	1.20 (0.91, 1.60)	$\le 60$	37	44	1.22 (0.91, 1.64)
	$>60$	43	40	0.83 (0.65, 1.06)	$>60$	43	39	0.79 (0.61, 1.02)
Terbufos	—	40	42	1.12 (0.93, 1.34)	$\le 60$	42	49	1.59 (1.20, 2.10)
	—	—	—	—	$>60$	39	36	0.83 (0.64, 1.08)
Carbamates
Aldicarb	—	10	10	1.00 (0.75, 1.34)	—	10	11	1.09 (0.81, 1.47)
Carbaryl	—	62	70	1.12 (0.93, 1.35)	—	63	68	1.07 (0.88, 1.30)
Carbofuran	—	31	32	0.94 (0.79, 1.13)	—	30	30	0.91 (0.75, 1.10)
Organochlorines
Aldrin	—	25	36	1.12 (0.93, 1.35)	—	23	34	1.13 (0.92, 1.37)
Chlordaneb	$\le 60$	20	26	0.76 (0.56, 1.04)	$\le 60$	20	26	0.75 (0.53, 1.03)
	$>60$	45	53	1.29 (1.04, 1.61)	$>60$	45	53	1.28 (1.02, 1.62)
Dieldrin	—	9	13	0.93 (0.72, 1.20)	—	9	13	0.94 (0.72, 1.23)
DDT	—	32	47	1.01 (0.84, 1.21)	—	31	46	1.02 (0.84, 1.23)
Heptachlor	—	21	27	0.92 (0.75, 1.13)	—	21	27	0.90 (0.73, 1.12)
Lindane	—	26	33	1.25 (1.04, 1.49)	—	26	32	1.21 (1.00, 1.46)
Toxaphene	—	18	22	0.98 (0.81, 1.20)	—	17	21	0.94 (0.76, 1.16)
Fumigants
Methylbromide	—	14	18	1.34 (1.04, 1.74)	—	14	19	1.36 (1.03, 1.80)
Aluminum phosphide	—	6	6	1.14 (0.82, 1.59)	—	6	7	1.24 (0.87, 1.77)
Carbon tetrachloride/disulfide	—	8	13	1.22 (0.95, 1.56)	—	8	13	1.27 (0.97, 1.67)
Ethylene dibromide	—	7	10	1.46 (1.10, 1.93)	—	6	10	1.53 (1.13, 2.08)
Fungicides
Benomylb	—	11	14	1.34 (1.03, 1.73)	—	11	14	1.30 (0.99, 1.70)
Captan	—	12	13	1.12 (0.87, 1.44)	—	12	13	1.12 (0.85, 1.47)
Chlorothalonil	—	7	6	0.89 (0.62, 1.26)	—	7	6	0.86 (0.59, 1.25)
Maneb/mancozeb	$\le 60$	9	18	1.93 (1.36, 2.74)	$\le 60$	9	18	2.01 (1.40, 2.89)
	$>60$	14	15	1.10 (0.80, 1.51)	$>60$	14	14	1.11 (0.79, 1.56)
Metalaxylc	$\le 60$	23	30	1.14 (0.84, 1.56)	$\le 60$	23	30	1.24 (0.90, 1.71)
	$>60$	26	23	0.87 (0.66, 1.14)	$>60$	26	24	0.86 (0.64, 1.16)
Herbicides
Analides/analines
Alachlor	—	57	60	1.11 (0.93, 1.32)	—	56	59	1.06 (0.88, 1.28)
Metolachlorb	—	48	47	1.03 (0.86, 1.23)	$\le 60$	50	53	1.29 (0.97, 1.71)
					$>60$	45	43	0.90 (0.70, 1.16)
Pendimethalin	—	46	46	1.10 (0.94, 1.30)	—	46	45	1.06 (0.89, 1.26)
Trifluralin	—	54	58	1.18 (0.98, 1.41)	—	54	59	1.21 (1.00, 1.46)
Phenoxy
2,4-D	—	79	84	1.30 (1.03, 1.64)	—	79	84	1.26 (0.99, 1.61)
2,4,5-T	—	29	39	1.10 (0.92, 1.31)	—	29	39	1.15 (0.96, 1.39)
2,4,5-TP	—	12	18	1.28 (1.03, 1.59)	—	12	18	1.34 (1.07, 1.68)
Thiocarbamate
EPTC	—	22	20	1.01 (0.81, 1.25)	—	21	21	1.07 (0.85, 1.34)
Triazine/triazinone
Atrazine	—	74	74	1.00 (0.84, 1.22)	—	74	73	1.05 (0.85, 1.30)
Cyanazine	—	45	48	1.22 (1.01, 1.47)	—	45	49	1.29 (1.06, 1.58)
Metribuzin	—	50	53	1.18 (0.98, 1.41)	—	50	53	1.20 (0.99, 1.45)
Other
Butylate	—	37	39	1.12 (0.94, 1.33)	—	37	39	1.12 (0.93, 1.34)
Chlorymuron ethyl	—	40	40	1.12 (0.95, 1.33)	—	40	39	1.08 (0.90, 1.29)
Dicamba	—	54	54	1.05 (0.86, 1.28)	—	55	54	1.08 (0.87, 1.33)
Glyphosate	—	78	83	1.43 (1.14, 1.78)	—	78	83	1.40 (1.11, 1.77)
Imazethapyr	$\le 60$	50	44	0.88 (0.67, 1.16)	—	46	42	1.05 (0.86, 1.29)
	$>60$	40	41	1.18 (0.92, 1.50)	—	—	—	—
Paraquat	$\le 60$	22	27	1.15 (0.84, 1.56)	$\le 60$	22	27	1.20 (0.87, 1.65)
	$>60$	24	30	1.38 (1.08, 1.75)	$>60$	24	31	1.42 (1.10, 1.82)
Petroleum oil/distillates	—	51	53	1.11 (0.94, 1.31)		51	54	1.16 (0.97, 1.38)

Note: Estimates were derived using Cox proportional hazard models with age as the time scale, limited to pesticides with at least 10 exposed cases. When proportional hazards were not met, HRs were stratified by median attained age (60 y). All models were adjusted for state, with additional adjustment for education or correlated pesticides as indicated for specific exposures. —, Not applicable; 2,4-D, 2,4-dichlorophenoxyacetic acid; 2,4,5-T, 2,4,5-trichlorophenoxyacetic acid; 2,4,5-TP, 2,4,5-trichlorophenoxy) propionic acid; CI, confidence interval; DDT, dichlorodiphenyltrichloroethane; EPTC, S-ethyl dipropylthiocarbamate; HR, hazard ratio.

^aSensitivity analysis excludes those with prevalent autoimmune diseases or leukemia/lymphoma (645 controls and 58 cases); age group “overall” is listed when age stratification was no longer justified based on violation of proportional hazards assumption.

^bAdjusted for state and education.

^cAdjusted for state and methylbromide exposure.

Associations between pesticides categorized by IWLDs of exposure (relative to never-exposed) (Table S2) were positive and strongest or statistically significant for the two highest quartiles or the highest tertile of exposure to 10 of 18 pesticides, with HRs for ever-/never-exposure $>1.2$ overall or in one age group, including 4 insecticides [permethrin (use on crops), coumaphos, malathion, and lindane], two fumigants (carbon tetrachloride/carbon disulfide and methyl bromide), 1 fungicide (benomyl), and 3 herbicides (glyphosate, 2,4-D, and paraquat) (Figure 1). Some evidence of a positive exposure–response relation was also evident for pesticides that were not associated with shingles based on ever-/never-exposure, including the insecticides carbaryl and diazinon, the fungicide captan, and the herbicides metribuzin (at $\le 60\text{\hspace{0.17em}years}$ of age) and trifluralin (at $\le 60\mathrm{y}$ of age).

Figure 1.

Risk of shingles associated with greater cumulative intensity-weighted days of specific pesticide use. Pesticides (including insecticides, fumigants, fungicides, and herbicides) with at least 10 exposed cases per quartile (Q) or tertile (T), for selected results showing a pattern of increasing HRs in higher quartiles or tertiles (i.e., Q3 and $\mathrm{Q}4>\mathrm{Q}1$; T3 and $\mathrm{T}2>\mathrm{T}1$, or with one or more HRs for the upper quartiles or tertile with CIs excluding the null. Complete results are shown in Table S3. HRs are based on Cox proportional hazard models and 95% confidence limits with age as the time scale, were allowed to vary by median attained age (60 y). All models include state; metalaxyl was adjusted for methylbromide, metribuzin was adjusted for imazethapyr. Note: 2,4-D, 2,4-dichlorophenoxyacetic acid; HR, hazard ratio.

In sensitivity analyses of 149 cases diagnosed in the first 5 y of follow-up, we confirmed several associations seen for ever-use (Table S3), including associations with use in the preenrollment period only for permethrin use on crops, coumaphos, malathion, lindane, methylbromide, benomyl, cyanazine, and glyphosate (and also for more recent use in the enrollment year for malathion, 2,4-D, and glyphosate). Maneb/mancozeb was positively associated with preenrollment use among those $\le 53\text{\hspace{0.17em}years}$ of age, whereas there were too few exposed cases to estimate associations with preenrollment use in the older age strata (where an overall association was noted for those $>60\text{\hspace{0.17em}years of age}$ in Table 3). Models for chlorpyrifos continued to require age stratification, with an inverse association for preenrollment use among those with attained age $>53\mathrm{y}$ and a contrasting positive association for preenrollment use in younger participants. For pesticides identified based on the exposure–response analyses (Figure 1), shingles in the first 5 y was associated with recent use of diazinon at enrollment ($\text{at}\le 53\text{\hspace{0.17em}years}$ of age), and with preenrollment use of captan and metribuzin.

Discussion

Our findings in this cohort of licensed pesticide applicators provide novel evidence that use of specific pesticides was associated with risk of being diagnosed with shingles over more than a decade of follow-up. These results are supported by evidence of stronger associations for those in the higher quartiles or tertiles of cumulative IWLDs of use. These findings are consistent with the growing evidence on the possible role of pesticides in relation to the risk of chronic immune-mediated outcomes, such as systemic autoimmune disease and hematopoietic cancers (Alavanja et al. 2014; Meyer et al. 2017; Parks et al. 2016). Shingles may be a marker of diminished ability of the immune system to control latent VZV infection. Thus, our findings of risk associated with pesticide exposure warrant further investigation because, similar to risks associated with aging and immunosuppressant medications (Marin et al. 2016; Schmidt et al. 2018), the potential immune effects of pesticides on shingles risk could have broader implications regarding vaccine efficacy and susceptibility to other infections.

Shingles incidence was associated with different types of insecticides. We saw patterns suggestive of exposure response for increasing IWLDs of two organophosphates: coumaphos and malathion. Malathion was the most frequently reported insecticide in the study sample, and it is currently approved for uses ranging from agricultural crops, to home and garden, as well as environmental eradication of insects, whereas coumaphos is a restricted-use insecticide that is typically used on animals and is not approved for residential use. Shingles was also associated with greater use of permethrin on crops but not for use on animals. Permethrin belongs to a large class of pyrethroid insecticides, with widespread and increasing use in recent decades. We also observed positive associations with ever-use of two persistent organochlorine insecticides: chlordane (in older participants) and lindane. Chlordane was used in agriculture starting in the 1950s through the 1980s, and it was widely used through the 1980s for residential termite treatment, but because it is a persistent chemical, long-lasting exposures are possible. Lindane was widely used in agriculture until it was phased out in the early 2000s, but it is still used to treat lice in humans. None of these insecticides were associated with RA in AHS applicators (Meyer et al. 2017). In a biomarker substudy of antinuclear antibodies (ANAs) in 668 AHS applicators, chlordane was among the cyclodienes associated with increased ANA prevalence, whereas malathion was inversely associated with ANA prevalence (Parks et al. 2019). In AHS applicators, insecticides associated with non-Hodgkin lymphoma (NHL) or subtypes of NHL included lindane (follicular lymphoma and total NHL), permethrin (multiple myeloma), and dichlorodiphenyltrichloroethane (DDT) (total NHL) (Alavanja et al. 2014). Several insecticides, including lindane and organophosphates, such as malathion, diazinon, and fonofos, have been associated with NHL in meta-analyses and in analyses of pooled data from North American and European studies (Koutros et al. 2019; Leon et al. 2019; Schinasi and Leon 2014). The International Agency for Research on Cancer (IARC) has concluded there is sufficient evidence for the carcinogenicity of lindane, based in part on associations in humans with NHL (IARC 2018a), and the IARC has classified malathion and diazinon as probably carcinogenic given sufficient evidence in animals and limited evidence of associations in humans, including NHL (IARC 2018b).

We also saw consistent associations, including evidence of increased HRs for higher cumulative use, for two fumigants (methyl bromide and carbon tetrachloride/disulfide) and the fungicide benomyl. Some evidence was also seen for the fungicides maneb/mancozeb (ever-use, in older participants) and for the highest tertile of cumulative use of another fungicide, captan. Methyl bromide is a highly toxic gas used to fumigate soil and is used in agricultural production of fruits and horticultural settings. Fungicides are widely used on similar crops (as well as in residential settings). Use of these chemicals was correlated in our sample, requiring mutual adjustment in some analyses for methybromide (metalaxyl) and benomyl (maneb/mancozeb). These pesticides were not associated with lymphoid cancers in AHS applicators (Alavanja et al. 2014). Methylbromide was associated with elevated ANA prevalence in male applicators (Parks et al. 2019), and maneb/mancozeb was associated with RA in female spouses (Parks et al. 2016), but not in male applicators (Meyer et al. 2017). Most of the fungicides and fumigants associated with shingles in our primary analysis (ever-use) have not been identified as risk factors for NHL, although there is some evidence of associations of with general fungicide use and some specific fungicides, including captan (identified in our IWLD analyses) (Chiu et al. 2006; Schinasi and Leon 2014; Schroeder et al. 2001). Although some of these fungicides and fumigants were relatively uncommon in our analyses, resulting in imprecise estimates of association, the findings across multiple types warrant further investigations in larger samples.

Finally, shingles was associated with ever-use or with IWLDs of several herbicides, including 2,4-D, cyanazine, and glyphosate; paraquat (in those $>60\text{\hspace{0.17em}years of age\hspace{0.17em}}$); and trifluralin (in those $\le 60\text{\hspace{0.17em}years of age}$). Associations with ever-use of other herbicides (e.g., cyanazine, paraquat, and glyphosate) were supported by elevated HRs for higher cumulative use, but patterns of increasing HRs with increasing use were less apparent. Herbicide use has increased in recent decades, with wide use of herbicides such as 2,4-D and glyphosate in both agricultural and residential settings (Kniss 2017). Across quartiles of IWLDs, glyphosate did not meet our criteria for exposure–response (HRs for Q4 and $\mathrm{Q}3>\mathrm{Q}1$); however, we did not account for increasing use in follow-up and exposures preceding shingles diagnosis. Authors of a recent review hypothesized that potential immune effects of 2,4-D may account for possible associations with NHL (Smith et al. 2017), but in another recent review, the IARC concluded that evidence was inadequate to support an association between 2,4-D and NHL (IARC 2018b). In its review on glyphosate, the IARC (2018a) noted limited evidence of cancer in humans, based primarily on associations with NHL; however, glyphosate has not been associated with NHL or autoimmune outcomes in AHS applicators (Andreotti et al. 2018; Meyer et al. 2017; Parks et al. 2019). Given their ongoing widespread and frequent use in agriculture and residential settings, future investigation of specific herbicides and shingles risk is warranted.

Associations between diverse pesticides and shingles in the present study might be explained by the effects on immune control of VZV through a range of potential mechanisms and pathways. Potential mechanisms by which pesticides may impact shingles risk can generally be viewed through the lens of diminished cellular immunity (i.e., cytotoxic T cells), a key causal factor in VZV reactivation (Gershon et al. 2015), although recent studies also suggest a role for natural killer cells and regulatory T-cells (Nikzad et al. 2019; Vukmanovic-Stejic et al. 2015; Weinberg et al. 2019). The well-established effects of aging-related changes in the immune system associated with shingles risk lends plausibility to the idea that past and chronic pesticide exposures could have cumulative and lasting effects on immune control of VZV. (Ogunjimi et al. 2014). In our sensitivity analyses of early incident shingles in the first 5 y of follow-up, associations with specific pesticide use only in the preenrollment period were generally consistent with overall findings for ever-use. This indirect evidence, together with dose–response associations, suggests a role for longer-term, past exposures on susceptibility to herpes zoster reactivation (but does not rule out acute or recent effects of these or other pesticides). Although some pesticides may have direct immunotoxic effects, others may indirectly influence the immune response and shingles risk through neurotoxic or endocrine mechanisms (Bansal et al. 2018; Galloway and Handy 2003). Organophosphate insecticides, for example, affect cholinergic pathways that can interface with the immune system (Galloway and Handy 2003). Pesticides have been shown to have neuroendocrine effects on other chronic viral infections, including cytomegalovirus, which may contribute to immunosenescence and thereby indirectly influence shingles risk (Ryu et al. 2018).

Several insecticides associated with shingles in our study have been shown to affect the immune system. Malathion can have immunosuppressive effects in rodents and other species (Banerjee et al. 1998; Krieger 2010). Chlordane has been linked to markers of increased cell-mediated immunity in an adult rodent model and oxychlordane (the primary toxic metabolite) was associated with increased leukocyte counts in humans (Serdar et al. 2014; Tryphonas et al. 2003), whereas chlordane exposure lead to decreased cytotoxic effects of human natural killer cells in vitro (Reed et al. 2004). Lindane effects may depend on dose and time, with a biphasic response of initial immune enhancement followed by later inhibition (Meera et al. 1992; Mokarizadeh et al. 2015). Limited experimental data suggests permethrin may suppress cellular immunity in vivo, with additional evidence on acute toxicity of permethrin to thymocytes in vitro and potentiated effects by UV exposure or other insecticides (Blaylock et al. 1995; Olgun et al. 2004; Olgun and Misra 2006; Prater et al. 2002, 2003). The immune effects of methyl bromide are difficult to study in experimental in vivo models owing to its high acute toxicity, and information from human observational studies is also limited. Use of ethylenbisdithiocarbamate fungicides (mancozeb and related chemicals, which have low acute toxicity relative to methyl bromide), was associated with an absolute increase in total leukocytes and CD8 (but not CD4) T-lymphocytes in a study of 248 agricultural workers compared with 231 nonexposed controls (Steerenberg et al. 2008). Immune effects may vary for the herbicide 2,4-D, which together with propanil (a common mixture), was associated with thymic atrophy in mice (de la Rosa et al. 2005). In a study of 10 farmers, acute exposure to 2,4-D in a commercial mix with another phenoxy herbicide (4-Chloro-2-methylphenoxyacetic acid) was associated with decreased levels of helper and suppressor T-cells, cytotoxic T-cells, and natural killer cells for up to 2 wk after use and a decreased mitogenic response that persisted 4–6 wk (Faustini et al. 1996).

Our study has limitations as well as strengths. The primary analysis focused on exposures reported at enrollment in association with incident shingles reported after 10 y of follow-up, thus, differential recall by participants who developed shingles is unlikely. Although we were able to examine detailed data on many specific pesticides, exposure assessment remains a challenge for many reasons, including a lack of temporal precision and changes in products and active ingredients used over time. Nondifferential exposure misclassification is likely, however reported pesticide use in the AHS has been shown to be reliable (Blair et al. 2002). Although we adjusted some analyses for correlated pesticides, we cannot rule out the possibility of confounding by other pesticide exposures. Farmers often use more than one pesticide at a time and over a lifetime, and we did not consider joint effects or mixtures. Unmeasured confounders could have influenced our results and we cannot rule out a role for ubiquitous types of farm exposures, such as UV light, which has been previously associated with shingles in men (Kawai et al. 2020). We saw no association with metrics of overall pesticide use as a possible proxy for other types of farm work, and scenarios in which sun exposure would confound only specific pesticide associations seem unlikely. Having a history of high pesticide exposure events was associated with shingles in older farmers, but we had an insufficient number of exposed cases to look at individual pesticides implicated in these events. Differential findings by age could indicate differences in susceptibility but may also be due to chance. Multiple testing can lead to chance results; therefore, we present and evaluate our findings based on strength of association and consistency across sensitivity analyses and in the context of prior evidence for immune effects. We did not estimate associations for pesticides with fewer than 10 exposed cases, so rarer pesticides were not included in the study or results were not shown when smaller subsets were considered in sensitivity analyses.

Our findings are based on self-reported shingles diagnosis, which may be prone to error. However, self-reported shingles was confirmed in medical records for 97% of a small sample of cases in health professionals and nurses cohorts, and self-report was also highly specific compared with Medicare claims data in a cohort of retired persons (98% confirmed), although with diminished sensitivity with more time since diagnosis (Hales et al. 2016; Kawai et al. 2020). Some cases in the AHS may have been missed, but the influence of false-negative cases may be tempered by low incidence rates. At follow-up, participants were asked about age at the most recent diagnosis, so some individuals may have had more than one episode during this time frame. In a clinical study of adults (median of 59 years of age) 6.2% of cases had a recurrence within 8 y of follow-up (Yawn et al. 2011); in our study, about twice as many (13%, $n=95$) of 709 prevalent cases reported a new diagnosis at follow-up, which could be higher owing to mistaken recall of diagnosis age. The estimated incidence based on self-report was within a reasonable range (not shown), and expected associations were seen for age, non-White race/ethnicity (mostly African Americans, for whom lower rates have been described in other populations), female sex, and history of an autoimmune disease or leukemia/lymphoma (Gershon et al. 2015; Kawai and Yawn 2017). We lacked data on other potential risk factors, such as specific medications or major stressors but consider these to be unlikely confounders because they are not expected to be strongly related to specific pesticides. Our case ascertainment occurred prior to widespread vaccination for VZV in the United States (first approved in 2006), so findings may not be generalizable to those receiving vaccination. Furthermore, these results may not extend to other nonagricultural populations and to more diverse samples given that a vast majority of AHS applicators were White men; only 2% were non-White (mostly African American from North Carolina), and 2% were female. Research is warranted on more diverse populations.

Although various biological mechanisms may plausibly link pesticides to altered VZV immunity, it is unclear how longer-term and past exposures might influence shingles risk. Further research is warranted on the chronic and long-lasting immune effects of pesticides on cell-mediated immunity in experimental models and human studies. Some pesticides, such as organochlorines, are persistent and others may transiently impact immune response in the absence of symptoms, and thus repeated or accumulated exposures could contribute to changes in VZV immunity over time. Subclinical VZV reactivation does occur (Gershon et al. 2015), but it is unclear what determines this and whether it provides any endogenous boosting, like vaccination, to reduce clinical VZV reactivation. Shingles risk does not rise abruptly on initiation of immunosuppressive therapies, suggesting a nuanced relationship between immune surveillance, VZV reactivation, and symptoms. We cannot predict whether our findings for shingles will translate to other diseases; however, other factors leading to diminished VZV immunity also increase risk of diverse infectious outcomes and decreased vaccine efficacy (Allen et al. 2020; Day et al. 2020).

Finally, our study showing that several specific agricultural pesticides were associated with shingles in farmers, especially at higher levels of cumulative exposure, supports the idea that some pesticides may influence immunological control of latent varicella-zoster. These novel findings may have important implications, especially for the associations with insecticides. Malathion, for example, is widely used across agricultural, residential, and public health settings, and although organochlorine insecticides are no longer used in many countries, they persist in the body and environment, and some remain a tool for vector control in preventing the spread of infectious diseases. Further research is warranted to confirm whether these associations are seen in other populations because any potential immune effects of pesticides could have widespread consequences given the global burden of infectious diseases on individual and public health.