Occupational exposure to pesticides increases the risk of amyotrophic lateral sclerosis: a systematic review and meta-analysis

Abstract

Objective

To systematically review the evidence on the association between occupational exposure to pesticides and the risk of amyotrophic lateral sclerosis (ALS).

Methods

A systematic search, conducted in eight bibliographic databases for publications between 1990 and 2025, identified observational studies estimating the risk of ALS after occupational pesticide exposure. Study quality was assessed using the WHO Risk of Bias (RoB) assessment instrument for systematic reviews, with the ROBINS-E (RoB in non-randomised studies of exposure) tool domains of bias. Pooled risk estimates were produced using random-effects models with restricted maximum likelihood, heterogeneity was assessed with I² statistics, and meta-regressions and publication bias explored with funnel plots and Egger’s test.

Results

Eight case-control studies (1734 cases) were retained for meta-analysis from 767 initially screened articles. ‘Ever’ occupational exposure to pesticides was associated with an increased risk of ALS (n=6 studies, pooled OR (pOR)=1.6; 95% CI 1.1, 2.2; I²=57%), for combined sexes. The risk for exposure to herbicides was slightly greater (pOR=1.7, I²=0.0%) than for exposure to insecticides or fungicides (pORs=1.6, I²=0.0%). Based on three studies, ever exposure to high levels of pesticides was associated with a higher risk (pOR=2.7; 95% CI=1.4, 5.0) than exposure to low levels (pOR=1.9; 95% CI=1.0, 3.7). Self-reported exposure assessment methods and older publication dates (<2015) were statistically significant predictors of the effect size.

Conclusion

Despite the small number of studies and some heterogeneity, our results add to the evidence suggesting that occupational exposure to pesticides may increase the risk of ALS.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Amyotrophic lateral sclerosis is a rare and rapidly fatal motor neuron neurodegenerative disease whose incidence has increased in recent decades.

• Acute neurotoxic effects of pesticides, particularly insecticides, have been known for decades, but the evidence of their potential long-term neurotoxicity is scarce.

WHAT THIS STUDY ADDS

• This meta-analysis reports increases of 61% to 71% in the risk of developing ALS in persons ever occupationally exposed to pesticides and specifically to herbicides or to insecticides, compared with unexposed persons. This increase was larger for workers ever exposed to high levels than for those ever exposed to low levels.

• In studies that reported risk estimates stratified by sex, the meta-risk in exposed men was twice as high as in unexposed men (three studies), whereas no increased meta-risk was reported in exposed women compared with unexposed women (two studies).

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study can guide improvements in aetiologic research on ALS by identifying certain sources of heterogeneity that can be addressed to strengthen the evidence on causal factors of the disease.

• Combined with other sources of evidence, these findings should encourage the implementation of interventions aimed at reducing exposure during occupational pesticide use.

Introduction

With increasing pressure on food production and uncertainties related to climate change, pesticide use has steadily increased over the past thirty years.¹ Insecticides have long been known for their acute neurotoxic effects, while certain herbicides and fungicides, presumed not to target humans, are now increasingly being investigated.² However, their role in the aetiology of neurodegenerative diseases following chronic exposure is still debated, with the possible exception of Parkinson’s disease.³

Amyotrophic lateral sclerosis (ALS) is a progressive, fatal neurodegenerative disorder that affects central and peripheral motor neurons.⁴ Incidence of ALS increases with age, with higher rates in males.⁵ Its aetiology is not well established, but a few genetic factors have been associated with an increased risk of sporadic ALS,⁶ as well as non-genetic factors such as repeated head injuries and smoking.⁷ Evidence is mounting for the importance of exogenous risk factors, such as exposure to cyanotoxins, electromagnetic fields and heavy metals, highlighting the contribution of environmental and occupational exposures.^{6 7}

Exposure to pesticides was first proposed as a risk factor for ALS in 1980,⁸ and since then, numerous studies and various systematic reviews have explored their potential role in the onset of ALS. The most recent review by Duan et al⁶ found some evidence of a positive association between ALS and exposure to pesticides, but with considerable between-study heterogeneity based on 10 studies of genetic and non-genetic risk factors. One review of systematic reviews by Belbasis et al⁹ examined pesticide exposure among several environmental risk factors for ALS and concluded that there was weak evidence of association with pesticides. Most reviews, however, do not provide separate pooled risk estimates for domestic vs occupational exposure to pesticides^10–12 or account for differences in exposure assessment methods with different reliabilities (eg, job title, expert assessment of occupational history).^{13 14}

It is important to distinguish between occupational and non-occupational exposures. Occupational exposures often involve the absorption of high doses of active ingredients and their adjuvants over short periods of time, mainly through inhalation and dermal routes. These factors in turn influence the resulting toxicokinetics and toxicodynamics.¹⁵

We therefore undertook a systematic review and meta-analysis of epidemiological studies to assess the strength of evidence on the association between occupational exposure to pesticides and the risk of ALS, paying particular attention to exposure assessment.

Methods

We adopted guidelines and tools from the Joanna Briggs Institute,¹⁶ the WHO and the International Labour Organisation (ILO) risk of bias in studies of estimates of exposure to occupational risk factors (RoB-SPEO) tools,¹⁷ and reported results following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines.¹⁸ Two authors independently performed the screening, selection and extraction for each article, without automation tools (FL and SG), using EndNote 20.6 (Clarivate, Philadelphia, Pennsylvania, USA) and an Excel spreadsheet. For each retained article, the adjusted risk estimates and CIs reported were independently verified by another author (PP). The protocol was registered on the PROSPERO international prospective register of systematic reviews (protocol CRD42021285041).¹⁹

Search strategy

The following databases were interrogated for articles published between January 1990 and February 2025: BIOSIS Toxicology, CAB Abstracts, Current Contents Search, Embase, Information SST (ISST), ProQuest Dissertations and Theses Global, PubMed and Scopus. The three search concepts, in French and in English, were disease (amyotrophic lateral sclerosis, Charcot disease, Gehrig and the MeSH term ‘motor neuron disease’); pesticides (MeSH term and subtypes based on target pest, ie, ‘herbicides’, ‘insecticides’, ‘fungicides’, etc.); and workplace (agriculture, occupational exposure, etc., including MeSH terms). The full search strategies used for all databases are given in online supplemental table S1.

Eligibility criteria

Eligible articles had to present specific risk estimates for ALS or for motor neuron diseases (MND) in relation to occupational exposure to pesticides and not only job title or industry (see online supplemental table S2 for a list of inclusion criteria and online supplemental methodological information). Eligible study designs included observational studies (case-control, cohort, case-cohort, cross-sectional or ecological studies) and systematic reviews (umbrella reviews and meta-analysis of observational studies). Peer-reviewed theses and scientific reports were examined, and a search for articles published by the authors was carried out to identify relevant articles for screening. Reference lists of retrieved articles were also scanned to identify any additional pertinent articles.

Non-systematic reviews, commentaries, editorials, conference abstracts, animal or in vitro studies, book chapters or other types of articles not reporting new research results were not retained for screening. Exclusions applied on review of the full papers were studies that did not provide risk estimates with CIs, or that did not present sufficient data to calculate risk estimates pertinent to the research question; reports that were superseded by more recently published results from the same study sample; studies where exposure assessment relied only on having an occupation in agriculture or farming; study results concerning only non-occupational sources of pesticide exposure or in which the occupational exposure was not assessed or analysed independently of other sources of exposure. When the article did not specify that risk estimates concerned occupational exposures, the first or the corresponding author was contacted by email to obtain further information; the article was excluded if no response was received after a follow-up email (three out of six authors responded).

Appraisal of outcome and exposure assessments

Disease assessment was considered of high quality when the ALS diagnosis was confirmed by a neurologist according to El Escorial²⁰ or Belsh criteria²¹; of moderate quality when there was no mention of El Escorial criteria but confirmation of ALS or MND by a neurologist or when extracted from medical records or health statistics with an International Classification of Diseases (9th or 10th revisions, ICD-9 or ICD-10) code (ALS: ICD-9 code 335.20, ICD-10 code G12.21; MND: ICD-9 code 335.2, ICD-10 code 12.2).²² A lower quality was attributed when the diagnosis was self-reported or relied only on death certificates.²³

Exposure assessment was categorised in three levels, adapted from Sutedja et al²⁴: high quality when based on expert assessment or on a job-exposure matrix (JEM) applied to each participant’s complete work history; medium quality when inferred by research team members (not experts in occupational hygiene) based on work history or task description; and low quality when based on self-reported questionnaire data.

Data extraction and quality assessment

The extracted information included the article reference (eg, author, title, year), methodological details (eg, population, recruitment, inclusion and exclusion criteria, exposure and outcome assessment and response rate), statistical details (eg, covariables and sensitivity analyses), and general results (eg, number of cases and controls, adjusted risk estimates).

Evaluation of the quality of systematic reviews was carried out using the Critical Appraisal Checklist for Systematic Reviews and Research Syntheses as recommended by Aromataris et al.¹⁶ Overlap in primary studies included in the systematic reviews was assessed using the overall corrected covered area (CCA) illustrated with the Graphical Representation of Overlap for OVErviews (GROOVE) tool.²⁵ For primary studies, the risk of bias was assessed using the Risk Of Bias In Non-randomised Studies of Exposures (ROBINS-E) tool,¹⁷ with sub-domains based on the WHO risk of bias assessment instrument for systematic reviews informing WHO global air quality guidelines²⁶ (see online supplemental table S3). Discrepancies in quality assessment were discussed between the two extracting authors until consensus was reached. Studies retained for analysis and reporting had to have received the overall score of ‘fair’, ‘average’ or ‘good’.

The variables extracted as moderators included study type (case-control or cohort); type of risk estimate (odds, hazard or risk ratio); case type (incidence, prevalence or mortality); method of exposure assessment (self-reported exposure, derived from occupational history or expert assessment/JEM); diagnostic criteria (El Escorial criteria and/or medical records reviewed by a neurologist, or death certificates); pesticide exposure class (pesticides as a non-specific class or categorised by target pest: herbicides, fungicides, insecticides, etc.); exposure intensity (ever exposed, low or high, or other units of analysis); and sex (no study collected information on gender).

Statistical analysis

The criteria used and decisions made in selecting risk estimates to include in the meta-analysis are detailed in the online supplemental methodological information and in online supplemental table S4. Briefly, adjusted risk estimates were favoured, risk estimates derived with less than five exposed cases/controls were excluded, and meta-analysis was performed when a minimum of two studies reported on the same pesticide type and/or covariable.

Meta-analyses and tests were performed using a random-effect linear regression model with restricted maximum likelihood (REML) to account for anticipated heterogeneity among studies due to differences in study design, populations, and exposure assessment methods.²⁷ This model performs well in meta-analyses with few studies or imbalanced sample sizes and introduces low bias in heterogeneity estimates.²⁸ After transforming the risk estimates and 95% CI from each study, we calculated the pooled risk estimates and their 95% CIs using the REML model and then exponentiated the pooled estimates and their 95% CIs to present the results in the original ratio scale.

Subgroup meta-analyses were conducted on the following covariates, when at least two studies reported results for subgroups of sex (all/male/female), class of pesticide (fungicides/herbicides/insecticides), exposure intensity (high/low), exposure assessment method (self-reported/expert-assessment), type of cases (incident/incident and prevalent/prevalent) and period of publication (before/after the median year, 2016). Apart from the subgroup analysis stratified on sex, all subgroup analyses used estimates on combined sexes when possible. Likewise, except for the class of pesticide, all other subgroup meta-analyses used estimates for pesticides as a non-specific class. In order to compare studies despite disparities in reported exposure intensity units and categories, newly created low and high intensities were grouped as follows: below/above the median level; less than once a month for <10 years/more than once a month for ≥10 years; and <400 exposure-days/> 2000 exposure-days (see online supplemental methodological information for details). The I-squared (I²) statistic was calculated to estimate the percentage of variation across studies, considering heterogeneity to be low (I²<25%), moderate (I²≥25%–49%), high (I²≥50%–74%) or extreme (I²≥75%).²⁹ The risk of publication bias and small-study effects were assessed with funnel plots, Egger’s tests and Galbraith plots.^{30 31}

Meta-regressions were conducted to quantify the relationship between the risk estimates and the covariates. The residual I² (I²_res) was calculated to measure the proportion of variability in effect size that remains unexplained after accounting for the covariate. The adjusted R² statistic reports the proportion of the between-study variance explained by the covariate. Finally, the Wald’s Χ² assesses the statistical significance of the covariate in the model, assuming a p-value threshold of 0.10.²⁹

Additional analyses were conducted to assess the robustness of findings. First, as the number of available studies was small, meta-analyses were repeated using a fixed-effect inverse-variance model to investigate how heterogeneity influenced the pooled estimate.³² The influence of individual studies was also explored with the “leave-one-out” method to assess overall heterogeneity and risk estimate changes. Finally, meta-analyses were redone including three articles that studied all motor neuron diseases (MND) rather than only ALS.^33–35 In categories where only two risk estimates were available, pooled risk estimates were deemed unreliable and are not presented.

All analyses were performed using STATA 18.5 (StataCorp LLC, Texas, USA).

Results

Study selection

Our literature search produced 1053 records, of which 160 primary studies and systematic reviews were identified for full-text screening. Figure 1 illustrates the PRISMA flow diagram of the identified, excluded and retained articles. Seventeen reviews and 38 primary studies were analysed, covering the periods 2007–2023 and 1997–2022, respectively. Of these, 12 reviews and 27 articles on primary studies were excluded for reasons presented in figure 1 and detailed in online supplemental table S5.

Figure 1. PRISMA Flow chart of the selection process of articles. ALS, amyotrophic lateral sclerosis; MND, motor neuron diseases; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analysis; RoB, risk of bias.

Five retained reviews met at least four quality criteria: a search in more than one bibliographic database, the selection of studies by two independent assessors, followed by their data extraction by the two same assessors, and a risk of bias assessment of primary studies.^{6 13 24 36 37} These reviews are summarised in online supplemental table S6; they retained a total of 26 primary studies, and the overlapping primary studies are illustrated in online supplemental table S7. There was a moderate to very high overlap of primary studies according to the CCA corrected for structural missingness (overall CCA=16.4%), and only one review focused on agricultural pesticide exposure,¹³ whereas the others had wider objectives. Given the disparate importance attached to exposure assessment in systematic reviews and their significant overlap, we opted to conduct a systematic review with meta-analysis using only primary studies instead of performing an umbrella review.

Eleven primary studies, conducted in the USA (n=5), in Europe and in Australia (n=2 each), in Asia, and in both Canada and France, were deemed of fair to good overall quality based on risk of bias assessment (details in online supplemental table S8) and were retained for meta-analysis. Of those, eight were included in the main meta-analyses, and three were used only in additional analyses (studies on MND, one of which focused on Agent Orange).

Characteristics of studies

Table 1 briefly describes the 11 retained primary articles (see more details in online supplemental table S9). They reported risk estimates for the association between ALS and occupational exposure to pesticides, generally assessed as a non-specific class.

Table 1. Selected characteristics of retained case-control and cohort-based primary studies on the association between occupational exposure to pesticides and amyotrophic lateral sclerosis (ALS).

Author, Country; Design	Study population	Reference population	Exposure assessment method	Exposure index	Adjustment variables
Beard et al (2016),44 USA; case-control	621 ALS cases (98% males), war veterans from the GENEVA study. Records reviewed by neurologist	958 controls (93% males), matched on diagnostic age and use of Veterans Affairs healthcare system	Self-reported exposure: herbicide, Agent Orange; telephone interview	Ever/never exposed Number of days exposed at work? In field?	Age, use of the VA healthcare system, sex, race/ethnicity, military branch of longest service
Beaudin et al (2022),38 Canada, France; case-control	404 ALS cases (57% males), treated in three tertiary-care referral centres (2003–2012)	381 controls (38% males), spouses of neurology patients+non MND neurology patients	Self-reported exposure: insecticides, herbicides, fungicides, rodenticides; telephone interview	Ever/never exposed (non-specific pesticides) Times per month and years exposed	Age, sex, region (Canada or France), education
Chen et al (2022),33 New Zealand; case-control	321 MND cases (64% males), recruited 2013–2016 from MND Association & hospital discharge database	605 controls (55% males), frequency-matched (age, sex) from 2008 New Zealand electoral list	Self-reported exposure: pesticides, insecticides, herbicides, fungicides, fumigants; interview or questionnaire	Ever/never exposure in one job Duration of exposed job in years	Age, sex, education, ethnicity, SES, smoking, sports, alcohol, head/spine injury, other self-reported exposures
Fang et al (2009),39 USA; case-control	109 ALS cases (61% males), recruited 1993–1996 from two referral centres	253 controls (62% males), matched on age, sex, New England region; from random digit dialling	Self-reported exposure: insecticides, herbicides, fungicides, fumigants; face-to-face/telephone interview	Ever/never exposed≥10 times Duration of exposed job in days	Age, sex, region of residence; some models: smoking, education
Filippini et al (2020),45 Italy; case-control	95 ALS cases (54% males), from registries, death certificates, hospital records (2008–2011 and 2002–2012)	135 controls (53% males), frequency-matched (sex, age, province) from health services database, recruited by mail	Self-reported exposure: pesticides, insecticides, herbicides, fungicides; questionnaire	Ever/never exposed	Sex, age, education.
Goutman et al (2022),40 USA; case-control	381 ALS cases (55% males), recruited from the University of Michigan ALS Clinic, 2010–2020	272 controls (47% males), from university online recruitment database	Self-reported exposure: pesticides or arsenic-containing compounds; questionnaires	Exposure score attributed by industrial hygienists for each of four reported jobs	Age, sex, military service
Koeman et al (2017),34 Netherlands; case-cohort	136 MND deaths (56% males), Netherlands Cohort Study on diet and cancer (NCSDC), 1986–2003	4166 controls (48% males), from random subset of participants in NCSDC	Job history reported in 1986; exposure from JEM: pesticides, insecticides, herbicides, fungicides	Ever/never exposed	Smoking, education, BMI, physical activity at recruitment
Malek et al (2014),41 USA; case-control	66 ALS cases (68% males), from three neurology clinics, 2008–2010	66 controls (68% males), from clinics waiting rooms & online database, matched by region, age, sex & race	Self-reported exposure: pesticides, insecticides, herbicides, fungicides, fumigants; interview	Ever/never exposed Duration of exposed job in days	Smoking, education, metals, organic/aromatic solvents, electrical, electronic, electro-magnetic machinery
McGuire et al (1997),42 USA; case-control	174 ALS cases (55% males), from surveillance system, 1990–1994	348 controls (55% males), recruited by random digit dialling or medicare lists, matched for county, sex & age group	Job history, exposures (self-reported & by industrial hygienists): pesticides, insecticides, herbicides, fungicides	Ever/never exposed (self and panel assessment) Low/high exposure (panel assessment)	Age, sex, education, proxy response or not
Pamphlett 2012,43 Australia; case-control	614 ALS cases (62% males), neurologist diagnosis; recruited by newsletter posts 2000–2011	708 controls (53% males), partners, friends or community groups	Self-reported exposure: pesticides, herbicides; questionnaire	Ever/never regularly exposed	No adjustment (“… reasonable similarities in age and gender…” between cases and controls.
Yi et al (2014),35 Korea; cohort	102 MND cases (100% males), from Korean Veterans Health Study; diagnosis from Veterans Health Service claims	Cohort of 111 726 veterans from Vietnam War, alive in July 2004 and with information on deployments	Deployment history, question on military unit: Agent Orange exposure from GIS model using spraying area	Exposure Opportunity Index model (low/high)	Age, military rank, smoking, alcohol, physical activity, domestic herbicide use, education household income

ALS, amyotrophic lateral sclerosis; BMI, body mass index; GENEVA, Genes and Environmental Exposures in Veterans with ALS; GIS, geographic information system; MND, motor neuron disease; SES, socioeconomic status; U. of Michigan, University of Michigan; VA, Veterans Affairs.

Participants were recruited or identified from varied sources such as neurology clinics^38–42 and existing databases created by patients’ associations,^{33 43} for specific studies⁴⁴ or for health services purposes.^{35 45}

Eight articles that focused on ALS were case-control studies, with participation rates between 55%–97% for cases and 35%–96% for controls, except for one study with an overall response rate of 19%.⁴⁵ Three studies (two cohort-based) included the broader diagnosis of MND (ICD-10 G12.2) for their case definition.^33–35 Five studies presented risk estimates combined for men and women,^{38–41 44} three others presented only risk estimates stratified by sex,^{34 35 43} and three studies presented both separate estimates by sex and combined estimates.^{33 42 45} Three studies did not present exposure to pesticides as a non-specific class but as separate classes of pesticides,^{35 39 44} and seven articles made an attempt to classify exposure as ‘high’ and ‘low’.^{33 35 38 39 41 42 44}

Overall results

Figure 2 presents forest plots of risk estimates and pooled odds ratios (pORs) for associations between ‘ever exposure’ to pesticides (non-specific class) and ALS risk, as a combined sex category and separately for women and men. Whereas sex-specific pooled estimates presented no heterogeneity, pORs for studies combining male and female participants presented extreme heterogeneity (I²=87%). There was a two-fold pOR for exposed men and a 58% increased risk overall. In figure 3, forest plots present risk estimates and pORs stratified by ever exposure to specific classes of pesticides: the pOR for herbicides was larger (71% increased risks) than for insecticides and fungicides (57% to 61% risk increase), with no heterogeneity. The overall pOR for these three classes of pesticides combined (pOR=1.61) is similar to the pOR for non-specific pesticides (pOR=1.58).

Figure 2. Association between ever-occupational exposure to pesticides (non-specific class) and a diagnosis of amyotrophic lateral sclerosis. Studies with estimates for combined sexes and stratified by sex. REML, restricted maximum likelihood.

Figure 3. Association between ever-occupational exposure to fungicides, herbicides and insecticides and diagnosis of amyotrophic lateral sclerosis. Studies with estimates for combined sexes only. REML, restricted maximum likelihood.

Table 2 presents the pORs for subgroup meta-analyses and the meta-regression effect size (exponentiated β) for exposure to pesticides and risk of ALS, stratified by sex, pesticide class, non-specific pesticide exposure intensity, exposure assessment method, type of ALS cases and period of publication; forest plots for these subgroup are presented in online supplemental figures S1–S4. There were insufficient studies to allow analyses by diagnostic method or by intensity of exposure to classes of pesticides.

Table 2. Subgroup meta-analyses and regression risk estimates for exposure to pesticides and ALS, with heterogeneity and variance statistics.

Grouping/adjustment covariable	Number of ORs	Subgroup meta-analysis		Meta-regression, variance and heterogeneity
		pOR (95% CI)	I² (%)	Effect size (95% CI)	Adj. R² (%)	I²res (%)	Wald χ² (P value)
Sex*	8	1.58 (1.13, 2.19)	57.3		7.40	49.1	2.54 (0.281)
Combined sexes	3	1.87 (0.746, 4.67)	87.1	(reference)
Female	2	1.05 (0.596, 1.85)	0.0	0.676 (0.280, 1.63)
Male	3	2.09 (1.13, 3.04)	0.0	1.40 (0.678, 2.90)
Class of pesticide	11	1.62 (1.26, 2.22)	0.0		0.0	0.0	0.08 (0.961)
Fungicides	3	1.61 (0.952, 2.73)	0.0	(reference)
Herbicides	4	1.71 (1.06, 2.74)	0.0	1.06 (0.522, 2.15)
Insecticides	4	1.57 (1.11, 2.22)	0.0	0.974 (0.519, 1.83)
Pesticide exposure intensity*	6	2.30 (1.46, 3.62)	0.0		0.0	0.0	0.481 (0.481)
High	3	2.68 (1.44, 5.01)	0.0	(reference)
Low	3	1.94 (1.00, 3.74)	0.0	0.722 (0.291, 1.79)
Exposure assessment method*	8	1.43 (1.09, 1.89)	43.5		100	0.0	6.57 (0.010)
Self-reported	6	1.62 (1.24, 2.13)	0.0	(reference)
Expert-assessment	2	1.05 (0.863, 1.28)	0.0	0.647 (0.463, 0.903)
Type of case*	8	1.57 (1.18, 2.09)	52.8		8.60	47.1	2.83 (0.243)
Incident	3	2.43 (1.26, 4.70)	49.6	(reference)
Incident & prevalent	2	1.51 (0.977, 2.32)	0.0	0.644 (0.292, 1.42)
Prevalent	3	1.31 (0.860, 2.01)	64.0	0.563 (0.287, 1.10)
Publication period*	7	1.60 (1.15, 2.23)	59.1		67.2	28.1	4.38 (0.036)
1997-2014	4	2.08 (1.36, 3.19)	30.0	(reference)
2016-2022	3	1.19 (0.873, 1.63)	35.2	0.581 (0.350, 0.966)

Figures in bold indicate statistical significance.

^*Includes only risk estimates for exposure to pesticides as a non-specific class.

Adj., adjusted; pOR, pooled odds ratio; REML, random effects model with restricted maximum likelihood.

Despite a significantly elevated pOR for males, the meta-regression using studies combining results for both sexes (reference) explained little between-study variance or residual heterogeneity (R²=7.4% and I²=49%), likely because the reference group mostly included men. However, when studies with combined results (‘Combined sexes’ category) were removed, the R² was 100% (female as a reference, risk estimate for male=1.99, 95% CI=1.01, 3.93, I²_res=0%, Wald χ²=4.0, p=0.047; results not tabulated).

The meta-regression by class of pesticides showed small differences in risks for herbicides and insecticides compared with fungicides, with no additional variance explained (R²=0.0%). High-intensity exposure to non-specific pesticides and to insecticides was associated with greater pORs than low-intensity exposures (n=3 and 2 studies, respectively), but again meta-regression analyses showed no between-study variance in risk estimates with these two covariables.

Self-reported assessment of exposure produced a greater risk (pOR=1.6) than expert-based assessment (pOR=1.4), but this difference was not significant in the meta-regression partly owing to heterogeneity between studies with expert-based assessments. Studies based on incident cases reported greater pooled risk estimates (pOR=2.4) than the ones including incident and prevalent cases (pOR=1.5) or prevalent cases only (pOR=1.3). However, the meta-regression showed no differences in risk between studies with 8.6% of the between-study variance explained by this variable. Finally, older studies reported a greater meta-risk estimate (pOR=2.1) than those published in the last 10 years (pOR=1.2). The meta-regression showed that period of publication explained two-thirds of the between-study variance.

Potential bias and sensitivity analyses

Funnel plots were inspected for all analyses and are reported for the main pooled effects and by class of pesticides (online supplemental figure S5). All funnel plots appeared symmetrical and revealed no outliers other than those clearly evident in the forest plots, namely those of Malek et al⁴¹ and Goutman et al⁴⁰ (figure 2) and of Fang et al³⁹ (figure 3). Galbraith plots (not presented) did not reveal additional information on heterogeneity or outliers. Egger’s tests for small study effect are reported in online supplemental table S10, where analyses adjusted for sex, for exposure assessment method, for type of case, and for period of publication suggest some asymmetry (p value <0.05), possibly linked to the Malek et al study.⁴¹ The “leave-one-out” analysis confirmed the influence of the same studies on risk estimate changes (online supplemental figure S6).

Repeating the analysis exploring pesticide classes and exposure intensity using fixed-effect inverse-variance models (online supplemental table S11) produced pORs comparable to those obtained with REML. Finally, the random-effects models reported in table 2 were repeated after adding risk estimates from the three cohort-based studies that considered the more general diagnosis of MND.^33–35 Online supplemental table S12 shows that all pooled risk estimates (pREs) remained the same or decreased slightly, and that heterogeneity increased for all adjustment covariables; the pRE for mortality studies was equal to 1.0.

Discussion

This systematic review of observational studies examined articles published between 1990 and 2025 on the association between occupational exposure to pesticides and the risk of ALS. In most studies, ALS diagnosis was made by a neurologist according to El Escorial criteria, and exposure was generally self-reported from a list of agents that included pesticides as a non-specific class. However, selection bias and misclassification of exposure were more concerning, with only one out of eight ALS studies considered to present low or low-medium risk of bias for both domains (online supplemental table S8). The overall direction of bias was impossible to assess given the heterogeneity of primary studies (see additional discussion in the Methodological considerations below, and in the online supplemental methodological information).

Based on eight case-control studies, overall analyses for ‘ever exposure’ to non-specific pesticides and to classes of pesticides gave increased pooled risks of 58% and 61%, respectively (71% for herbicides and 57% for insecticides).^{38 42 45} A semi-quantitative assessment of non-specific pesticide exposure intensity showed risk increases of 168% and of 94% at high and low intensities, respectively. The consideration of results from studies including MND did not notably influence the pooled risk estimates (online supplemental table S12).

Our pooled risk estimate of 1.58 for “ever exposure” to non-specific pesticides is slightly greater than those reported by three systematic reviews (online supplemental table S6). Focusing on agricultural exposures, Kang et al¹³ obtained a meta-risk estimate of 1.4 (95% CI 1.2, 1.7; I²=41%; 15 studies, including three overlapping with our review). Targeting several occupational exposures, Gunnarsson et al³⁷ calculated a pooled estimate of 1.4 (1.0, 1.8; I²=58%; five studies, one overlapping with ours). Summarising the effects of genetic and non-genetic risk factors, Duan et al⁶ reported a pooled risk estimate of 1.5 (1.1, 1.9; I²=79%; 10 studies, three overlapping with ours). Lastly, the Belbasis et al⁹ umbrella review had concluded that the evidence for an association between environmental exposure to pesticides and the onset of ALS was weak, based only on the above-mentioned Kang et al review.

Several chemical classes of insecticides and herbicides, including organochlorine-based, organophosphorus-based and glyphosate-based compounds, act on target pests through mechanisms that are neurotoxic to humans; for example, malathion, paraquat and glyphosate produce alterations, including oxidative stress, neuroinflammation and reactive oxygen species,^{2 46 47} which are among proposed mechanisms for pesticide exposure-mediated ALS,⁴⁸ in addition to neuronal apoptosis and inhibition of critical neurotransmission enzymes, such as cholinesterases.⁴⁹ Few primary studies assessed the risk associated with exposure to pesticides classified by target pest. As reported for Parkinson’s disease,⁵⁰ exposure to insecticides and to herbicides was associated with greater pooled risks of ALS (increased risks of ~60%). Some studies have evaluated the association between ALS and occupational exposure to specific chemical classes or to individual pesticides. Chen et al³³ calculated larger risk estimates for the association between MND and self-reported pesticide exposure to organochlorines (OR 3.3; 95% CI 1.2, 9.1), organophosphates (OR 3.1; 95% CI 1.4, 6.9) and pyrethroids (OR 6.4; 95% CI 1.1, 36.0), based on fewer than 23 cases for each group. It is possible that differences in data collection procedures (self-administered questionnaires, telephone/face-to-face interviews) and type of cases (incident/prevalent) may have influenced the size of risk estimates.

A detailed dose-response analysis, other than stratification in approximate low and high levels, was not possible given the disparity of results presentation in primary studies^{38 41 42}; only one study reported a significant dose-response trend with ALS.⁴² Our findings are compatible with a dose-response trend (greater intensity of exposure being linked to larger risks than lower intensity). Agent Orange exposure during the Vietnam war has also been associated with increased risks of ALS in American and Korean veterans^{35 44} with ORs ranging from 1.2 for low-intensity to 3.3 for high-intensity exposures. None of the five retained reviews presented pooled estimates based on intensity of exposure.

In this review, pooled risk estimates based on self-reported exposures were larger than those based on expert assessment, which could be explained by recall bias or a preexisting belief that pesticides can increase the risk of ALS, especially in studies published after the first meta-analysis of Kamel et al⁵¹ that showed an association between ALS and exposure to pesticides. Conversely, McGuire et al⁴² published before the Kamel et al article, found slightly greater risk estimates for expert assessment compared with self-reported exposure, and even greater estimates when results of both methods were combined, which reflects different assessments of exposure between experts and lay persons. In a recent review on pesticide exposure and Parkinson’s disease, Ohlander et al also found greater pooled risk ratios for expert assessments compared with job title/industry or self-reports, although there was no statistical difference between the three methods.⁵²

We found an elevated risk estimate in studies with incident cases only (with moderate heterogeneity (I²=50%) and lower risk estimates in studies with a combination of incident and prevalent cases or with prevalent cases only (no heterogeneity). If exposure influences both disease onset and its survival, it is likely that prevalent cases have characteristics that favoured their survival, and if this (unmeasured) characteristic is also associated with disease onset, the measured association may be biased.⁵³

The median publication year, 2016, was used as a cut-off point to separate more and less recent publication periods for the primary studies. The resulting analysis produced lower pooled risk estimates in recent years. In their review of occupational pesticide exposures and three chronic diseases, Ohlander et al⁵² did not report significant differences in summary estimates for Parkinson’s disease, prostate cancer or non-Hodgkin’s lymphoma from recent publications (defined as ≥2007), although the more recent estimates were slightly lower. A more recent period of publication may be an indicator of more refined study methods, but this did not seem to be the case in our review, as in each period, only one study used expert occupational exposure assessment, and the older studies were judged to have, overall, a lower risk of bias and a better quality (online supplemental table S8).⁴² However, older studies had a lower mean sample size but generally reported higher participation rates (mean 85%) compared with more recent ones (52%).

The sensitivity analyses corroborated our findings of increased risks with exposure to pesticides, and they highlight variability in pooled risk estimates linked to heterogeneity. The two studies that changed pooled risk estimates the most when omitted were the same for all analyses,^{40 41} and both recruited patients from neurology clinics and used self-reported exposure to pesticides in their analyses that produced divergent risk estimates. One study derived exposure from a self-administered question asking about exposure ‘to pesticides or arsenic’⁴⁰; this non-specific question, together with a small number of exposed subjects (five cases and seven controls), could have reduced the risk estimate towards the null. The other study assessed exposure “10 times or more” through a question on occupational exposure to 21 agents, including pesticides.⁴¹ Their analysis was adjusted for several potential confounders, but not for residential well water use (although cases were twice as likely as controls to have used that water source for >20 years). Pesticide levels in residential well water of that area had been measured for a few decades⁵⁴; if pesticide exposure is causally associated with ALS, failing to adjust for residential well water use could have overestimated the risk associated with occupational exposure to pesticides.

The analyses including studies on MND only slightly reduced the pooled risk estimates, consistent with ALS constituting approximately 85% of MNDs,³³ but introduced more heterogeneity.

Additional analyses to test other covariables that could explain some heterogeneity were not possible because of a lack of information in the publications. For example, only Beaudin et al³⁸ tested the effect of modifying the reference group, reporting that excluding cases’ spouses and siblings from the control group increased the risk estimates (from 1.67 to 2.04), confirming similarities of exposure to pesticides of spouses. They also found that participants who did not use personal protective equipment had a risk 15% to 40% greater, depending on the control group, than the other participants.

Methodological considerations

Based on our risk of bias assessment, the overall quality was rated as satisfactory or average for five of the eight primary ALS studies and two of the three MND studies (see online supplemental table S8 and supplemental methodological information). Overall, outcome assessment was unlikely to introduce major bias, and most studies adjusted for a similar minimal set of potential confounders. However, selection bias and misclassification of exposure were more concerning, with medium or medium-high risk of bias in seven of the eight ALS studies and one of the three MND studies; these biases were just as likely to increase or decrease the risk estimates. This meta-analysis has several strengths. We searched eight bibliographic databases for reviews and primary studies and retained those of fair to satisfactory quality, based on systematic assessment methods. We insisted on consensus diagnostic criteria for ALS and on individual exposure assessments rather than group-level ones, such as job title or industry, that are less accurate and may be an unreliable indicator of actual pesticide use (see additional discussion in online supplemental methodological information).⁵⁵ The case-control studies all relied on neurologist-confirmed diagnoses based on published criteria, and the cohort-based studies used a MND diagnosis, deemed to be reasonably accurate for research purposes.⁵⁶ Our efforts to distinguish between occupational and non-occupational sources of exposure were aimed at facilitating the selection of targeted preventive measures to reduce exposure. The overall pooled risk estimates for pesticides as a non-specific class were based on 1734 cases of ALS and 457 cases of MND, depending on the model.

The major limitations of this review are the small number of available primary studies (with limited statistical power and imprecision of between-study variance)⁵⁷ and their crude exposure assessment generally based on answers to few limited questions, which reduced our main pooled risk estimates to the crude indicator of ever/never exposure. In addition, most studies did not specify whether diagnoses were of sporadic ALS or included familial cases, but familial cases constitute 4% to 10% of ALS cases, depending on geography, study design, definition of familial cases and decade of case ascertainment⁵⁸ and should not invalidate our conclusions. The average age at onset is 10 to 20 years older for sporadic ALS, but all available risk estimates were adjusted for age.⁷

A large part of the limitations is attributable to the retrospective nature of observational studies and the associated uncontrolled biases. Some lifestyle and environmental factors are suspected to increase the risk of ALS, and the highest level of evidence has been attributed to head injury.⁵⁹ A few primary studies have reported increased risks associated with potential risk factors (eg, previous head trauma and electrocution³⁸ or consumption of well water⁴¹) but did not adjust their risk estimates for these suspected risk factors, which may have led to uncontrolled confounding. Some risk overestimation can be attributed to information bias derived from better recall or false memories of exposure among cases (who sought to explain their illness or may have been aware of a suspected link between pesticides and their illness).⁵³ On the other hand, most available primary studies relied solely on self-reported pesticide exposure, leading to misclassification that tends to bring risk estimates closer to the null.^{52 60}

Conclusion

In our review, the limited detail and heterogeneity in exposure metrics prevented us from conducting more refined analyses, including assessments of interactions, intensity of exposure, or more detailed dose-response patterns. Hence, to strengthen the evidence on risk factors for ALS, primary studies particularly need methodological improvements on exposure assessment, on the selection of reference groups and on adjustment for potential confounders, including analysis for interactions between risk factors. The findings of this review add to the evidence that occupational exposure to pesticides may increase the risk of ALS and should encourage the implementation of interventions aimed at reducing exposure during occupational pesticide use.