Association Between Exposure to Particulate Matter and the Incidence of Parkinson’s Disease: A Nationwide Cohort Study in Taiwan
Article information
Abstract
Objective
Emerging evidence suggests that air pollution exposure may increase the risk of Parkinson’s disease (PD). We aimed to investigate the association between exposure to fine particulate matter (PM2.5) and the risk of incident PD nationwide.
Methods
We utilized data from the Taiwan National Health Insurance Research Database, which is spatiotemporally linked with air quality data from the Taiwan Environmental Protection Administration website. The study population consisted of participants who were followed from the index date (January 1, 2005) until the occurrence of PD or the end of the study period (December 31, 2017). Participants who were diagnosed with PD before the index date were excluded. To evaluate the association between exposure to PM2.5 and incident PD risk, we employed Cox regression to estimate the hazard ratio and 95% confidence interval (CI).
Results
A total of 454,583 participants were included, with a mean (standard deviation) age of 63.1 (9.9) years and a male proportion of 50%. Over a mean follow-up period of 11.1 (3.6) years, 4% of the participants (n = 18,862) developed PD. We observed a significant positive association between PM2.5 exposure and the risk of PD, with a hazard ratio of 1.22 (95% CI, 1.20–1.23) per interquartile range increase in exposure (10.17 μg/m3) when adjusting for both SO2 and NO2.
Conclusion
We provide further evidence of an association between PM2.5 exposure and the risk of PD. These findings underscore the urgent need for public health policies aimed at reducing ambient air pollution and its potential impact on PD.
INTRODUCTION
Air pollution is a complex mixture of gases, particulate matter (PM), organic compounds, and metals that can detrimentally impact human health. Exposure to air pollution, especially fine particulate matter (PM2.5), has been associated with an increased risk of neurological disorders, such as Alzheimer’s disease, stroke, and dementia, as well as accelerated progression of neurodegenerative diseases [1-4]. PM2.5 is of particular concern because it can enter the brain through the nasal olfactory mucosa or systemic circulation, triggering neuroinflammatory responses [2,5-7].
Parkinson’s disease (PD) is a common neurodegenerative disorder characterized by progressive degeneration of dopamine-generating neurons in the substantia nigra [8]. The pathogenesis of PD may involve environmental and genetic factors related to oxidative stress, excitotoxicity, and mitochondrial defects [8]. Exposure to air pollution has been linked to neurotoxic effects in the brain, including blood‒brain barrier breakdown, microglial activation, oxidative stress, neuroinflammation, and hypothalamic–pituitary–adrenal axis dysregulation [5,9-11]. Evidence from numerous experimental studies has indicated that PM exposure can contribute to the pathogenesis of PD by disrupting proteostasis, injuring mitochondria, and inducing inflammation [10,11]. Some epidemiological studies have reported a positive association between PM2.5 exposure and PD incidence; however, there are inconsistencies across studies, possibly due to differences in methodological approaches, cohort characteristics, sample size, and follow-up duration [11-27].
To address these issues, we conducted a large-scale nationwide epidemiological investigation with a long follow-up duration to examine the potential association between long-term exposure to PM2.5 and the risk of incident PD. We carefully selected air pollutant covariates to prevent collinearity and controlled for potential confounding factors. Our study aimed to test the hypothesis that exposure to PM2.5 increases the risk of incident PD.
The underlying rationale focused on PM2.5 as the primary air pollutant of interest, in contrast to other air pollutants. This decision was grounded in the growing body of evidence that highlights the particularly insidious role of PM2.5 in contributing to various adverse health outcomes, especially neurological disorders such as PD. Unlike larger particulates, PM2.5 can penetrate deeply into the respiratory system and even enter the bloodstream, posing significant risks to human health. Moreover, the ability of PM2.5 to carry a variety of toxic substances due to its small size and large surface area further exacerbates its potential neurotoxic effects. The selection of PM2.5 over other pollutants was also informed by epidemiological studies that have increasingly linked PM2.5 exposure to cognitive decline, neuroinflammation, and the acceleration of neurodegenerative processes. By concentrating on PM2.5, our study aligns with current research priorities, aiming to elucidate the mechanisms through which PM2.5 contributes to the incidence of PD, thereby offering valuable insights into public health strategies aimed at mitigating exposure and reducing the burden of neurological diseases.
MATERIALS & METHODS
National Health Insurance Database
The National Health Insurance Research Database (NHIRD) is a nationally representative cohort database that captures the complete medical records of 99.99% of Taiwan’s population through the National Health Insurance Program. The NHIRD is managed by the Health and Welfare Data Center of Taiwan’s Ministry of Health and Welfare, which facilitates applications for the usage of data in research. For this study, we extracted longitudinal beneficiary registries and medical records from outpatient and inpatient visits for a randomly selected sample of two million insured beneficiaries from the NHIRD. Each subject in the database is assigned an encrypted personal identifier that enables linkage of their historical medical records. Disease diagnoses before 2016 were classified according to the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM), and those after 2016 were classified according to the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM). Further details about the NHIRD are available elsewhere [4,28].
Air quality data
This population-based national cohort study included individual health data for each participant in the random sample. To estimate the daily ambient concentrations of air pollutants in all cities, districts, or townships in Taiwan, we used a kriging method with an external trend. Each participant’s monthly PM2.5 exposure was based on the city, district, or township of residence or occupation type (for the working population) reported in the NHIRD.
Air quality data are available from the Taiwan Environmental Protection Administration, which maintains 83 air monitoring stations throughout Taiwan. These data include measurements of temperature (°C), relative humidity (%), and concentrations of various air pollutants, including SO2 (ppb), CO (ppm), O3 (ppm), NOx (ppb), NO (ppb), NO2 (ppb), PM10 (μg/m3), and PM2.5 (μg/m3).
Spatiotemporal estimation of air pollutant concentrations
Spatiotemporal techniques have been widely used in ambient pollutant concentration estimation across space and time [29]. Stochastic approaches generally consider the distribution of ambient pollutants within the framework of random field theory, which allows for the formation of a multivariate joint distribution among the relevant attributes. One popular stochastic method for estimating air quality in space and time is the kriging method, which is often referred to as the best unbiased linear estimator because it has minimum estimation uncertainty. Most stochastic methods assume homogeneity and stationarity, and in modeling spatiotemporal processes, a deterministic spatiotemporal mean trend is typically decomposed from a stochastic residual component [30].
In this study, we utilized a simple kriging method with an external trend to estimate the daily ambient concentrations of air pollutants in all townships across Taiwan. To determine the deterministic temporal trends of pollutants at each monitoring station, we employed a local weighted regression method that smoothed the temporal variation by fitting a series of locally loworder polynomial functions to the scattered data [31]. To ensure the normality of the space-time residuals prior to applying the kriging method, we also performed a normal score transformation on the residual components of the space-time data. The spatiotemporal estimation of ambient pollutant concentrations was conducted at the center of each township within the study area.
Due to the lack of PM2.5 data prior to 2005, we obtained air quality data for all years beginning in 2005 and aggregated the spatiotemporally estimated data into monthly averages. The beneficiary registry dataset contains monthly records detailing demographic information for each insured individual, including their birth year, sex, residential county, township, and insured status. Consequently, the residential county and township for each subject were reviewed on a monthly basis and linked to corresponding air quality data based on the township and month of the year. These data were then integrated into average yearly values for each subject during the follow-up period. We drew a map of the spatiotemporal distribution of average PM2.5 concentrations in Taiwan in a recently published paper [4].
In our study, air pollution exposure estimation was conducted utilizing the Bayesian Maximum Entropy (BME) method. This method was chosen for its robustness in integrating diverse data sources and effectively managing the inherent uncertainties within environmental datasets. The BME model was applied at the township level with an hourly temporal resolution, from which annual and daily mean values were subsequently derived through averaging. This advanced modeling technique offers significant improvements over conventional methods such as remote sensing or land-use regression models, particularly in terms of accommodating spatial and temporal variability and reducing reliance on extensive ground-truth data. The application of the BME method in our analysis ensured a highly accurate and nuanced estimation of air pollution exposure, aligning with the rigorous demands of epidemiological research in accurately assessing environmental risk factors.
Study subjects
Our study cohort consisted of 454,583 insured individuals aged 50 years and older as of January 1, 2005, who were followed up until PD diagnosis, insurance termination, or December 31, 2017. Subjects who were previously diagnosed with PD or missing air quality data were excluded from the analysis (Figure 1). In our study, we included individuals with a minimum of two outpatient visits or those admitted with a diagnosis of PD (ICD-9-CM code: 3322; ICD-10-CM code: G20) from January 2005 to December 2017. Comorbidities such as hypertension, diabetes, hyperlipidemia, chronic obstructive pulmonary disease, and lung cancer were assessed as potential confounding factors. The Charlson Comorbidity Index (CCI) score was calculated using hospital admission records both before and during the study period.
Study enrollment flowchart.
Statistical analysis
Baseline characteristics are described as means (standard deviations [SDs]) and frequencies (%). We also calculated Pearson correlation coefficients for two-variable combinations of temperature, relative humidity, and air pollutant concentrations to avoid collinearity. Only variables that had lower correlations with each other (Pearson’s correlation coefficients < 0.7) were selected for further analysis. Age, air quality and the CCI score were considered time-dependent covariates in this study. Therefore, we used extended Cox models based on the Andersen‒Gill counting process to measure the association between exposure to air pollutants and the risk of developing PD. We created yearly records for each subject, consisting of baseline characteristics, annual concentrations of air pollutants per interquartile range (IQR), annual temperature and relative humidity from the index date to the end of follow-up. At most, a subject could had 13 yearly records. The beneficiary registry dataset contains monthly records including demographic data for each insured individual, such as birth year, sex, residential county and township, and insured status. We linked air quality data by the corresponding township and month of the year for each subject and aggregated the monthly air quality data into yearly data. The CCI score of each subject during the corresponding year was also calculated. All the statistical analyses were performed using R software, version 4.1.0 (R Core Team, 2021; https://www.R-project.org/) and SAS software, version 9.4 (SAS Institute Inc., Cary, NC, USA).
RESULTS
This study included 454,583 subjects aged over 50 years with a mean follow-up period of 11.1 ± 3.6 years. The mean age at baseline was 63.1 years, and 50% of the subjects were male. Hypertension, diabetes, and hyperlipidemia were present in 45.4%, 20.9%, and 26.7% of the subjects, respectively. The majority (81.3%) of the subjects had a CCI score of 0 (Table 1).
Baseline characteristics of the entire longitudinal cohort
We checked for collinearity among the individual environmental determinants by calculating Pearson correlation coefficients (Table 2). The results showed that some air pollutants were highly correlated with each other. The correlations among the annual exposures were high between PM and other air pollutants (PM2.5−PM10, CO−NO, CO−NO2, CO−NOx, NO−NO2, NO−NOx, NO2−NOx, NO2−O3, and NOx−O3), with correlation coefficients ranging from 0.72 to 0.96. Temperature, relative humidity, PM2.5, SO2, and NO2 had lower correlations with each other (correlation coefficients < 0.7) and were thus included in the subsequent Cox regression analysis.
Collinearity between ambient air pollutants tested using pearson correlation matrix
Table 3 shows the yearly average air pollution exposure levels among individuals in the study population, including the mean, SD, median, range, and IQR over the entire follow-up period, with a maximum duration of 13 years. Figure 2 shows the concentrations of PM2.5, SO2, and NO2 between 2005 and 2017.
Participants’ mean exposure levels to air pollutants during the 13 years before the event or censoring
Trends of the concentrations of PM2.5, SO2, and NO2 between 2005 and 2017 in Taiwan.
In this study, 4% (18,862) of the subjects developed PD during the mean follow-up of 6.2 years. Table 4 shows the association between PM2.5 exposure and the risk of PD after adjusting for relevant demographic, medical, and environmental covariates. After adjusting for demographic, medical, and environmental covariates, the results showed a significant association between PM2.5 exposure and the risk of PD. The hazard ratio (HR) for PD was 1.15 (95% CI, 1.13–1.16) per IQR increase in PM2.5 exposure. Adding SO2 or NO2 exposure to the model increased the HR to 1.23 and 1.13 per IQR increase in PM2.5 exposure, respectively. When both SO2 and NO2 exposure were added, the HR for PD was 1.22 per IQR increase in PM2.5 exposure. The HR for PD was greater in the male group than in the female group. These findings suggest a positive association between PM2.5 exposure and the risk of PD, which may be further increased by coexposure to SO2 and NO2.
Our Cox regression analysis, as depicted in Table 4, investigated the association between PM2.5 exposure (alone or in combination with SO2 and/or NO2) and the risk of developing PD over a 13-year follow-up period, with a particular focus on sex differences. In males, PD risk was consistently associated with increased exposure to air pollutants across these models. For instance, exposure to PM2.5 alone significantly increased the risk in the fully adjusted Model 2 (HR: 1.14; 95% CI: 1.12, 1.16), while coexposure to PM2.5, SO2, and NO2 additively increased PD risk. On the other hand, the analysis of females revealed a notably lower risk, and exposure to PM2.5 alone or in combination with SO2 and/or NO2 did not contribute to differing risks, suggesting a less pronounced effect of air pollution exposure on PD risk in women.
Figure 3 presents the forest plot for the results of Model 2 from Table 4, highlighting the most important findings of our study.
Hazard ratios (95% confidence intervals) for the association between PM2.5 exposure and the risk of PD during the 13-year follow- up period. The model was adjusted for age, sex, occupation type, hypertension status, diabetes mellitus status, hyperlipidemia status, chronic obstructive pulmonary disease status, lung cancer status, Charlson Comorbidity Index score, temperature, relative humidity, and air pollutant concentrations in the corresponding year. HR, hazard ratio; PD, Parkinson’s disease.
DISCUSSION
This study investigated the link between air pollution exposure and PD incidence in Taiwan. After a follow-up period of up to 13 years and controlling for relevant factors, the results showed a significant association between PM2.5 exposure and a 13%–24% increased risk of PD by per IQR increase. This association persisted after adjusting for exposure to SO2 and/or NO2. Overall, the findings suggest that long-term exposure to PM2.5 may contribute to the incidence of PD.
The burden of PD has been increasing worldwide, potentially due to the aging of the population, a longer PD duration, and changes in environmental and social risk factors [32-34]. Our study revealed a generally greater incidence of PD than earlier studies conducted in Taiwan, South Korea, Australia, Canada, the United States of America, and Italy [14,15,17-19,35-37]. However, comparisons between studies are difficult due to differences in data sources, cohort sizes, PD diagnostic criteria, follow-up durations, and statistical analyses.
Various studies have investigated the relationship between long-term exposure to air pollutants and PD risk, but the results are inconsistent. Some studies reported a positive association of PD risk with PM2.5, PM10, O3, NO2, NOx, CO, and SO2 exposure, while others reported a negative or no association with these pollutants [14,17,18,22,36,38-40]. These discrepancies may be due to differences in study design, population characteristics, pollutant compositions, and exposure measurements. Our nationwide study included a larger cohort with a longer follow-up period than did previous studies and performed collinearity analysis to avoid overestimating the association of air pollution exposure with PD risk, as many pollutants originate from the same sources and have temporal correlations [13,21].
The sex-specific analysis within our Cox regression framework revealed insightful disparities in the risk of PD attributable to air pollution exposure. The increased HRs observed among males suggest a heightened vulnerability to the neurotoxic effects of pollutants such as PM2.5, SO2, and NO2. This discrepancy may highlight underlying biological differences in susceptibility to environmental toxins or differences in exposure levels between the sexes. The relatively low HRs among females imply either a reduced sensitivity to these pollutants or potential sex-related protective factors that mitigate the risk of PD. These findings necessitate a deeper exploration into the mechanisms driving these sex-specific responses to air pollution exposure. Furthermore, our results underscore the importance of incorporating sex as a key variable in environmental health research, ensuring that public health policies and interventions are tailored to effectively protect the most susceptible populations from the adverse effects of air pollution on neurological health.
PM2.5 is a small particle that can enter the lungs and harm human health. It is associated with detrimental effects on the respiratory, cardiovascular, and central nervous systems. PM is a complex mixture of organic and inorganic particles that may be more important than the total amount. Among a variety of air pollutants, PM is an important contributor to neurological diseases, particularly PD. Animal studies suggest that PM can reach the brain and induce widespread neuroinflammation and neurotoxic reactions that affect multiple brain regions. Exposure to ambient PM causes a loss of dopaminergic neurons, a hallmark of PD. Human autopsy studies have shown increased neuroinflammation and oxidative stress indices; increased biomarkers of accelerated brain aging; amyloid beta, hyperphosphorylated tau protein, and α-synuclein accumulation; olfactory bulb inflammation; and increased metal concentrations [41-44]. Our study supports the notion that PM2.5 could influence the pathogenesis of PD through direct and systemic neurotoxicity after entering the central nervous system, either via the systemic circulation by crossing the blood‒brain barrier or through nasal passages by impacting the olfactory mucosa.
PD has a long prodromal period, and our study covered the preclinical stage of PD. Medical comorbidities during this period may make confirming the link between air pollution exposure and PD risk challenging. Our findings suggest that sustained exposure to PM2.5 is a risk factor during the prodromal stage of PD, and increased recognition of this risk is needed. Minimizing population-level exposure could lead to neurological benefits and cost savings in health care.
It is pertinent to acknowledge that while our study extensively utilized the BME method for spatiotemporal air quality modeling, we did not explicitly detail the model’s accuracy or sensitivity metrics within our report. The BME method, renowned for its advanced statistical framework, allows for refined environmental exposure estimations by addressing spatial heterogeneity and temporal variability. This approach significantly mitigates uncertainty in environmental data, a crucial aspect of epidemiological research. Although the effectiveness of the BME method is well supported in the literature, particularly in comparison to traditional models, the absence of specific accuracy and sensitivity values in our analysis may limit the interpretability of our findings to some extent. Future work could benefit from directly incorporating these metrics, offering a clearer understanding of the model’s performance and further validating the reliability of the exposure assessments derived from this sophisticated modeling approach.
An intrinsic limitation of our study is the absence of a lag analysis to meticulously explore the effects of PM2.5 exposure during the premotor phase of PD. This phase, characterized by nonmotor symptoms that precede clinical diagnosis, can last several years, making the timing of exposure critical for understanding disease etiology. The lack of longitudinal depth and precise timing of exposure in our dataset made it impossible to conduct a lag analysis to accurately model these effects. Consequently, our findings may not fully capture the delayed impact of PM2.5 exposure on the risk of developing PD, highlighting the need for future research involving detailed longitudinal data to address this gap. Other limitations include the lack of data on certain factors associated with PD, potential underdiagnosis, limited information on PD stage and progression, incomplete data on lifetime exposure to air pollution, and the possibility of commuting between different areas. However, our large-scale, nationally representative study provides valuable epidemiological evidence for environmental policy-making. Future investigations should consider the association between nocturnal air pollution exposure and PD risk while adjusting for diurnal variation [45,46].
In summary, our study indicated that long-term PM2.5 exposure is linked to a greater risk of developing PD, with a 1.13- to 1.24-fold increased risk after considering various factors. The neurotoxic effects of PM2.5 exposure, which is a modifiable risk factor, could contribute to the development of PD. Further research is needed to determine the demographic and genetic factors that may increase susceptibility to PM2.5-related neurotoxicity and PD pathogenesis.
Notes
Conflicts of Interest
The authors have no financial conflicts of interest.
Funding Statement
The study received support from HungKuang University and Kuang Tien General Hospital (HK-KTOH-109-05). The funders did not participate in the study design, analysis, write-up, or decision to submit for publication.
Author Contributions
Conceptualization: Chih-Sung Liang, Chun-Pai Yang. Data curation: Ting- Bin Chen, Ching-Mao Chang, Cheng-Chia Yang, Chun-Pai Yang. Formal analysis: Cheng-Chia Yang, Yuh-Shen Wu, Winn-Jung Huang, Yuan-Horng Yan, Hwa-Lung Yu, I-Ju Tsai, Chih-Sung Liang, Chun-Pai Yang. Writing—original draft: Ting-Bin Chen, Ching-Mao Chang, Cheng-Chia Yang. Writing—review & editing: Ting-Bin Chen, Cheng-Chia Yang, Hwa-Lung Yu, I-Ju Tsai, Chih-Sung Liang, Cheng-Yu Wei, Chun-Pai Yang.
Acknowledgements
We would like to express our gratitude to the Health and Welfare Data Science Center, China Medical University Hospital for providing administrative and technical support.