Systematic literature review of the epidemiology of glyphosate and neurological outcomes

Purpose Human health risk assessments of glyphosate have focused on animal toxicology data for determining neurotoxic potential. Human epidemiological studies have not yet been systematically reviewed for glyphosate neurotoxicity hazard identification. The objective of this systematic literature review was to summarize the available epidemiology of glyphosate exposure and neurological outcomes in humans. Methods As of December 2021, 25 eligible epidemiological studies of glyphosate exposure and neurological endpoints were identified and assessed for five quality dimensions using guidance from the U.S. Environmental Protection Agency. Studies that assessed personal use of glyphosate were prioritized, whereas those assessing indirect exposure (other than personal use) were rated as low quality, since biomonitoring data indicate that indirect metrics of glyphosate exposure almost always equate to non-detectable glyphosate doses. Results Overall, the scientific evidence on glyphosate and neurotoxicity in humans is sparse and methodologically limited, based on nine included epidemiological studies of neurodegenerative outcomes (two high quality), five studies of neurobehavioral outcomes (two high quality), six studies of neurodevelopmental outcomes (none high quality), and five studies of other and mixed neurological outcomes (one high quality). The five high-quality studies showed no association between glyphosate use and risk of depression, Parkinson disease, or peripheral nerve conduction velocity. Results were mixed among the eight moderate-quality studies, which did not demonstrate consistent associations with any neurological endpoints or categories. Low-quality studies were considered uninformative about possible neurotoxic effects due primarily to questionable assessments of indirect exposure. Conclusions No association has been demonstrated between glyphosate and any neurological outcomes in humans. To move the state of science forward, epidemiological studies should focus on scenarios involving direct and frequent use of glyphosate while collecting information on validated health outcomes, concomitant agricultural exposures, and relevant personal characteristics. Supplementary Information The online version contains supplementary material available at 10.1007/s00420-022-01878-0.


Introduction
Glyphosate is a non-selective phosphonomethyl amino acid herbicide that is widely used to control weeds in agriculture, forestry, and lawn and garden care. Following glyphosate's initial registration in 1974, the U.S. Environmental Protection Agency (EPA) has routinely reviewed and reassessed the safety and uses of glyphosate, in part by implementing the registration review process every 15 years (U.S. EPA 2022). In January 2020, U.S. EPA released its latest interim decision for registration review of glyphosate, in which the Agency concluded that there are no risks of concern to human health when glyphosate is used in accordance with its current label. Likewise, the European Food Safety Authority (EFSA) and European Chemicals Agency recently began their renewal reassessments for glyphosate (EFSA 2021).
While substantial scientific and public attention has been focused on the carcinogenic potential of glyphosate (U.S. EPA 2017b), neurological outcomes have been less frequently studied. In its most recent draft human health risk assessment in support of registration review for glyphosate, U.S. EPA concluded: "There was no evidence that glyphosate is neurotoxic" (U.S. EPA 2017a). This conclusion was based primarily on a limited number of acceptable toxicity studies in animals.
Nevertheless, neurological outcomes spanning the age spectrum, including neurodevelopmental disorders and chronic neurodegenerative diseases, have become increasingly prominent issues in human health risk assessments of pesticides, including glyphosate. Uncertainties and limitations in the extrapolation of data from laboratory animals to humans make it important to consider whether informative epidemiological data exist and, if so, to incorporate epidemiological findings into the process of human health risk assessment (Déglin et al. 2021). To date, however, a thorough review of the epidemiological evidence on glyphosate and neurotoxicity has been lacking. Therefore, we undertook this systematic literature review to identify, assess, and summarize the current state of the epidemiology on glyphosate exposure and neurological outcomes in humans.

Literature search
This systematic literature review was conducted in accordance with procedures described in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement (Page et al. 2021). Following PRISMA guidance, we developed a Population-Exposure-Comparator-Outcome-Study Design (PECOS) statement to delineate the objectives of our review, as follows: • Population: humans • Exposure: glyphosate exposure • Comparator: absence of glyphosate exposure • Outcome: chronic neurological conditions, including central and peripheral nervous system disorders, excluding acute poisoning and intoxication events, acute nonspecific neurological symptoms (e.g., headache, dizziness), and nervous system neoplasms • Study design: comparative epidemiological studies, including cross-sectional, case-control, and cohort studies, excluding case reports and case series.
Studies considered eligible for inclusion in our literature review were human epidemiological studies that reported a measure of association (or data sufficient to calculate a measure of association) between any chronic neurological condition, as described above, and glyphosate exposure, including direct exposure via personal use (i.e., spraying or mixing) of glyphosate, as well as possible indirect exposure via routes other than personal use. Measures of association included relative risks (e.g., odds ratios, risk ratios, hazard ratios), risk differences, correlation coefficients, etc. We excluded animal, in vitro, and mechanistic studies, because this review is limited to human epidemiology. We also excluded case reports and case series, because they lack comparators, and thus cannot estimate exposure-outcome associations, and reviews, editorials, commentaries, letters, and other articles that did not report original data, although we examined the reference lists of these sources to identify additional potentially relevant studies.
This search string identified 602 results in PubMed as of December 8, 2021. We included 16 additional potentially relevant articles identified from reference lists, bringing the total number of articles considered to 618. After excluding 28 non-human studies, 304 studies of irrelevant exposures (e.g., non-specific categories of pesticides or herbicides, specific pesticides other than glyphosate, occupational titles, or medications), and 63 studies of non-neurological outcomes (e.g., cardiovascular diseases, neoplasms), we considered 223 articles for full-text review. Based on our review of the complete texts of these 223 articles, we excluded 196 studies that did not report original associations with glyphosate exposure and two studies that did not report associations with neurological conditions, leaving 25 articles eligible for inclusion in our systematic review. A flow chart of the literature search process is shown in Fig. 1.

Data collection
Study characteristics and results were extracted into a spreadsheet with separate fields for first author, publication year, study country, study design, study population (including size, setting, and participation rates), comparison population, exposure assessment method, exposure levels assessed, neurological outcomes assessed, outcome assessment method, confounders adjusted and considered, statistical approach, estimates of association (with interval estimates, e.g., 95% confidence intervals or credible intervals), and funding source. Where applicable, results with multivariable statistical adjustment were prioritized for reporting over unadjusted results, and results for multiple quantitative or semi-quantitative categories of glyphosate exposure were prioritized for reporting over those comparing ever vs. never exposure. Studies were categorized according to whether they assessed neurodegenerative outcomes [e.g., Parkinson disease, amyotrophic lateral sclerosis (ALS)], neurobehavioral outcomes (e.g., depression, cognitive function), neurodevelopmental outcomes (e.g., autism spectrum disorder, neural tube defects), or other or mixed neurological outcomes. Data extraction was conducted by one investigator (ETC) and reviewed for accuracy and completeness by the two other investigators (NUO, JFA).

Glyphosate exposure assessment
Exposure assessment is critically important in the epidemiology of glyphosate and is, therefore, introduced here in detail as background to inform the exposure assessment quality rating used in our systematic literature review (described in the next section). In general, exposure to pesticides, including glyphosate, is variously defined in epidemiological studies. Here, we refer to "direct exposure metrics" as those that classify exposure based on first-hand or personal application or mixing of a specific pesticide. In contrast, we refer to "indirect exposure metrics" as those that classify exposure to a pesticide based on possible routes other than personal use. For example, some studies classify indirect exposure based on having a household member who mixed or applied a pesticide, or working or living on a farm where a pesticide was applied, without necessarily having applied it oneself. Other studies classify exposure indirectly based on having a residential address within a specified distance from a reported pesticide application. Some studies use an ecological indirect exposure definition, such as total weight of a pesticide applied annually to certain crops within a state or county. The common assumption across all of these exposure metrics is that they provide a valid indication of glyphosate entering the body of persons classified as exposed.
Biomonitoring for glyphosate sheds light on the validity of these various exposure definitions and provides an estimation of the potential range of corresponding doses (namely, the amount of chemical that is biologically internalized). The most comprehensive biomonitoring study for glyphosate is the Farm Family Exposure Study (FFES) . The FFES included 48 farmers who applied glyphosate on their farms in Minnesota and South Carolina, along with their immediate family members residing on those farms. Family members aged 4 years and older collected 24-h urine samples for measurement of glyphosate levels on the day before, the day of, and 3 days after the on-study pesticide application. On the day of glyphosate application, 60% of farmers had quantifiable glyphosate in urine [the predominant route of excretion (U.S. EPA 2017b)], whereas 40% of farmers did not; the prevalence of detectable glyphosate declined on post-application days 1-3 (Acquavella et al. 2004). The limit of detection for glyphosate in urine was one part per billion (ppb). The distribution of urinary glyphosate concentrations for applicators was highly skewed, with only a small percentage of values that were appreciably different from the 3-ppb geometric mean. The maximum systemic dose in any farmer was 0.004 mg/kg, and the median systemic dose was 0.0001 mg/kg; methods for calculation of systemic dose, based on amount of glyphosate excreted and adjusting for incomplete excretion, pharmacokinetic recovery, and individual body weight, are described in an appendix to .
These results from the FFES are consistent with those subsequently published (Curwin et al. 2007;Niemann et al. 2015;Solomon 2016). Because all of the farmers personally applied glyphosate, the 40% of farmers with undetectable urinary glyphosate may reasonably be assumed to have had some biological dose of glyphosate below the 1-ppb limit of detection. Accordingly, in the present literature review, we considered personal use of glyphosate to be a valid exposure metric, corresponding to a median systemic dose that is approximately 3-4 orders of magnitude below the level set by various regulatory agencies as an acceptable daily intake [~ 0.3-1 mg/kg/day (EFSA 2015;FAO and WHO 2016;U.S. EPA 2017a)].
Many epidemiological studies of glyphosate employ indirect metrics of exposure, as opposed to direct exposure via personal glyphosate use. For FFES spouses and children who did not take part in the on-study glyphosate application (i.e., excluding those who helped with or were present for glyphosate mixing or application), only two (4%) of 48 spouses and one (2%) of 52 children had detectable glyphosate in their urine on the day of application (Acquavella et al. 2004).
For family members without a detectable urinary dose of glyphosate, it is difficult to determine whether any were actually exposed. For instance, some could have been away from the farm at work or at school on the day of glyphosate application, in which case no exposure might have occurred, as evidenced by the lack of a detectable amount of glyphosate in their urine. For indirectly exposed family members in the FFES who had detectable urinary glyphosate values, the maximum systemic doses were 0.00004 mg/kg for spouses and 0.0008 mg/kg for the one child with detectable glyphosate. Thus, an indirect exposure metric is not supported by glyphosate biomonitoring data, which suggest no appreciable dose from indirect exposure scenarios for residents on farms where glyphosate was applied. Accordingly, we determined a priori that in this literature review, we would classify any indirect exposure metrics as being low quality. Although other glyphosate exposure scenarios besides personal occupational use and indirect residential exposure are plausible-for example, spraying at home, ingestion, and dermal contact with recently sprayed crops-no such alternative exposure pathways were assessed in any epidemiological studies of glyphosate and neurotoxicity.
The FFES and other biomonitoring data also cast doubt on the validity of exposure metrics based on residential proximity to a glyphosate application. If measurable glyphosate is extremely rare among family members who were not involved directly in the glyphosate application on their farms, as indicated by these studies (Acquavella et al. 2004;Curwin et al. 2007;Niemann et al. 2015;Solomon 2016), then it is highly unlikely that those residing at some distance from an application would receive any appreciable dose of glyphosate from the attributed application. Reliance on residential proximity to a glyphosate application also introduces uncertainty about whether such individuals were at home and, if so, whether they spent any time outdoors on the day of the proximal application. Accordingly, in our literature review, we determined a priori that all studies with a proximity-based exposure metric would be rated as having lowquality exposure assessment. Similarly, ecological studies do not assign exposures to individuals; hence, there is no way of knowing the likelihood of contact with glyphosate on an individual level. Therefore, we also determined a priori that all ecological studies would be rated as having low-quality exposure assessment because of the obvious limitations of such an approach to exposure classification. may influence the magnitude of the risk estimate) (U.S. EPA 2016). Each of these five domains was counted once. Ratings were assigned based on consensus among all three investigators.
Exposure assessment According to the OPP framework, high quality (i.e., low risk of bias) in exposure assessment calls for use of an exposure metric with an "[a]ccurate and precise relationship with external exposure, internal dose, or target dose, possibly associated with a [mode of action/adverse outcome pathway]" (U.S. EPA 2016). However, none of the epidemiological studies included in this literature review measured glyphosate dose in biological specimens or provided validation of self-reported glyphosate use based on application records or biomonitoring. Therefore, to allow for variation in ratings among the available studies, and consistent with biomonitoring results from the FFES (Acquavella et al. 2004), we considered assessment of self-reported first-hand use of glyphosate, with details on frequency and/or duration of use of glyphosate, to be high quality. We rated assessment of self-reported first-hand use of glyphosate without such details (i.e., ever vs. never use) as being of moderate quality. For the scientific reasons provided in the prior section of this article, any indirect exposure measures other than direct personal use of glyphosate, including assessment based on geographic proximity to glyphosate applications, were rated as low quality.
Outcome assessment In line with the OPP framework, we rated outcome assessment as high quality if neurological conditions were classified based entirely on medical records and/or clinical examination by a medical professional. We considered outcome assessment to be moderate quality if it involved a self-reported doctor's diagnosis of a medical condition, a component of initial self-report followed by validation based on medical records, or ascertainment based on death certificates or diagnosis codes without additional review of medical records. We rated outcome assessment as low quality if it was based entirely on self-report or ecological (population-level) data.
Confounder control OPP's criteria for high-quality confounder control require comprehensively providing "[g] ood control" for "important confounders relevant to [the] scientific question" (U.S. EPA 2016), i.e., potential risk factors for neurological outcomes that may be correlated with glyphosate exposure, as well as standard confounders. Because none of the epidemiological studies identified in this literature review strictly met this definition, we allowed for a range of ratings among the available studies by classifying confounder control as high quality if statistical analyses were adjusted for multiple personal characteristics and other pesticides besides glyphosate; moderate quality if analyses were adjusted for multiple personal characteristics, but no other pesticides besides glyphosate; and low quality if analyses were adjusted for few or no personal characteristics and no other pesticides.
Statistical approach Following OPP guidance, we rated the statistical approach as high quality if study authors performed a statistical analysis that was appropriate to the study question and design, supported by adequate sample size, with consideration of approaches to address potential sources of bias. The statistical approach was considered moderate quality if study authors performed an appropriate statistical analysis without consideration of potential bias or with limited sample size; and low quality if statistical comparisons were minimal or not described clearly.
Other sources of bias In line with the OPP framework, we considered other potential sources of bias based on a combination of study design and study population, bearing in mind that if properly executed, both cohort and case-control study designs can yield valid results. We rated studies as high quality if they were prospective in design with minimal concerns for selection bias or other sources of bias, for example, based on study recruitment and follow-up methods, inclusion/ exclusion criteria, and participation rates; moderate quality if they were prospective in design with moderate concerns for selection bias, or retrospective and population-based in design with high participation; and low quality if they were ecological, cross-sectional, retrospective hospital-based, retrospective population-based with low participation, or case-only in design.
Overall Study quality was classified based on the balance of quality ratings across all five domains. Studies that relied on low-quality information on glyphosate exposure are inherently unable to determine whether individuals did or did not receive any dose of glyphosate, rendering them uninformative about possible health impacts. Therefore, we considered all studies with low-quality exposure assessment to be uninformative.

Results
The methodological characteristics and results of the 13 moderate-and high-quality epidemiological studies of glyphosate and neurological outcomes are summarized in Table 1, grouped by category of neurological outcome. The remaining 12 low-quality studies are summarized in Supplemental Table 1. A heat map of study-specific ratings of the risk of bias in each of the five domains of exposure assessment, outcome assessment, confounder control, statistical approach, and other aspects, as well as the overall study quality rating, is provided in Table 2, along with a brief rationale for each rating. Three studies that used different methods depending on the exposure metric (Beard et al. 2013;Montgomery et al. 2017) or the neurological outcome (Garry et al. 2002) received two ratings in a single domain. A simplified version of the heat map without explanatory text is provided as Supplemental Fig. 1.

Neurodegenerative outcomes
We rated two studies of neurogenerative outcomes as high quality overall (Tables 1, 2). These were overlapping analyses of Parkinson disease in the Agricultural Health Study (AHS) cohort: the first study evaluated prevalent and incident cases in the full initial cohort (Kamel et al. 2007), and the second study was an updated analysis including only incident cases, based on cohort members with at least one follow-up assessment (Shrestha et al. 2020). The AHS is a prospective cohort study of more than 57,000 licensed pesticide applicators (mostly farmers) from Iowa and North Carolina and more than 32,000 of their spouses, all of whom were initially enrolled in 1993-1997 by completing questionnaires about agricultural practices, including personal mixing or application of 50 specific pesticides, as well as lifestyle and health factors (Agricultural Health Study 2021). Pesticide applicators reported additional information on frequency and duration of pesticide use, as well as mixing practices, application methods, repairing of application equipment, and use of personal protective equipment. (Most studies of neurological outcomes in the AHS, however, provided results only for ever vs. never glyphosate use.) Follow-up interviews or questionnaires were completed in 1999-2003, 2005-2010, and 2013-2015, and cohort members were linked to vital statistics and state cancer registries to ascertain cause-specific mortality and cancer incidence. Certain self-reported health outcomes were confirmed by medical records or clinical examination.
As summarized in Tables 1 and 2, although both of the AHS studies of Parkinson disease relied to some extent on self-reported health endpoints, resulting in a moderatequality rating for outcome assessment, they were considered to have relatively high-quality (i.e., low-bias) exposure assessment, confounder control, statistical approach, and other aspects of study design and study population. Both high-quality studies of neurodegenerative outcomes estimated near-null associations (i.e., relative risks close to 1.0) between glyphosate use and risk of Parkinson disease (Kamel et al. 2007;Shrestha et al. 2020) (Table 1).
Two other AHS cohort studies of neurodegenerative outcomes, ALS mortality and age-related macular degeneration, were rated as moderate quality overall (Kamel et al. 2012;Montgomery et al. 2017), as was a hospital-based case-control study of Parkinson disease in East Texas (Dhillon et al. 2008) (Tables 1, 2). These AHS studies were downgraded to moderate quality based on their limited adjustment for confounding, combined with ascertainment of ALS based primarily on death certificates (Kamel et al. 2012), which are prone to classification error (Chio et al. 1992;Middleton et al. 2011); or ascertainment of age-related macular degeneration based initially on self-report, followed by medical confirmation of only positive reports (Montgomery et al. 2017). Risk of age-related macular generation was positively associated with the highest category of glyphosate exposure (> 100 lifetime days), but ever use of glyphosate was not associated with risk after more rigorous confounder adjustment (Montgomery et al. 2017). No association was observed between glyphosate use and death from ALS (Kamel et al. 2012).
The hospital-based case-control study of Parkinson disease in East Texas, also rated as moderate quality, found no association between ever use of glyphosate and risk of neurologist-diagnosed Parkinson disease (Dhillon et al. 2008) (Tables 1, 2). This study design had substantial problems with selection bias and confounding, but had high-quality exposure and outcome assessment.
The remaining four studies of neurodegenerative outcomes were rated as low quality overall (Table 2 and Supplemental Table 1). These included three ecological or semi-ecological studies of ALS or Parkinson disease that used regional data on agricultural applications of glyphosate (Andrew et al. 2021;Caballero et al. 2018;Wan and Lin 2016); and a small, pilot hospital-based case-control study of Parkinson disease that was rated as low quality due to its inclusion of some unvalidated cases identified from patient support groups, as well as its absence of any adjustment for confounding (Wechsler et al. 1991). Due to their serious methodological weaknesses, which limited their ability to provide scientific insight on any potential neurotoxic effects of glyphosate, these low-quality studies were not considered further in our weight-of-evidence evaluation.

Neurobehavioral outcomes
Two studies of neurobehavioral outcomes were rated as high quality overall; these were non-overlapping analyses of depression (one in women and one in men) in the prospective AHS cohort (Beard et al. 2013;Beard et al. 2014) (Tables 1, 2). Both of these studies reported near-null results, with relative risks close to 1.0, for associations between glyphosate use and risk of depression (Beard et al. 2013;Beard et al. 2014).
In a moderate-quality cross-sectional study of a convenience sample of 288 Ugandan farmers, self-reported glyphosate use, which was classified semi-quantitatively using detailed exposure information, was associated with poorer visual retention, but not with several other neurobehavioral outcomes assessed using a standard battery of tests (Fuhrimann et al. 2021) (Tables 1, 2).
The other two studies of neurobehavioral outcomes were considered uninformative for our weight-of-evidence evaluation due to their poor methodological quality (Table 2 and Supplemental Table 1). These were cross-sectional analyses with low-quality exposure assessment based on farm-level but not personal glyphosate use in relation to self-reported minor psychiatric disorders (Faria et al. 2014); and selfreported acute occupational glyphosate poisoning in relation to self-reported depressive symptoms (Kim et al. 2013). In the latter study (Kim et al. 2013), farmers were classified as exposed (i.e., poisoned) if they retrospectively reported having experienced any of 21 non-specific symptoms and signs (nausea, vomiting, diarrhea, sore throat, runny nose, dyspnea, headache, dizziness, hyperactivity, profuse sweating, blurred vision, paresthesia, slurred speech, paralysis, chest pain, syncope, muscle weakness, skin irritation, eye irritation, lacrimation, and fatigue) within 48 h of using pesticides in the prior year. Subjects who did not report such episodes were classified as non-poisoned; however, no distinction was made between non-poisoned farmers who did and did not apply glyphosate. Given the non-specific nature of the exposure assessment and the inability to distinguish between glyphosate use and non-use, we rated the exposure metric in this study as low quality.

Neurodevelopmental outcomes
None of the epidemiological studies of glyphosate exposure and neurodevelopmental outcomes were rated as high quality (Table 2). Among the six identified studies of neurodevelopmental endpoints, only one-a hospital-based case-control study of suspected developmental delay among young children in Thailand (Juntarawijit et al. 2020)-merited a rating of moderate quality. This study found no association between maternal self-reported prenatal, postnatal, or overall use of glyphosate and suspected developmental delay among children under age 5 years who participated in Thailand's National Child Developmental Screening Program (Juntarawijit et al. 2020) (Tables 1, 2). Notably, five of the six studies of neurodevelopmental outcomes were rated as low quality (Table 2 and Supplemental Table 1). These included one ecological study (ranked as lowest in quality) that was considered low quality due to its use of state-level estimates of glyphosate agricultural applications and autism prevalence over time (Nevison 2014); three semi-ecologic studies that were considered low quality due to their use of residential proximity to glyphosate applications as the exposure metric in relation to neural tube defects, autism spectrum disorder, and neural tube defects and orofacial clefts, respectively (Rull et al. 2006;von Ehrenstein et al. 2019;Yang et al. 2014); and a crosssectional analysis that was rated as low quality due to its reliance on unvalidated parent-reported data on childhood autism or attention deficit/hyperactivity disorder (as well as confirmed parent-reported data on congenital central nervous system anomalies), cross-sectional design, and no or minimal confounder adjustment (Garry et al. 2002). Due to their low overall quality, these studies were not considered further in our weight-of-evidence evaluation.

Other and mixed neurological outcomes
One high-quality short-term prospective cohort study evaluated glyphosate exposure in relation to peripheral nerve conduction (categorized as an "other" neurological outcome) among Chinese farmers, with relatively high-quality exposure assessment, outcome assessment, confounder control, and statistical approach (Zhang et al. 2018) (Tables 1, 2). In this study, 218 farmers in three provinces in China were followed for the 2012 planting season. The farmers kept detailed records of pesticide use throughout the duration of the study, and nerve conduction parameters were measured at the beginning and near the end of the planting season. Strengths of this study include the use of well-controlled nerve conduction tests and careful real-time assessment of pesticide use, including biweekly checking of farmers' records and pesticide containers. Multivariable regression was used to control for several potential confounding factors, including other pesticide classes. Limitations are the short 1-year timeframe considered for potential chronic effects of pesticides, the lack of measurement of biological dose of glyphosate, and not having more extended pesticide histories for the subjects. Use of glyphosate was not found to be associated with any measure of nerve conduction, whether classified as abnormal vs. normal or based on the total number of abnormal parameters noted in a given domain.
Two AHS analyses were rated as moderate quality due to their reliance on exclusively self-reported, unvalidated neurological outcomes, namely, dream-enacting behaviors and olfactory impairment (Shrestha et al. 2018;Shrestha et al. 2021) (Tables 1, 2). Both of these studies reported mostly positive associations with glyphosate use, including semi-quantitative measures of intensity-weighted cumulative glyphosate use in the analysis of olfactory impairment (Shrestha et al. 2021).
A cross-sectional analysis in the Ugandan farmer population, also rated as moderate quality, showed that selfreported glyphosate use in the past week, but not otherwise in the past year, was associated with a higher prevalence of self-reported overall sleep problems in the past week (Fuhrimann et al. 2022) (Tables 1, 2). However, several other measures of recent sleep problems assessed in this study were not associated with frequency of self-reported glyphosate use in the past week.
Otherwise, a low-quality ecological study that linked U.S. national-level data on glyphosate agricultural applications and various neurological outcomes, including hospital discharge rates, mortality rates, and childhood autism case counts, was identified but not considered further in our 1 3 information on neurological conditions, some exclusively so (Beard et al. 2013(Beard et al. , 2014Faria et al. 2014;Fuhrimann et al. 2022;Garry et al. 2002;Kamel et al. 2007;Kim et al. 2013;Shrestha et al. 2018Shrestha et al. , 2021, thereby introducing a degree of misclassification that could have been extensive in some studies and, potentially, differential by exposure status. Confounding remains a concern in virtually all studies, both for neurological disorders with many known risk factors (e.g., sleep problems, depression) and those with as yet few identified risk factors (e.g., Parkinson disease, ALS). None of the available studies had strictly highquality confounder control based on "[g]ood control for important confounders relevant to [the] scientific question, and standard confounders" (U.S. EPA 2016). When infrequent use of a specific pesticide, such as glyphosate, takes place in a milieu of numerous other chronic exposures with potential biological activity, including other pesticides and aspects of an agricultural occupation or lifestyle, uncontrolled confounding and residual confounding (the latter due to insufficient adjustment for measured confounders) are difficult to eliminate.
In conclusion, based on our systematic review of the available epidemiological literature, we found no consistent evidence of a causal effect, or even a convincing statistical association, between glyphosate exposure and any neurological outcome in humans. To meaningfully advance our understanding of the potential neurotoxicity of glyphosate in humans, more epidemiological studies are needed in populations that have frequent direct exposure, validated outcome assessment, minimization of selection bias through achievement of high participation and retention rates, and rigorous statistical adjustment for potential confounding, especially for important personal factors and correlated pesticides. Past biomonitoring studies indicate that poor-quality exposure assessment or biologically implausible hypotheses involving ever-vs.-never use of glyphosate or indirect exposure pathways are unlikely to be informative. Nevertheless, additional biomonitoring studies could provide valuable insight into the validity of indirect exposure metrics based on geographic models or reported use by household members. Epidemiological research in this realm requires scientific prudence in determining which causal hypotheses may have merit, and whether the appropriate epidemiological data are available or attainable to investigate those hypotheses in a manner that yields valid and informative results.