Ambient Trichloroethylene Exposure and Parkinson Disease Risk in Medicare Beneficiaries

Abstract

Background and Objectives

Trichloroethylene (TCE) is an important environmental contaminant in the United States due to widespread use industrially. Epidemiologic studies suggest that occupational exposure to TCE and TCE-contaminated drinking water may increase the risk of Parkinson disease (PD). The aim of this study was to investigate the nationwide relationship between ambient TCE and PD risk.

Methods

We performed a nationwide, population-based, case-control study to investigate the association between incident PD in US Medicare beneficiaries aged 67 years and older in 2016–2018 and their residential exposure to ambient (outdoor) TCE in 2002. We assigned residence based on the latitude and longitude of Medicare beneficiaries' zip + 4 center 2 years before diagnosis/reference. We assigned TCE exposure based on census tract–level data from the US Environmental Protection Agency's (EPA's) National Air Toxics Assessment program. We used logistic regression to estimate relative risk (RR) adjusted for age, sex, race, smoking, health care utilization, rural-urban commuting area (RUCA), and PM_2.5. We also mapped the nationwide geospatial pattern of ambient TCE and then explored high-resolution local PD risk patterns using MapGAM for the 10-mile radius around the top 3 TCE-emitting facilities from the EPA's Toxic Release Inventory in 2002.

Results

We identified 221,789 incident PD cases (median age 78.8 years; 45% female) and 1,132,765 matched controls (median age 75.7 years; 57% female). We found a dose-dependent positive association between ambient TCE concentrations and PD risk, wherein beneficiaries exposed to the top decile of ambient TCE levels (0.14–8.66 μg/m³) had an RR for PD of 1.10 (95% CI 1.08–1.13) compared with those exposed to the lowest decile of TCE (0.005–0.01 μg/m³). We observed high levels of ambient TCE in the rust belt region of the United States and several smaller areas throughout the nation. Our MapGAM results suggested greater PD risk in the area surrounding 2 of the 3 highest ambient TCE-emitting facilities, one of which demonstrated a marked decreasing gradient of risk with increasing distance from the facility.

Discussion

We identified a positive association between ambient TCE and PD risk, suggesting that TCE may contribute to PD. We identified relatively higher risk of PD near 2 TCE-emitting facilities. This study was limited to Medicare-aged individuals.

Introduction

Parkinson disease (PD) remains largely an idiopathic disease although most of the cases are suspected to result from exposure to environmental neurotoxicants. Trichloroethylene (TCE) is a widespread environmental toxicant in the United States, commonly found in both water and air.¹ TCE is a chlorinated solvent used in a wide variety of industrial processes and products. In the 1920s, TCE was used commonly as a dry cleaning and degreasing agent and could be found in many household cleaning products.² The use of TCE in dry cleaning decreased in the 1950s, and in the 1970s, the US Food and Drug Administration banned the use of TCE as an inhaled anesthetic.¹ Nevertheless, TCE was still used as a degreasing agent and for other purposes in a wide variety of settings including metal and textile manufacturing³ in the United States as recent as 2025. Environmental exposure to TCE is relatively common. When assessed in the year 2000, the US Environmental Protection Agency's (EPA's) Toxic Release Inventory Program reported that up to 30% of US drinking water supplies were contaminated with TCE.^4,5

Owing in part to its ubiquitous use in manufacturing, most human studies of TCE and PD have focused on occupational exposure.^6-9 A recent study found that occupational TCE exposure was highest in regions with a high-middle sociodemographic index, and that occupational exposure has been increasing over the past 3 decades, with the fastest rate of increase in North American and Southern Sub-Saharan Africa.¹⁰ Occupational TCE exposure is anticipated to increase globally through 2050.¹⁰ Nevertheless, TCE was set to be banned in the United States in 2024 (although this ban was paused a few months later in 2025), but whether developing nations follow this example remains to be seen. It will take many years for the negative health impacts of TCE to subside, given the long prodromal period of disease and the ubiquitous presence of TCE in the environment. In addition, the potential ban stipulates a planned phase-out period of up to 10 years, with some workplace uses granted a longer phase-out time frame.¹¹ In addition, TCE is often difficult to remediate from the environment, and in some cases, it can take decades to complete remediation of a TCE-contaminated site.¹² Thus, those living near existing TCE-contaminated sites may continue to be exposed long after a ban takes effect.

When inhaled or ingested, TCE readily crosses the blood-brain barrier¹³ and causes dopaminergic neurodegeneration in animal studies.^14-16 Notably, human studies have implicated TCE in PD pathogenesis.^6-9 Recently, a study found greater risk of PD among those who had lived in Camp Lejeune, a military base in North Carolina with TCE-contaminated water, compared with a military base in California without TCE water contamination.¹⁷ However, each year, millions of people are exposed to TCE through indoor and outdoor air pollution¹ and, in fact, inhalation is the most common route of TCE exposure.^3,18 We sought to determine whether exposure to outdoor ambient TCE is associated with PD risk in a nationwide study using US Medicare data. We hypothesized that we would find a positive association between outdoor ambient TCE exposure and PD risk.

Methods

Standard Protocol Approvals, Registrations, and Participant Consents

This study was approved by the institutional review boards at Barrow Neurological Institute and Washington University School of Medicine in St. Louis and by the Centers for Medicare and Medicaid Services, which deidentified data before release. Participant consent in this administrative data–based study was not required.

Study Population and Design

We constructed a population-based case-control study using Medicare research files¹⁹: master beneficiary summary files from 2014 to 2018, comprehensive Medicare claims data (inpatient, outpatient, carrier, skilled nursing facility, home health care, hospice, and durable medical equipment) from 2014 to 2018, and an extract of residential location (zip + 4) histories from the Medicare enrollment database. We identified and included all incident PD cases in 2016–2018 who met case and study eligibility criteria detailed further. We enumerated all noncases who met the same study criteria and randomly selected 5 controls for each case while frequency matching by month and year of diagnosis, that is, first diagnosis code for PD. We assigned as a control reference date a randomly selected date in the respective month and year. Our eligibility criteria were designed to ensure a population-based sample with complete data: age-eligible for Medicare ≥2 years before diagnosis/selection (age ≥67), no Part C (Medicare Advantage plan/health maintenance organization) coverage, age 110 years and younger, and US residence, all in the respective month and year of case diagnosis or control reference, and no atypical parkinsonism or Lewy body dementia. We also required sufficient geographic and covariate data for inclusion in this study, that is, demographic data and residential zip + 4 information 2 years before PD diagnosis/control reference date, that is, on a date in 2014–2016.

PD Case Identification

Incident PD cases included all study-eligible beneficiaries with ≥1 ICD-10-CM diagnosis code G20 in 2016–2018, but no ICD-9-CM (332.0) or ICD-10-CM code for PD in previous years to exclude prevalent cases. This case definition maximizes sensitivity without materially affecting specificity²⁰ and aims to ensure representativeness of cases, including regarding environmental exposures.

Ambient TCE Exposure Estimation

Our exposure of interest was annual average ambient TCE exposure, specifically environmental concentrations in residential outdoor air in the years before PD diagnosis, that is, at an etiologically relevant time point, given the long prodromal period of PD.²¹ We obtained ambient TCE estimates from the US EPA National Air Toxics Assessment (NATA),²² which derived TCE estimates based on air dispersion and photochemical models before aggregating to the census tract level.²³ For our primary analyses, we used NATA data from 2002 (14–16 years before PD diagnosis/control selection), the earliest year available after recent modeling advancements in TCE estimation were used. For comparison, we used the most recent NATA data in or before the residence year (NATA 2014 data, 2–4 years before diagnosis/selection and aligned with the residence data). We then used geographic information systems (ArcMap version 10.8.2) to link the respective ambient TCE concentration to Medicare beneficiaries' residence locations based on the census tract associated with the latitude and longitude of each beneficiary's respective zip + 4 location.

Assessment of Demographics and Other Covariates

We determined beneficiary demographic information (age, sex, race) from the master beneficiary summary file for the diagnosis/reference year. We defined health care utilization as the number of unique diagnosis codes for each beneficiary from 2014 to 2 years before the diagnosis/reference date, as calculated from the complete claims data. We previously demonstrated that adjusting for the number of diagnosis codes to capture confounding by health care utilization provides more complete adjustment in studies of PD than using more traditional measures, such as the Elixhauser comorbidity index.²⁴ We used these same claims data to estimate previous traumatic brain injury (TBI) (eTable 1A) and the probability of ever smoking for each beneficiary with a validated claims-based algorithm.²⁵ All the abovementioned covariates were at the individual level. We also obtained 2 geographic-level covariates: average annual mean particulate matter with a diameter <2.5 μm (PM_2.5) for the years 1998–2000 (1-km resolution)²⁶ and census tract–level data on rural-urban commuting areas (RUCAs) for 2010 from the US Department of Agriculture.²⁷

Statistical Analysis: Case-Control Study

We performed logistic regression, with PD as the outcome and census tract–level ambient TCE concentration as the independent variable, modeled as detailed further. We obtained the odds ratio (OR) and 95% CI and expressed this OR as the relative risk (RR) because PD is a rare outcome. We adjusted a priori for age, sex, race, smoking, health care utilization, RUCA, and average annual mean PM_2.5.²¹ We modeled age continuously, race as a categorical variable (6 categories), smoking and health care utilization in deciles (modeled linearly based on the Akaike information criterion), RUCA as a dichotomous variable (metropolitan core area vs other areas), and average annual mean PM_2.5 with 2 linear splines as previously.²¹

Based on results from the Box-Tidwell test for nonlinearity, we modeled TCE nonlinearly. Specifically, we modeled TCE as a continuous variable after applying a natural log transformation, which provided better model fit as determined by the Akaike information criterion. We also allowed for a less constrained relationship by modeling TCE as a categorical (decile) variable. We selected the lowest decile of TCE (0.005–0.01 μg/m³) as the reference group. The variance inflation factor (VIF) indicated that spline 1 of PM_2.5 was highly co-linear with TCE, and thus, we omitted spline 1 from our models. To illustrate the impact of the adjustments, we provide models of the relationship between TCE and PD with and without adjustment for PM_2.5 (through spline 2) and RUCA. To assess whether restriction to Medicare-aged beneficiaries could limit generalizability, we tested whether the TCE-PD association differed by age. We also assessed whether the relationship differed by sex. Moreover, we assessed RUCA as a potential effect modifier by modeling it as a dichotomous variable (major metropolitan core area vs nonmetropolitan core area). We tested for an interaction using a likelihood ratio test comparing models with and without an interaction term. As a sensitivity analysis, we performed our analysis using TCE data from 2014.

Statistical Analysis: Clinical Features

We assessed the association between TCE and several clinical features of PD recorded before diagnosis, including tremor, gait abnormalities, falls, and dementia, which were defined using ICD-9-CM and ICD-10-CM codes (eTable 1, B–E). Models were adjusted for the same factors detailed above.

Mapping Nationwide TCE Patterns and Local PD Risk Maps Around Facilities

To identify the national pattern of ambient TCE exposure, we mapped TCE at the census tract level. We compared patterns of exposure with patterns of PD risk (eFigure 1). To better understand the effect of outdoor ambient TCE at a local level, we supplemented our nationwide analysis with an exploratory study of the local patterns of PD risk around the top 3 ambient TCE-emitting facilities in 2002. Specifically, we used location data on ambient TCE-emitting facilities in 2002 from the EPA's Toxic Release Inventory²⁸ and the zip + 4 locations of PD cases and all noncases in a 10-mile radius who met study eligibility criteria to create a high-resolution risk map around the facility. We used MapGAM, a novel geostatistical method that uses generalized additive models (GAMs) and spatially referenced beneficiary information for cases and controls to create high-resolution PD risk surfaces that incorporate smoothing and adjust for known confounding factors. Our MapGAM models adjusted for age, sex, race, health care utilization, smoking probability (predicted from demographic information and smoking-specific diagnoses),²¹ and average annual mean PM_2.5 from 1998 to 2000.²¹ All maps were created using all incident PD cases and all available noncases within a 10-mile radius of the facility. High-risk regions were symbolized in ArcPro, and “zones of significance” were outlined using dashed lines. MapGAM provides “zones of significance” to delineate regions where the 95% CIs do not include an RR of 1.00. Maps were only created if results contained at least 1 zone of significance. MapGAM is computationally intensive, so we focused on the top 3 TCE-emitting facilities only. We also incorporated data on estimated average hourly wind direction from 1970 to 2018 into maps to visualize the potential for emissions transport.²⁹ As a sensitivity analysis, we performed our MapGAM analysis after restricting to women to assess whether risk patterns differed according to sex. In doing so, we aimed to gain insight into the extent to which risk around the facility might be due to occupational clustering, because of the fact that women aged 65+ were expected to be less likely to work or have previously worked at the facilities compared with men aged 65+. We hypothesized that risk maps restricted to women would reflect primarily environmental exposures.

Data Availability

The Centers for Medicare and Medicaid Services does not permit data sharing under the data use agreement.

Results

We identified 286,589 incident PD cases and selected 1,432,945 (5:1) controls who met our study eligibility criteria for our case-control analysis. After excluding 64,800 cases and 300,180 controls without valid zip + 4 information for the contiguous United States, we were left with 221,789 cases and 1,132,765 controls for our study. In this case-control sample, PD cases were male (55%), White (86%), and older (mean 78.8 vs 75.7); the percentages of these demographic characteristics were higher than in controls (Table 1). The proportion of PD cases who reside in a metropolitan core area is higher than controls. Conversely, the percentages of any of the other 9 RUCA categories were lower in cases than in controls (Table 1). Of the beneficiaries with address information available for up to 8 years before diagnosis/selection, we found that 82% of cases and 84% of controls maintained the same 5-digit zip code for all 8 years before their diagnosis/selection. For our MapGAM analysis, 11,018 noncases lived within a 10-mile radius of the 3 facilities.

Table 1.

Characteristics of Incident PD Cases and Controls With Census Tract–Level Ambient TCE Data Based on Zip + 4 Location Information (US Medicare 2016–2018)

	PD cases (N = 221,789)	Controls (N = 1,132,765)
Female, n (%)	100,492 (45)	640,791 (57)
Race/ethnicity, n (%)
White	190,769 (86)	955,230 (84)
Black	12,373 (6)	80,441 (7)
Hispanic	11,130 (5)	54,657 (5)
Asian/Pacific Islander	5,527 (2)	30,422 (3)
Native American	482 (0.2)	3,087 (0.3)
Other	1,508 (0.7)	8,928 (0.8)
Age, y, mean (SD)	78.8 (7.30)	75.7 (7.33)
RUCA (code), n (%)
Metropolitan core area (1)	143,107 (64.52)	715,062 (63.13)
Metropolitan area high commuting (2)	23,249 (10.48)	127,299 (11.24)
Metropolitan area low commuting (3)	1,674 (0.75)	9,072 (0.80)
Micropolitan core area (4)	18,904 (8.52)	94,680 (8.36)
Micropolitan high commuting (5)	7,687 (3.47)	41,148 (3.63)
Micropolitan low commuting (6)	1,401 (0.63)	7,670 (0.68)
Small town core (7)	9,910 (4.47)	51,343 (4.53)
Small town high commuting (8)	2,818 (1.27)	15,431 (1.36)
Small town low commuting (9)	1,972 (0.89)	10,288 (0.91)
Rural area (10)	11,039 (4.98)	60,588 (5.35)

Abbreviations: PD = Parkinson disease; RUCA = rural-urban commuting area; TCE = trichloroethylene.

When modeling ambient TCE exposure as a natural log-transformed variable, we found a positive linear association between TCE exposure and PD risk (RR 1.03; 95% CI 1.02–1.032); that is, there was approximately 3% greater risk of PD with every 1-unit increase in the log scale of TCE exposure (Table 2; p_trend < 0.0001). When TCE was modeled as deciles, we found that individuals living in census tracts with TCE levels in the top decile (0.14–8.66 μg/m³) had an RR of 1.10 (95% CI 1.08–1.13) compared with the lowest decile (0.005–0.01 μg/m³) (Figure 1; Table 2). When modeling ambient TCE exposure as a natural log-transformed variable, we found a positive linear association between TCE exposure and prodromal gait abnormalities (OR 1.07; 95% CI 1.06–1.08), falls (OR 1.04; 95% CI 1.03–1.05), and dementia (OR 1.05; 95% CI 1.04–1.06) and a slight inverse association with tremor (OR 0.98; 95% CI 0.97–0.99) (eTable 2). When modeled as a categorical (decile) variable, the association with tremor was positive up until the 8th and 9th decile (eTable 3). We did not find an interaction with sex (p_interaction = 0.5502), age (p_interaction = 0.3092), or RUCA (p_interaction = 0.1707). Additional adjustment for previous TBI did not have a notable impact on the point estimates (eTable 4). As a sensitivity analysis, we also performed our analysis using TCE exposure data from 2014 and found that results attenuated but remained significant (results not shown).

Table 2.

Outdoor Ambient TCE and PD Risk (US Medicare 2016–2018)

Census tract–level ambient TCE, μg/m3, range (median)	Cases (N = 221,789)	Controls (N = 1,132,765)	Basic adjusted model RR (95% CI)a	Further adjusted for RUCA and PM2.5 RR (95% CI)a,b,c
D1: 0.005–0.01 (0.01)	20,647	114,775	1.00 (reference)	1.00 (reference)
D2: 0.01–0.02 (0.02)	21,993	113,475	1.04 (1.02–1.06)d	1.04 (1.02–1.07)d
D3: 0.02–0.03 (0.02)	21,622	113,836	1.06 (1.04–1.08)d	1.06 (1.04–1.09)d
D4: 0.03–0.04 (0.03)	22,008	113,460	1.08 (1.06–1.10)d	1.08 (1.06–1.11)d
D5: 0.04–0.05 (0.04)	21,871	113,583	1.09 (1.06–1.11)d	1.08 (1.06–1.11)d
D6: 0.05–0.07 (0.06)	22,020	113,431	1.08 (1.06–1.11)d	1.08 (1.06–1.11)d
D7: 0.07–0.09 (0.08)	22,606	112,833	1.10 (1.08–1.12)d	1.09 (1.07–1.12)d
D8: 0.09–0.11 (0.10)	22,769	112,693	1.08 (1.05–1.10)d	1.06 (1.04–1.09)d
D9: 0.11–0.14 (0.13)	23,281	112,179	1.10 (1.08–1.13)d	1.09 (1.06–1.11)d
D1: 0.14–8.66 (0.18)	22,972	112,500	1.12 (1.09–1.14)d	1.10 (1.08–1.13)d
			ptrend < 0.0001	ptrend < 0.0001
TCE (natural log)			1.03 (1.02–1.032)d,e	1.03 (1.02–1.032)d,e
			ptrend < 0.0001	ptrend < 0.0001

Abbreviations: PD = Parkinson disease; PM_2.5 = particulate matter; RR = relative risk; RUCA = rural-urban commuting area; TCE = trichloroethylene.

^aAdjusted for individual-level variables (age, sex, race, ever/never smoking, and health care utilization), based on cases and controls with geocodable zip + 4 location for the residence 2 years before PD diagnosis or control reference date and census tract–level TCE data.

^bExcludes 212 beneficiaries with an unknown RUCA code.

^cFurther adjusted for RUCA and spline 2 of average annual mean PM_2.5. Spline 1 was omitted from the model because of multicollinearity.

^dp < 0.05.

^ePer-unit change, that is, approximately risk per doubling of TCE exposure.

Figure 1. Outdoor Ambient TCE and Relative Risk of Incident PD in United States.

The relationship between annual average ambient TCE concentrations and risk of PD nationwide, with TCE on the x-axis and the relative risk of PD on the y-axis. TCE categories are based on deciles of TCE exposure in Medicare beneficiaries. The map depicts census tracts with ambient TCE estimates in the top decile (0.14–8.66 μg/m³) in 2002. PD = Parkinson disease; TCE = trichloroethylene.

Census tract–level mapping revealed the highest levels of ambient TCE clustering in the “rust belt” region of the United States (Figure 2). In our exploratory MapGAM study of local patterns of risk around the top 3 TCE-emitting facilities in the United States in 2002, we found relatively greater PD risk proximate to 2 of the TCE emitters that both manufacture separators for lead-acid batteries compared with areas further from the facilities. One of these 2 TCE-emitting facilities exhibited a marked gradient in risk, wherein risk of PD decreased with increasing distance from the facility (Figure 3). This facility was contained within a “zone of significance.” For this facility in Oregon, the mean RR of PD was 4.41 (95% CI 1.40–14.36) greater within approximately 1–5 miles surrounding the facility (the zone of significance) when compared with areas further away from the facility (up to 10 miles). The highest RRs of PD were observed southeast (downwind) of the Oregon facility. The facility in Indiana was not contained within the zone of significance; however, there was a zone of significance 0.2 miles east of the facility (Figure 4) and a general pattern of relatively higher risk near the facility. In addition, MapGAM also identified a separate zone of significance approximately 8 miles northeast of the facility. There were no zones of significance detected in the MapGAM models of the facility in Wisconsin, which manufactures steel tubing, and thus, no map results are provided. Sensitivity analyses restricted to women beneficiaries revealed high risk of PD surrounding all 3 facilities; however, these analyses relied on too few beneficiaries to be presented because of CMS restrictions. Patterns of high risk spanned across areas of both high and low population densities (eFigures 2 and 3).

Figure 2. Ambient TCE, United States 2002.

A map of census tract–level annual average ambient TCE concentrations in the United States in 2002 based on US EPA NATA data. EPA = Environmental Protection Agency; NATA = National Air Toxics Assessment; PD = Parkinson disease; TCE = trichloroethylene.

Figure 3. High-Resolution PD Risk Surface Centered on the Ambient TCE-Emitting Facility in Oregon.

High-resolution PD risk maps created using the zip + 4 location information for all study-eligible incident PD cases and noncases in Medicare for the years 2016–2018 within a 10-mile radius of the facility. The map (10-mile radius) is centered on a facility that emitted 198,327 pounds of ambient TCE in 2002. The map adjusts for age, sex, race, smoking probability, health care utilization, and PM_2.5. Wind is represented as flow vectors based on wind roses of hourly wind speed and direction (average wind speed 6.2 mph) for 1972–2018 in Corvallis Muni (∼17 miles west of facility).²⁹ The dashed line area denotes significant high-risk regions where the 95% CIs do not include an RR of 1.00. PD = Parkinson disease; PM_2.5 = particulate matter; RR = relative risk; TCE = trichloroethylene.

Figure 4. High-Resolution PD Risk Surface Centered on the Ambient TCE-Emitting Facility in Indiana.

High-resolution PD risk maps created using the zip + 4 location information for all study-eligible incident PD cases and noncases in Medicare for the years 2016–2018 within a 10-mile radius of the facility. The map (10-mile radius) is centered on a facility that emitted 1,098,811 pounds of ambient TCE in 2002. The map adjusts for age, sex, race, smoking probability, health care utilization, and PM_2.5. Wind is represented as flow vectors based on wind roses of hourly wind speed and direction (average wind speed 8.0 mph) for 1970–2018 in Louisville (∼22 miles east of facility).²⁹ Dashed line areas denote significant high-risk regions where the 95% CIs do not include an RR of 1.00. PD = Parkinson disease; PM_2.5 = particulate matter; RR = relative risk; TCE = trichloroethylene.

Discussion

In this large, population-based case-control study investigating the relationship between ambient TCE exposure and PD risk, we found a nationwide, dose-dependent positive association. PD risk was 10% greater for populations exposed to levels of ambient TCE in the top decile compared with populations exposed to the lowest decile of TCE. In addition, prodromal symptoms of gait abnormalities, falls, and dementia were more common in PD cases living in neighborhoods with higher levels of ambient TCE. A key strength of our study is that we used a large, population-based data set restricted to incident disease, which is important for studies designed to investigate disease etiology. Furthermore, we leveraged innovative geographic information systems methods including census tract–level mapping and MapGAM to better understand the spatial patterns of ambient TCE exposure and its relation to PD. Our study adds to the existing literature that has demonstrated associations between PD and occupational TCE exposure^6,7,9,30 and potential exposure to TCE-contaminated tap water.¹⁷ Our innovative geographic analyses also provide insight into specific regions of the nation for further investigation.

A previous study of occupational solvent exposure found that previous exposure to TCE (determined using a questionnaire) was associated with markedly increased risk of PD, and that the onset of PD symptoms could occur 10–40 years after initial TCE exposure.⁶ Murine studies corroborate these findings, demonstrating that TCE exposure results in dopaminergic neuronal cell death in the midbrain,⁷ α-synuclein accumulation, and neuroinflammation^14,31—including a recent study of low-dose TCE inhalation in rodents.¹⁶ In addition, studies have linked potential environmental TCE exposure to increased risk of PD at Camp Lejeune, a US Marine Corps Base in North Carolina.^2,17,32-35 At Camp Lejeune, TCE exposure is suspected to have occurred from a water treatment plant that was contaminated with TCE between 1953 and 1987. Thousands of people living and working at the base were potentially exposed to TCE during this time. A recent study found that overall PD risk was 70% higher in service members stationed at Camp Lejeune between 1975 and 1985 compared with those stationed at Camp Pendleton, where water was not contaminated with TCE. Investigators found that PD risk was 28% higher when restricting to PD diagnoses made before PD becoming a presumptive service-connected condition among the exposed group.¹⁷ Our study adds to this previous work by demonstrating an association between ambient TCE exposure and risk of PD.

In our study, we identified ambient TCE clustering primarily in the rust belt region of the United States. Notably, the geographic distribution of ambient TCE, which is highest in parts of the Midwest, Northeast, and California, overlaps partially with state-level PD risk. For example, higher relative risk of PD is observed in similar parts of the Midwest and Northeast—and lower relative risk seen in the Northern Great Plains and Maine. However, there are exceptions such as low TCE concentration in North Dakota despite higher PD relative risk and high TCE concentrations in California where low risk of PD is observed. This suggests that other environmental neurotoxicants beyond TCE contribute to PD-related risk, including certain pesticides previously linked to risk of PD.³⁶

Moreover, our study provides insight into the impact of TCE on clinical PD subtypes. Our positive association with gait abnormalities aligns with a previous study, which found that environmental exposure to PM_2.5 was associated with the akinetic-rigid PD type and not the tremor-predominant clinical subtype.³⁷ Similarly, TCE exposure may manifest as an axial-predominant clinical PD subtype that includes dementia.

Of the top 3 TCE-emitting facilities in the nation, we identified 2 with a relatively greater PD risk in the area surrounding the facilities. Most notably, in Lebanon, OR, the region with the greatest risk of PD was downwind (southeast) of a facility that emitted 198,327 pounds of TCE in 2002. The pattern of PD risk in Lebanon was centered around this facility and exhibited a marked decreasing risk gradient extending outward from the facility. This facility manufactures lead-acid battery separators and has been in operation since 1984. In 2008, a multiyear study conducted by the National Institute for Occupational Safety and Health (NIOSH) found evidence of adverse neurologic health outcomes in workers at this facility.³⁸ Like the NIOSH investigation in Lebanon, the EPA brought attention to TCE emissions coming from the facility in Corydon, IN.^39,40 In Corydon, we estimated a region with relatively greater risk of PD east of the facility, which, in 2002, emitted the highest levels of ambient TCE (1,098,811 pounds) in the nation. However, the pattern of high PD risk surrounding the Corydon facility was less pronounced compared with the risk pattern around the Lebanon facility, with some of the highest risks in Corydon observed approximately 8 miles northeast of the facility. The potential impact of wind direction is difficult to assess, given the variability of prevailing winds in Corydon. Moreover, the terrain to the west of the facility is more rugged than in the east. Nevertheless, we were unable to determine the potential impact of terrain and wind strength on the pattern of risk in Corydon. The region surrounding the facility in Troy, WI (which emitted 209,879 pounds of TCE in 2002), exhibited a geospatial pattern inconsistent with increased risk of PD due to ambient TCE. Notably, this facility shut down operations in 2009, which was 7–9 years before our case ascertainment, and therefore, our map of Troy represents the PD risk of populations with up to nearly a decade less TCE exposure compared with the Lebanon and Corydon facilities. In addition, during its operation, a state-level investigation suggested that there were no concerning concentrations of TCE in much of the surrounding community.⁴¹ It is possible that community knowledge regarding risk of occupational TCE exposure around the facilities could have created a screening bias for PD, including for those who worked at the facilities. However, after restricting to women, our MapGAM analysis identified patterns of high risk surrounding the facilities, which is largely inconsistent with confounding by occupational exposure for women in that age group (i.e., women from the birth cohorts in our study typically did not work in factories). Nevertheless, we acknowledge the potential for confounding by occupational exposures, including previous military-related exposures. While additional facilities other than the top 3 selected may demonstrate similar patterns of high risk, the MapGAM methods are computationally intensive when applied to large population data sets, so this study focused on the top 3 facilities with the highest TCE emissions. One strength of our facility risk map approach is that, unlike our case-control analysis, our high-resolution maps are not vulnerable to exposure misclassification from using census tract–level data because our PD risk maps do not use NATA TCE estimates. Nevertheless, it is possible that other correlated exposures may contribute to the PD risk pattern around these facilities.

Our study was limited by the geographic resolution of the exposure data available. We were unable to account for the variation of TCE within each census tract, and census tracts can vary substantially in size resulting in exposure misclassification, especially in nonmetropolitan areas where census tracts tend to be larger. We were also unable to consider unmeasured indoor TCE exposure. Nevertheless, NATA data have been shown to accurately estimate the national median levels of most hazardous air pollutants,⁴² with 1 study in Detroit, Michigan, finding that NATA data predicted the average ambient concentrations of an air toxic within 5% of that measured by the research group using passive sampler monitors.⁴³ Other studies suggest that NATA data are unable to capture high concentrations and thus underestimate air toxics in some areas.^42,44 We also acknowledge that, although we did not identify an interaction between TCE exposure and age, our data do not allow us to confirm that TCE is an important risk factor of younger onset PD. Thus, our results cannot be generalized to populations younger than 65. Future studies are needed that focus on the association of ambient TCE exposure in population-based studies in those with younger onset PD. Although we used ambient TCE in our models, in some regions, exposure through drinking water or vapor intrusion may also be important sources of exposure. Although we found a compelling risk pattern surrounding a TCE-emitting facility in Oregon in our facility-level analyses, we cannot rule out the possibility that community knowledge regarding risk of occupational TCE exposure could create a screening bias for PD. We also cannot exclude the possibility of an unidentified urban exposure that is correlated with ambient TCE exposure. Other limitations include drawbacks inherent to Medicare data such as the restriction on coverage to patients aged 65 years or older, although the TCE-PD association did not differ by age within our study population. In addition, owing to the long prodromal period of PD,⁴⁵ we focused on TCE estimates 14–16 years before incident PD diagnosis and assumed residential stability between that point and 2 years before PD diagnosis/reference. However, we were able to confirm in a previous study that most Medicare beneficiaries are generally nonmobile (reside in the same zip code) for at least 5 years before PD diagnosis/reference.²¹ With that said, we note that the true period of relevant neurotoxicant exposure for PD is unknown. Finally, the validity of our data on PD diagnosis requires beneficiaries to obtain medical care and relies on competent diagnoses and data entry by health care providers and staff. Nonetheless, our case ascertainment approach likely has a very high specificity and relatively high sensitivity.²⁰ Despite limitations, our study provides new evidence of the potential importance of ambient TCE exposure on PD risk and identifies specific areas of the nation that may be targets for exposure remediation.

Glossary

EPA

Environmental Protection Agency

GAM

generalized additive model

ICD-9-CM

International Classification of Diseases, 9th Revision, Clinical Modification

ICD-10-CM

International Classification of Diseases, 10th Revision, Clinical Modification

NATA

National Air Toxics Assessment

NIOSH

National Institute for Occupational Safety and Health

OR

odds ratio

PD

Parkinson disease

PM_2.5

particulate matter with a diameter <2.5 μm

RR

relative risk

RUCA

rural-urban commuting area

TBI

traumatic brain injury

TCE

trichloroethylene.

Author Contributions

B. Krzyzanowski: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. K.M. Beyene: analysis or interpretation of data. J.R. Turner: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. B.A. Racette: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data.

Study Funding

This study was supported by the US Department of Defense (PD190057), the Kemper and Ethel Marley Foundation, Barrow Neurological Foundation, and a gift from the Moreno Family.

Disclosure

The authors report no disclosures relevant to the manuscript. Go to Neurology.org/N for full disclosures.