The association of long-term air pollution with a composite neurodegenerative risk score among Chinese elderly: evidence from the CHARLS

Abstract

Background

The relationship between long-term exposure to air pollution and the risk of neurodegenerative diseases has drawn increasing attention. However, the relative effects of various pollutants and the potential role of systemic inflammation remain unclear. This study aims to explore the association between various air pollutants and the comprehensive risk score of neurodegenerative diseases in the middle-aged and elderly population in China, and to examine the effect modification of baseline high-sensitivity C-reactive protein (hs-CRP) and white blood cell count (WBC).

Methods

The data were sourced from the China Longitudinal Study of Health and Retirement (CHARLS) and the National Earth System Science Data Center. The average concentrations of fine particulate matter (PM_2.5), inhalable particulate matter (PM10), nitrogen dioxide (NO₂), and ozone (O₃ ) from 2012 to 2015 were obtained through processing by the ArcGIS platform. The outcome was the comprehensive risk score and its ternary scale constructed by grip strength, the Central Depression Scale (CES-D-10), the Mini-Mental State Examination (MMSE), and instrumental activities of daily living (IADL) at the follow-up in 2015. A two-stage analysis was adopted: Firstly, multiple linear regression (OLS) was used to analyze the association between pollutants and continuous scores; Then, ordered Logistic regression was applied. After adjusting for the baseline score in 2011, age, gender, education, urban and rural residence, and the number of chronic diseases, the association between pollutants and risk levels was analyzed, and age stratification analysis of pollutants was conducted (30-59 years old, 60-74 years old, ≥75 years old). The effect modification effect was examined by incorporating the interaction term between pollutants and baseline inflammatory markers into the model.

Results

OLS regression analysis showed that long-term PM2.5 exposure was significantly associated with an elevated comprehensive neurodegenerative risk score (adjusted β = 0.010, 95% CI: 0.008-0.016), while the associations with PM10, NO2, and O3 were not significant. Ordered Logistic regression analysis showed that, after adjusting for covariates, only PM2.5 was significantly associated with a higher neurodegenerative risk score grade (OR= 1.52, 95% CI: 1.01-2.31), and this association was specific to the 60-74 age group. Baseline hs-CRP and WBC levels had no significant effect modification on the association between pollutants and the comprehensive risk score of neurodegeneration (P > 0.05).

Conclusion

In the study population, long-term PM2.5 exposure was associated with an elevated comprehensive neurodegenerative risk score, especially among the elderly aged 60-74. No significant modifying effect of baseline inflammatory levels on this association was found. The research results suggest that air pollution control and targeted intervention for the elderly population have potential public health significance.

Introduction

Neurodegenerative diseases, such as Alzheimer’s disease, Parkinson’s disease, and other forms of dementia, remain a significant clinical challenge due to the lack of effective treatments and the difficulty in achieving early and accurate diagnosis [1]. According to the 2021 Global Burden of Disease (GBD) report [2], the prevalence of neurodegenerative diseases is continuously increasing worldwide, presenting a major public health challenge 0. These diseases not only impose a heavy caregiving burden on patients and their families but also place substantial pressure on healthcare systems. Alzheimer’s disease (AD), the most common neurodegenerative disease, is of particular concern [3]. A report based on Chinese data shows that among a sample of 626,276 elderly people, the overall prevalence rate of Alzheimer’s disease is 3.48%, with an incidence rate of 7.90 cases per 1,000 person-years [4], posing a severe threat to population health [5]. Against this backdrop, constructing a comprehensive risk score for neurodegeneration that integrates multi-dimensional information such as motor, emotional, cognitive and daily functional aspects holds significant preventive medical value for quantifying individual risks, identifying susceptible populations and exploring modifiable environmental risk factors in the preclinical stage.

Ambient air pollution is a leading global cause of mortality and is closely associated with various adverse health outcomes, including cardiovascular diseases [6], metabolic syndrome [7], neurodegenerative diseases [8], physical frailty [9], depression [10–12], and systemic inflammation [13–17]. Mechanistic studies further indicate that exposure to air pollution can elevate systemic inflammation levels [18–22], which in turn may influence the onset of depression [23, 24], cardiovascular health [5], and cardiopulmonary function [25]. Late-life depression (LLD), a common condition, frequently co-occurs with age-related neurodegenerative diseases [26]. Therefore, it is reasonable to hypothesize that air pollution may influence the risk of neurodegenerative diseases through the pathway of systemic inflammation. While emerging research has begun to explore the association between systemic inflammation and neurodegenerative diseases, suggesting that reducing neuroinflammation or systemic inflammation might help slow or even reverse neurodegenerative processes [27–31], evidence specifically investigating the role of systemic inflammation in the link between air pollution and neurodegeneration remains limited. Our study aims to address this gap.

However, there is still insufficient research on the combined impact of air pollution and neurodegenerative risks at present, which restricts its scientific value and public health implications. Existing causal analyses are relatively limited, with most studies employing cross-sectional designs, which struggle to adequately capture the long-term effects and dynamic changes of air pollution [9, 23], and there is a lack of robust longitudinal evidence. Furthermore, although numerous studies have linked air pollution to adverse health outcomes, most have focused on fine particulate matter (PM_2.5) [27, 32], while paying relatively less attention to other pollutants like NO₂ and O₃. This limits our in-depth understanding of the independent and interactive effects of different air pollutants.Concurrently, factors that may buffer or exacerbate the effect of air pollution on neurodegeneration have not been systematically explored. These include gender, history of chronic diseases, educational level, marital status, and physical activity participation, all of which can lead to variations in health outcomes. Notably, the prevalence, risk factors, and clinical manifestations of neurodegenerative diseases may differ significantly among different adult age groups. Particularly, “young-old” adults (aged 60–74) and “old-old” adults (over 75) differ markedly in terms of physiological reserve, comorbidity profiles, and treatment responses [33, 34]. However, most existing studies analyze the population aged 60 and above as a single group, which may mask heterogeneity within this broad category. In the exploration of the mechanism of systemic inflammation, existing studies often fail to clearly distinguish the different statistical and biological concepts of “mediators” and “effector modifications”, and the assessment of inflammation mostly relies on a single biomarker.

To advance the existing research, this study is based on the national representative prospective cohort of the China Health and Elderly Care Follow-up Survey, aiming to achieve the following goals: (1) To assess the longitudinal association between long-term (2012–2015) exposure to four air pollutants, namely PM_2.5, PM₁₀,NO₂ ,and O₃, and the comprehensive risk score of neurodegeneration in the middle-aged and elderly population; (2) Through age-stratified analysis, explore the potential modifying effect of age on the above-mentioned association; (3) To examine whether the baseline systemic inflammatory status (hs-CRP and WBC) is an effect modifier of the association between air pollution and the comprehensive risk score of neurodegeneration, that is, to explore whether the level of inflammation will change the intensity of this association.

This study aims to provide high-level evidence for understanding the heterogeneity of population damage to brain health caused by air pollution and to offer a scientific basis for formulating precise prevention strategies for high-risk subgroups by simultaneously examining multiple pollutants, adopting a longitudinal design, conducting detailed subgroup analyses, and rigor testing the effect modification of inflammation.

Methods

Study population

The study participants were drawn from the China Health and Retirement Longitudinal Study (CHARLS), a nationwide cohort survey. The CHARLS survey covers approximately 150 county-level units across 28 provinces in China and is designed to collect extensive, high-quality micro-level data on middle-aged and older adults. Following the baseline assessment in 2011, follow-up surveys were conducted in 2015 and 2018. The national baseline survey was carried out from June 2011 to March 2012 and involved 17,705 respondents. Blood samples were collected from CHARLS participants once every two follow-up waves [1]. Relevant blood biomarker data in the public CHARLS database are only available for the years 2011 and 2015.

Therefore, the study population for this analysis consisted of individuals who provided blood samples in 2011 and participated in both the blood collection and follow-up in 2015. Initially, 23,952 middle-aged and older adults who participated in the blood draws across both waves were identified. Because the ID coding system used in 2015 differed from that in 2011, the 2011 IDs were recoded to enable matching with the 2015 IDs. Individuals meeting one or more of the following criteria were excluded: (1) lack of community-level information; (2) missing data on cognitive function scores, depression scores, grip strength levels, instrumental activities of daily living scores, hs-CRP levels, WBC levels, or key demographic and covariate information; (3) loss to follow-up. Consequently, a final analytical sample of 6,180 middle-aged and older adults from 28 provinces in China was recruited for this study. The participant inclusion flowchart is shown in Fig. 1. Informed consent was obtained from all participants, and the CHARLS survey received approval from the Peking University Institutional Review Board (IRB Approval Code: IRB00001052-11015).

Fig. 1.

Participant Inclusion Flowchart

Consistent with the World Health Organization’s definition of older populations and commonly used age stratification criteria in clinical research [35, 36], participants in this study were categorized into three age groups: 30–59 years (middle-aged), 60–74 years (young-old), and 75 years and above. This stratification was implemented to investigate potential differences in the impact of air pollution across distinct demographic segments characterized by varying physiological stages, risks of multimorbidity, and health needs.

Neurodegenerative composite risk score

To construct a comprehensive and effective indicator for identifying early-stage neurodegenerative risk, this study developed a composite risk score. This score was calculated by integrating indicators from four core functional domains using an equal-weight averaging method: motor function (assessed by maximum grip strength), affective function (evaluated by the 10-item version of the Center for Epidemiologic Studies Depression Scale), cognitive function (measured by the Mini-Mental State Examination), and activities of daily living function (assessed by Instrumental Activities of Daily Living). All measures were obtained using standardized scales from the China Health and Retirement Longitudinal Study (CHARLS) project.

Maximum grip strength serves as a representative measure of motor function and a reliable proxy indicator of skeletal muscle mass. Substantial evidence demonstrates that sarcopenia, characterized by reduced muscle mass and strength, is closely associated with cognitive decline and dementia risk in epidemiological studies [37–39], reflecting the interrelationship between physical and brain health. Emotional function was evaluated using the Center for Epidemiologic Studies Depression Scale (CESD-10), as emotional disorders are common progenitor symptoms of neurodegenerative diseases [40], and the neuroendocrine and inflammatory dysregulation they represent themselves may also be important factors promoting the pathological process [41, 42]. The Mini-Mental State Examination (MMSE), as a global cognitive screening tool for dementia, is valuable for its sensitivity in detecting early decline across multiple cognitive domains (e.g., memory, orientation, executive function) associated with dementia [43]. The Instrumental Activities of Daily Living (IADL) score assesses instrumental activities of daily living [44], such as managing finances, shopping, and medication. Impairment in these higher-order functions is a critical indicator of the transition from mild cognitive impairment to clinical dementia, demonstrating the real-world functional impact of cognitive deficits. Integrating these four domains enables the construction of a multidimensional risk profile spanning physical, affective, cognitive, and functional dimensions, providing predictive validity superior to any single indicator alone.

In terms of data sources, all indicators in this study were derived from the CHARLS database, a nationally representative, high-quality data resource. CHARLS employed rigorously validated and internationally recognized instruments to measure the aforementioned functions, ensuring the scientific validity and comparability of the data. Specifically, maximum grip strength was measured using a standardized handheld dynamometer, with the highest value from multiple attempts of the dominant hand recorded, ensuring a unified and reliable method. The depression assessment utilized the 10-item version of the Center for Epidemiologic Studies Depression Scale (CES-D-10), which has demonstrated good reliability and validity in the Chinese elderly population for assessing the frequency of depressive symptoms over the past week [45]. Cognitive function was assessed using the CHARLS-adapted version of the Mini-Mental State Examination (MMSE), which has been culturally adapted for the Chinese context and covers core cognitive domains such as time orientation, place orientation, memory, and calculation [46]. The assessment of Instrumental Activities of Daily Living (IADL) was based on the widely used Lawton scale [47], inquiring whether respondents required assistance with activities such as cooking, managing finances, and taking medications. These standardized, high-quality measurements provide a solid data foundation for constructing the composite risk score in this study, enabling it to not only capture subtle functional changes but also ensure the generalizability of the findings to broader populations.

In terms of calculating the comprehensive risk score, we adopt the arithmetic mean method of Z-scores. Firstly, to ensure the consistency of the risk direction for all indicators (that is, an increase in the value indicates an increase in risk), the relevant scores were reversed in direction, transforming them from “capability indicators” to “damage indicators”. Subsequently, the four directionally aligned variables were standardized using Z-score normalization, converting them to a comparable scale with a mean of 0 and a standard deviation of 1. Finally, the standardized Z-scores were averaged arithmetically to generate a continuous composite risk score. This score intuitively reflects an individual’s deviation from the population average risk: a score of 0 represents average risk, positive values indicate above-average risk, and negative values denote below-average risk. This construction method ensures transparency and comparability while avoiding subjective weighting, comprehensively capturing the multidimensional nature of neurodegenerative risk.The composite risk score demonstrated a Cronbach’s alpha of 0.6, indicating that neurodegenerative risk is a composite concept comprising relatively independent domains. Although the internal consistency is moderate, all theoretically important domains were included to holistically capture multiple dimensions of risk. The composite score provides an overarching measure of overall risk.

Air pollution exposure

Ground-level air pollution concentrations for the years 2011 to 2015 were utilized. The study selected the following datasets: the China 1KM High-Resolution High-Quality Yearly PM_2.5 and PM₁₀ datasets, the China 10KM High-Resolution High-Quality Yearly Near-Surface Nitrogen Dioxide (NO₂) dataset, and the China 1KM High-Resolution High-Quality Yearly Near-Surface Ozone (O₃) dataset. These data were obtained from the National Earth System Science Data Center (link). The primary coverage of the datasets encompasses the entire region of China, with a spatial resolution of 1 km (except for NO₂ at 10 km), a temporal resolution of yearly, and units of µg/m³.

The PM_2.5 dataset was generated by developing a four-dimensional spatiotemporal deep forest model. Using the seamless daily 1-km PM_2.5 dataset from the CHAP (China High Air Pollution) project as a constraint, this model integrated data from a high-density ground-based observation network for PM_2.5 chemical components, satellite remote sensing products, meteorological reanalysis data, and model simulations. It established conversion models to relate different chemical components to total PM_2.5 concentration, successfully producing, for the first time, a daily seamless 1-km resolution satellite remote sensing product of four major inorganic PM_2.5 components for China. These components include sulfate (SO₄²⁻), nitrate (NO₃⁻), ammonium (NH₄⁺), and chloride (Cl⁻) [48].

The PM₁₀ dataset was generated using a spatiotemporal extreme random tree model. This model utilized supplementary data to fill spatial gaps in the satellite-based MODIS MAIAC AOD product, integrating big data sources such as ground-based observations, atmospheric reanalysis, and emission inventories to produce a seamless national ground-level PM₁₀ dataset [49].

The NO₂ dataset was developed using artificial intelligence techniques that account for the spatiotemporal heterogeneity of air pollution. An AI algorithm was employed to derive a seamless national ground-level NO₂ dataset by learning from big data, including ground-based observations, satellite remote sensing products, atmospheric reanalysis, and model simulation data [50].

The O₃ dataset was generated using an artificial intelligence model that primarily utilizes solar radiation intensity and air temperature as key predictors. This model integrated big data sources, including ground-based observations, atmospheric reanalysis, and emission inventories, to produce a seamless national dataset of the maximum 8-hour moving average ground-level O₃ concentrations [51].

The five-year concentration distributions of the four pollutants across China are illustrated in Fig. 2.

Fig. 2.

Distribution of Urban Exposure Levels for Four Air Pollutants in China (2011–2015)

Systemic inflammation

Serum high-sensitivity C-reactive protein (hs-CRP) and white blood cell (WBC) count were measured as inflammatory biomarkers. Blood sampling and laboratory procedures followed the guidelines outlined in the CDC Manual for Phlebotomy and Sample Processing, with participants instructed to fast overnight. WBC count, as part of a complete blood cell analysis, was measured using an automated analyzer within 141 min of sample collection. Hs-CRP levels were determined via an immunoturbidimetric assay, demonstrating intra-assay and inter-assay coefficients of variation of 1.3% and 5.7%, respectively. All quality control sample results fell within the predefined target ranges (within two standard deviations of the mean quality control concentration) [51]. All analytical values were derived from a multiply imputed dataset. WBC outliers (values < 2 or > 20 × 10⁹/L) were excluded from analysis, and hs-CRP values were log-transformed to approximate a normal distribution [8].

Covariates

Sociodemographic information, lifestyle behaviors, and health status data were collected using standardized self-administered questionnaires from the CHARLS database. This study adjusted for a range of potential confounding factors, including: (1) sociodemographic characteristics: age (categorized as middle-aged/young-old/oldest-old), sex (male/female), place of birth (urban/rural), educational attainment (junior high school and below/senior high school and above), and marital status (married or partnered/single or other); (2) lifestyle behaviors: sleep duration (continuous) and social activity engagement (such as visiting friends, playing mahjong, or exercising; the score reflected the number of different types of activities participated in); and (3) health-related variables: the number of chronic disease types (based on medical history, including hypertension, dyslipidemia, diabetes, cancer, chronic lung disease, liver disease, chronic heart disease, stroke, chronic kidney disease, gastric disease, emotional problems, memory-related diseases, rheumatism, or asthma). A score of 0 indicated no chronic diseases, 1 indicated one chronic disease, 2 indicated two chronic diseases, and 3 indicated three or more chronic diseases.

Statistical analysis

This study adheres to the standard practice of assessing long-term exposure effects in environmental epidemiology, aiming to reduce random fluctuations and measurement errors in a single year through multi-year average exposure assessment. To achieve this goal, we first based on the national pollutant monitoring data in the ArcGIS 10.8 (Environmental Systems Research Institute, Redlands, CA, USA) platform, The concentrations of air pollutants (PM_2.5, PM₁₀, NO₂, O₃) in 2012, 2013, 2014 and 2015 were respectively processed by spatial interpolation and meshing to generate high-resolution concentration distribution raster layers for each year. Subsequently, the raster data of each year is aggregated to the city scale, and the average concentration of each city for each year is calculated. On this basis, the arithmetic mean of pollutant concentrations in each participant’s city during the four-year period from 2012 to 2015 was calculated as an estimated indicator of their long-term exposure. This indicator is considered to be more stable in reflecting the contamination level an individual has experienced during the follow-up period than the exposure value of a single year, and it is a key step in the retrospective longitudinal exposure assessment at the core of this study.Pollutant concentrations were introduced as continuous variables in all regression models, and the scaling scale was uniformly set at “every increase of 10 µg/m³”. This definition applies to all relevant multiple linear regression (OLS) and ordered logistic regression models in the entire text.

After the above processing, the exposure data were precisely matched and integrated with the individual baseline (2011) covariate and follow-up (2015) outcome data of the CHARLS cohort in the R 4.2.1 (Lucent Technologies, USA) environment. After the matching was completed, the dataset was imported into SPSS 26.0 for basic data management, cleaning and descriptive statistical analysis. Ultimately, all multiple regression models involving correlation estimation (including linear regression and ordered Logistic regression) were completed in Stata 14.0 (StataCorp LLC, College Station, Texas, USA) to take advantage of its robust statistical modeling and standard error correction capabilities. The main analysis of this study focused on assessing the association between average pollution exposure from 2012 to 2015 and health outcomes in 2015. Under the premise of controlling for the baseline status in 2011, this essentially constitutes a longitudinal association study framework with retrospective exposure assessment.

In terms of statistical strategy, we adopted a two-stage analysis framework. Firstly, the ordinary least squares (OLS) regression model was adopted, with the comprehensive risk score of continuous neurodegenerative diseases as the dependent variable, to preliminarily explore the linear effect of air pollutant exposure. Subsequently, to enhance the clinical interpretability of the results, we converted the continuous risk scores at the 2015 follow-up into ordered categorical variables (low, medium, and high-risk groups) based on the fixed ternary cut-off points determined by the comprehensive score distribution of the baseline population in 2011 (cut1 = 2.98; cut2 = 4.85), and used these as outcome variables to construct an ordered Logistic regression model as the main analysis model. By adopting this “fixed cut-off point” approach, the aim is to establish a constant risk standard that is comparable across time, thereby enabling direct assessment of an individual’s absolute change relative to the baseline risk threshold from 2011 to 2015.

All final models were adjusted for covariates including age, sex, educational level, and the number of chronic conditions. The effects of continuous independent variables are presented as odds ratios (ORs) with their 95% confidence intervals (CIs). Furthermore, potential modifying effects were examined by introducing product interaction terms into the models, and stratified analyses were conducted by age group (30–59 years, 60- 74years, and over 75 years). All statistical tests were two-sided, with the significance level set at α = 0.05.

Results

Study population characteristics and covariate associations

A total of 6,180 participants from 28 province-level administrative divisions in China were included in this study. Their geographical distribution is shown in Fig. 3, and the participant inclusion flowchart is presented in Fig. 1. Spatial analysis conducted at the provincial scale revealed significant spatial heterogeneity in the sample distribution. From a macro perspective, the sample distribution closely aligns with China’s population density and socioeconomic development levels, exhibiting a typical spatial pattern characterized by higher density in the eastern regions and sparser coverage in the western areas.

Fig. 3.

Geographic distribution of the 6,180 study participants across China

Table 1 presents the summary statistics for the study population. Continuous variables are expressed as mean (standard deviation) or median (interquartile range), while categorical variables are described as frequency (percentage).The mean age of the participants was 71.76 (8.82) years, with the majority being young-old (57.86%) and oldest-old (36.25%) adults. Females accounted for 53.63% of the sample, and 66.32% were rural residents. The overall educational level was relatively low, with 69.78% having an education level of primary school or below. Most participants were married or had a partner (86.90%).Regarding chronic conditions, 33.19%, 29.45%, 19.29%, and 18.07% of participants had 0, 1, 2, and 3 or more chronic diseases, respectively. Concerning inflammatory biomarkers, the mean white blood cell (WBC) counts were 6.24 (0.02) ×10⁹/L in 2011 and 5.69 (0.03) ×10⁹/L in 2015. The levels of log-transformed high-sensitivity C-reactive protein (ln(hs-CRP)) were 0.13 (0.01) in 2011 and 0.39 (0.01) in 2015, suggesting an increase in systemic inflammation levels during the follow-up period. The mean activity scores were 1.43 (1.91) in 2011 and 1.78 (2.26) in 2015.The median (IQR) exposure levels to air pollutants in 2011 were as follows: PM_2.5: 59.94 (44.16–71.92) µg/m³, PM₁₀: 102.88 (75.37–123.46) µg/m³, NO₂: 29.53 (22.14–38.74) µg/m³, and O₃: 84.11 (81.54–88.88) µg/m³.

Table 1.

Summary Statistics of the Study Population (N = 6180)

Variables	Summary statistics
Age, mean(SD), yesrs	71.76(8.82)
Middle-aged	364(5.89)
Young-old adults	3576(57.86)
Old-old adults	2240(36.25)
Gender
Male	2866(46.37)
Female	3314(53.63)
Residence
Rural	4099(66.32)
Urban	2081(33.68)
Education level
Bachelor’s degree or above	83(1.34)
Junior high school or above	1784(28.88)
Primary school and below	4310(69.78)
Number of chronic disease
0	2051(33.19)
1	1820(29.45)
2	1192(19.29)
3	1117(18.07)
Marital status
Married or partnered	5372(86.90)
Others(Separated/divorced/widowed/never married)	808(13.10)
Activity
2011activies	1.43(1.91)
2015activies	1.78(2.26)
2011WBC, mean(SD),109/L	6.24(0.02)
2015WBC, mean(SD),109/L	5.69(0.03)
2011ln(hs-CRP), mean(SD)	0.13(0.01)
2015ln(hs-CRP), mean(SD)	0.39(0.01)
2011PM2.5, median(IQR),µg/m3	59.94(44.16–71.92)
2011PM10, median(IQR),µg/m3	102.88(75.37−123.46)
2011NO2, median(IQR),µg/m3	29.53(22.14–38.74)
2011O3, median(IQR),µg/m3	84.11(81.54–88.88)
2012−2015PM2.5, median(IQR),µg/m3	57.21(42.68–70.81)
2012−2015PM10, median(IQR),µg/m3	95.07(69.12−122.01)
2012−2015NO2, median(IQR),µg/m3	29.45(22.73–37.31)
2012−2015O3, median(IQR),µg/m3	84.39(80.02–88.37)
2011 Neurodegenerative Composite Risk Score
High risk	2060(33.33)
Medium risk	2060(33.33)
Low risk	2060(33.33)
2015 Neurodegenerative Composite Risk Score
High risk	2190(35.44)
Medium risk	1950(31.55)
Low risk	2040(33.01)

Continuous variables are presented as Mean (SD) or Median (IQR); categorical variables are presented as Frequency (%)

SD Standard Deviation, IQR Interquartile Range, WBC  White Blood Cell Count, hs-CRP  High-sensitivity C-reactive protein, ln(hs-CRP) Natural logarithm of hs-CRP, PM2.5 Particulate Matter with an aerodynamic diameter ≤ 2.5 μm, PM10 Particulate Matter with an aerodynamic diameter ≤ 10 μm, NO2 Nitrogen Dioxide, O3 Ozone

To assess the longitudinal changes in the neurodegenerative risk of the research subjects, we constructed a ternary risk level (low, medium, and high risk) based on a comprehensive functional score. Firstly, based on the comprehensive score distribution of all research subjects at the baseline survey in 2011, calculate the triquels (i.e., the 33rd and 66th percentiles) as fixed risk cut-off points. Subsequently, this set of fixed cut-off points was simultaneously applied to divide the individual comprehensive scores of 2011 and 2015 respectively, thereby obtaining comparable risk levels at the two time points. An individual rated as “high-risk” in 2015 means that their functional score is no lower than (or exceeds) the score threshold corresponding to the top 33% of individuals in the 2011 population. This approach ensures that the definition of risk levels remains consistent over time, enabling the Sankey plot (showing the flow of levels) and the ordered Logistic regression model (analyzing the influencing factors of levels) to be based on the same outcome variables.

Figure 4 illustrates the dynamic changes in neurodegenerative risk levels between 2011 and 2015. The results indicate significant temporal stability in risk status: 57.33% of individuals in the baseline low-risk group remained in the low-risk category at follow-up. Longitudinal observation revealed that although the size of the low-risk population remained relatively stable (decreasing slightly from 2,060 to 2,040 individuals), the cohort as a whole exhibited a net shift in risk distribution. This shift was characterized by a noticeable contraction in the medium-risk population (decreasing from 2,060 to 1,950 individuals) and a concurrent expansion of the high-risk group (increasing from 2,060 to 2,190 individuals). Analysis of transition pathways further revealed key characteristics of risk progression, including significant gender differences in transition patterns (with a higher proportion of females transitioning to or remaining in the high-risk category compared to males) and a notable stability within the high-risk state itself. In summary, the overall risk burden of the study population increased over time, with the proportion of high-risk individuals rising from 33.33% at baseline to 35.44%. These findings suggest that neurodegenerative risk status is not only persistent but also demonstrates an overall trend toward escalation.

Fig. 4.

Gender-Specific Transitions in Neurodegenerative Risk Levels (2011–2015)

Before running the ordered Logistic regression model, we evaluated the proportional dominance hypothesis through the Brant test. The test results of the main analysis model (i.e., the model adjusted for baseline score, age, gender, education, urban-rural, and chronic diseases) showed that the proportional advantage hypothesis held (P = 0.123). This result provides statistical support for the use of the standard ordered Logistic regression model. Through a multivariate ordered Logistic regression model, we analyzed the influence of sociodemographic and clinical factors on the risk flow of neurodegeneration. This model adopts the 2015 risk level defined based on the 2011 baseline standard as the ordered outcome variable to explore the predictive effect of each factor on the longitudinal change of the risk score level (Table 2).

Table 2.

Multivariable Ordinal Logistic Regression Analysis of Sociodemographic and Clinical Factors Associated with Neurodegenerative Disease Composite Risk Score Levels

Variables	Category	OR(95%CI)	P-valve
Age	Per year	1.03(1.03 to 1.04)	< 0.001
Gender	Male(Reference)
	female	2.15(1.94 to 2.40)	< 0.001
Education	Low(Reference)
	Middle	0.41(0.37 to 0.47)	< 0.001
	High	0.23(0.15 to 0.36)	< 0.001
Baseline risk level	Low(Reference)
	Middle	2.05(1.81 to 2.33)	< 0.001
	High	4.53(4.00 to 5.14)	< 0.001
Numbers of chronic diseases	0(Reference)
	1	1.59(1.40 to 1.80)	< 0.001
	2	2.83(2.44 to 3.28)	< 0.001
	> 2	7.04(5.91 to 8.40)	< 0.001

The results in this table are derived from a multivariable ordinal logistic regression model. The odds ratios (ORs) represent the relative change in the odds of being in a higher neurodegenerative risk category associated with a 10 µg/m³ increase in the long-term average pollutant concentration (2012–2015). All models were simultaneously adjusted for all variables listed in the table

The analysis results show that all the examined factors are significantly associated with the risk level in 2015. Specifically: The odds ratio (OR) for women to have a higher risk level compared to men was 2.15 (95% CI: (1.94–2.40), the ors of educational level were 0.41 (0.37–0.47) and 0.23 (0.15–0.36) respectively. The baseline (2011) risk status had a significant predictive effect, and the ORs of medium risk and high risk were 2.05 (1.81–2.33) and 4.53 (4.00-5.14) respectively The number of chronic diseases showed a significant dose-response relationship. The ORS of 1, 2, and ≥ 3 chronic diseases were 1.59 (1.40–1.80), 2.83 (2.44–3.28), and 7.04 (5.91–8.40), respectively. For each additional year of baseline age, the OR was 1.03 (1.03–1.04).The above results indicate that, after controlling for other factors, being female, having a low educational level, a high baseline risk status, a burden of multiple chronic diseases, and advanced age are independent risk factors for the increase in the comprehensive risk score of neurodegenerative diseases. That is, these characteristics can predict the flow of individuals to higher risk levels under a constant risk standard from 2011 to 2015.

The association between air pollution, inflammation and the comprehensive risk score of neurodegeneration

To comprehensively assess the long-term impact of air pollutant exposure on the comprehensive risk score of neurodegeneration, this study adopted a prospective cohort design. The main analysis took the comprehensive functional risk score (continuous variable) and ternary risk level (ordered variable) at the follow-up in 2015 as the outcome, and established multiple linear regression and ordered Logistic regression models respectively to enhance the clinical interpretability of the results. The baseline score of 2011, demographic and clinical confounding factors were uniformly adjusted in the model to analyze the independent association between average pollutant exposure from 2012 to 2015 and the outcome in 2015.

In univariate and multivariate OLS regression analyses, among the four air pollutants, only PM2.5 showed stable significance in its association with the comprehensive score of neurodegenerative risk. As shown in Table 3, in the unadjusted model, PM2.5 was significantly positively associated with the risk score (β = 0.012, 95% CI: 0.005 to 0.105). After further controlling for baseline scores and adjusting for confounding factors such as age, gender, educational level, and the number of chronic diseases, the association remained significant and the confidence interval was more precise (adjusted β = 0.010, 95% CI: ) (0.008 to 0.016), indicating that long-term PM2.5 exposure is an independent influencing factor for the risk of neurodegeneration. Based on this core and stable finding and considering the key role of PM2.5 in previous studies, this research will subsequently focus on PM2.5 and use an ordered Logistic regression model to further explore the age-stratified differences in its effects. Among other pollutants, NO₂ and O₃ only showed significant association in the univariate model (NO₂: β = 0.031, 95% CI: 0.0280 to 0.040; O₃: β = 0.023, 95% CI: 0.007 to 0.040), but after multivariate adjustment, their associations were NO longer statistically significant (NO₂ adjusted β = 0.026, 95% CI: -0.003 to 0.037; Adjusted O₃ β = 0.022, 95% CI: -0.004 to 0.042. PM10 was not significant in either model. Furthermore, inflammatory markers (hs-CRP and WBC) did not show statistical significance in all models.

Table 3.

Associations of Air Pollutants and Systemic Inflammation Markers with Neurodegenerative Risk Score (OLS Regression Models)

Variables	Model1(Crude) β coefficient(95%CI)	Model2(adjusted) β coefficient(95%CI)
PM2.5	0.012(0.005 to 0.105)*	0.010(0.008 to 0.016)***
PM10	0.003(−0.002 to 0.006)	0.011(-0.003 to 0.034)
NO2	0.031(0.028 to 0.040)*	0.026(-0.003 to 0.037)
O3	0.023(0.007 to 0.040)*	0.022(−0.004 to 0.042)
hs-CRP	0.020(−0.010 to 0.0310)	−0.005(−0.150 to 0.010)
WBC	0.002(−0.020 to 0.010)	0.006(−0.150 to 0.010)

Parentheses contain 95% confidence intervals based on robust standard errors.Model 1: Unadjusted crude model; Model 2: Multivariable model adjusted for age, sex, educational level, and chronic conditions. All effect estimates are expressed per 10 µg/m³ increase in pollutant concentration

Abbreviations: PM2.5 Particulate matter with an aerodynamic diameter ≤ 2.5 μm, PM10 Particulate matter with an aerodynamic diameter ≤ 10 μm, NO2 Nitrogen dioxide, O3 Ozone

*** p < 0.001,** p < 0.01, *p < 0.05

The impact of air pollution on the comprehensive risk score of neurodegeneration

Based on the stratified analysis results from the ordinal logistic regression model (Fig. 5), this study systematically evaluated the differential effects of exposure to four air pollutants (PM_2.5, PM₁₀, NO₂, and O₃) on neurodegenerative risk across different age groups.The results indicate that PM_2.5 exposure had a significant positive effect on neurodegenerative risk among young-old adults aged 60–74, with an adjusted odds ratio (OR) of 1.52 (95% Confidence Interval [CI]: 1.01–2.31, P = 0.049). This suggests that in this age group, when the PM_2.5 concentration increased from the first to the third quartile (44.16–71.92 µg/m³), the odds of moving to a higher neurodegenerative risk category increased by 52%. However, for middle-aged adults (30–59 years) and the oldest-old adults (> 75 years), the effect estimates for PM_2.5, while greater than 1 (OR = 2.06, 95% CI: 0.61–6.99; OR = 1.21, 95% CI: 0.71–2.06, respectively), did not reach statistical significance (P = 0.245; P = 0.476).For PM₁₀ exposure, risk estimates across all age groups showed positive trends (OR > 1) but also failed to reach statistical significance (all P > 0.05). The OR for the young-old group was 1.51 (95% CI: 0.93–2.45, P = 0.095). The effect patterns of NO₂ and O₃ were heterogeneous across age groups: the risk estimate for NO₂ in the oldest-old group was 1.73 (95% CI: 0.65–4.64), while it was 0.81 (95% CI: 0.38–1.75) in the young-old group, neither being statistically significant. O₃ showed slight protective or neutral effects (OR ≤ 1) in all age groups, but these were not statistically significant.It is worth noting that the insignificant results in the middle-aged group are very likely due to the limited sample size, which may lead to insufficient statistical power.

Fig. 5.

Association between PM2.5 Exposure stratified by Age Group and Neurodegenerative Risk Score grades. All effect estimates are expressed per 10 µg/m³ increase in pollutant concentration. Abbreviations: PM2.5: particulate matter with an aerodynamic diameter ≤ 2.5 μm; PM10: particulate matter with an aerodynamic diameter ≤ 10 μm; NO2: nitrogen dioxide; O3: ozone.

In summary, the stratified analysis indicates that among the four air pollutants examined, PM_2.5 exposure has a significant and specific effect on neurodegenerative risk in the young-old population aged 60–74.

The modifying effect of systemic inflammation

This study further explored the effect modification of baseline (2011) systemic inflammation levels between average air pollutant exposure from 2012 to 2015 and the risk of neurodegeneration in 2015. Based on adjusting the baseline risk score, age, gender, urban and rural residence, and the number of chronic diseases, the interaction terms of four pollutants ( PM_2.5, PM₁₀, NO₂, O₃) and two inflammatory markers (WBC, hs-CRP) were introduced respectively. The analysis results showed that the interaction terms between all pollutants and inflammatory markers were not statistically significant (all P < 0.05), and the regression coefficients (Beta) were mostly close to zero (Table 4). Specifically, the interactions of PM_2.5, PM₁₀, NO₂ or O₃ with baseline WBC or hs-CRP did not reach a significant level. This indicates that in the population of this study, no significant effect modification was found by baseline inflammation levels on the association between long-term air pollution exposure and neurodegenerative risk.

Table 4.

Interaction analysis between air pollutants and baseline inflammatory markers

Pollutant	Inflammation	Beta(95%CI)	P
PM2.5	WBC2011	-9.00(-37.21, 19.21)	0.527
PM2.5	hs-CRP2011	-0.01(-0.04, 0.02)	0.455
PM10	WBC2011	-5.14(-20.81, 10.53)	0.517
PM10	hs-CRP2011	-0.01(-0.03, 0.02)	0.608
NO2	WBC2011	0.01(-0.02, 0.04)	0.604
NO2	hs-CRP2011	-4.67(-14.84, 13.50)	0.945
O3	WBC2011	0.03(-0.05, 0.11)	0.503
O3	hs-CRP2011	0.04(-0.20, 0.28)	0.740

N = 6,180, contaminants: per 10 µg/m³; hs-CRP: converted by natural logarithm

Sensitivity analysis

To evaluate the robustness of the main analysis results for the comprehensive scoring construction method, we conducted a sensitivity analysis. We used principal component analysis (PCA) to reduce the dimensions of four standardized functional indicators (grip strength, CES-D-10, MMSE, IADL), and took the first principal component that explained the variance the greatest as the new "Comprehensive Risk Score for Neurodegeneration". The PCA score was significantly positively correlated with the original equal-weighted Z score (Pearson's R = 0.476, p < 0.001). With this PCA score as a continuous outcome, the re-fitted multivariate linear regression model showed that long-term PM2.5 exposure (for every 10 µg/m³ increase) was still significantly associated (adjusted β = 0.010, 95% CI: 0.004-0.016, p = 0.002). This result is consistent with the conclusion of the principal ordered logistic regression model, indicating that the association between PM_2.5 and the risk of neurodegeneration is insensitive to the weighted method of score construction, and the core findings are robust. Given that the "number of chronic diseases" and the "baseline score" may be on a causal path, we fitted a restrictive model that did not adjust for these two variables. Compared with the main model, the effect estimates of PM_2.5 remained significant, indicating the robustness of the core association.

Discussion

This study, through a prospective cohort design, systematically explored the longitudinal association between long-term air pollution exposure and the comprehensive risk score of neurodegeneration in middle-aged and elderly people, and obtained three major findings. Firstly, the individual’s comprehensive risk of neurodegeneration shows a certain degree of stability over four years, but there is a main trend of evolving towards high risk. Secondly, age, gender, educational level, baseline risk status and the burden of chronic diseases jointly constitute the multi-dimensional influencing factors of risk. Most importantly, among the four pollutants of PM_2.5, PM₁₀, NO₂ and O₃ evaluated systematically, only long-term PM_2.5 exposure was significantly associated with an increase in the comprehensive risk score of neurodegeneration, and this association showed obvious age heterogeneity - only statistically significant in the “young elderly” population aged 60–74. This age-specific association pattern suggests that early old age may be more sensitive to the potential impact of air pollution. The absence of a significant association in the middle-aged group may be attributed to intact neuroprotective mechanisms and preserved physiological function [35], whereas in the oldest-old adults, the signal of a weak environmental effect might have been masked by competing risk factors such as multimorbidity and genetic susceptibility [51].

It is noteworthy that the variations in effect estimates across different air pollutants may stem from heterogeneity in study population characteristics, regional pollution levels, outcome measurements, and analytical strategies [52]. Among the four pollutants systematically evaluated (PM_2.5, PM₁₀, NO₂, O₃), only PM_2.5 demonstrated a specific association. This may be attributed to its unique physicochemical properties – its ultrafine particulate fraction possesses a larger specific surface area, greater pulmonary penetration efficiency, and enhanced ability to translocate to the brain, enabling direct effects on the central nervous system via pathways such as crossing the blood-brain barrier or the olfactory bulb route [53–56].

Our exploratory analysis of the exposure time window provides an important reference for understanding the potential long-term effects of PM_2.5. This study used the average exposure from 2012 to 2015 to assess the outcome in 2015. This design was not strictly “completely prospective” in terms of time sequence (exposure measurements occurred entirely before the outcome). However, by uniformly adjusting the baseline neurodegenerative comprehensive risk score in 2011, we effectively controlled the health status of individuals at the start of the study, thereby directing the main association estimate towards the risk changes between 2011 and 2015. This analytical strategy helps to reduce confounding caused by initial health differences and enhance the rationality of causal inferences. The analysis results show that long-term average exposure to PM_2.5 is associated with an increase in the comprehensive risk score of neurodegeneration, which is more consistent with the characteristic that the physiological and pathological processes of neurodegeneration usually require a longer accumulation period, suggesting that its impact may be a chronic and progressive process.

Although we hypothesized that baseline inflammatory levels might modify the effects of PM_2.5, the interaction analysis of this study showed that inflammatory markers (hs-CRP, WBC) did not exhibit significant effector modification effects, a finding that is partially consistent with the results of other studies using the same cohort [12]. The interpretation of this negative result requires caution, as it may suggest: (1) In this study population, inflammation may not be the main modifying factor associated; (2) The influence of PM_2.5 may involve other mechanisms, such as directly acting on the central nervous system [57], or exerting effects through pathways such as oxidative stress; (3) The neuroinflammatory response may have compartmental characteristics [58], and the association between its local activation and changes in systemic circulatory markers is limited. Future research needs to further explore whether other potential biological or psychosocial factors modify the association between air pollution and neurological health.

From a public health perspective, this study found that long-term PM_2.5 exposure is associated with an elevated comprehensive risk score of neurodegeneration in people aged 60–74, suggesting that this age group may be a potentially susceptible group worthy of special attention. This discovery provides preliminary epidemiological clues for targeted air pollution health protection in early old age. If future research can verify this association through more detailed exposure assessment and longer follow-up, it will help improve the air quality standards mainly based on cardiopulmonary effects, thereby more comprehensively protecting the multi-dimensional health of the aging population.

Limitations

This study has several advantages: it is based on a large longitudinal cohort (CHARLS) with national representativeness; At the same time, the long-term average exposure to four major air pollutants was evaluated; Adopt a comprehensive risk score to capture multi-dimensional functional changes; The baseline risk status was prospectively adjusted, enhancing the rationality of causal inference. However, this study also has several limitations that require careful interpretation:

Firstly, to protect personal privacy, the assessment of air pollution exposure is only conducted at the urban level, failing to capture the internal differences within the city. This may lead to misclassification of non-differential exposure, and such misclassification usually causes the effect estimation to be biased towards zero, thereby possibly underestimating the true health risks. Secondly, the Cronbach’s alpha value of the constructed neurodegenerative composite risk score is 0.6, which is at a medium level. This reflects that the score integrates relatively independent functional domains such as movement, emotion, cognition, and daily activities. Although it can provide a multi-dimensional risk profile, its measurement reliability as a single continuous variable does not reach the ideal standard (≥ 0.7). Caution should be exercised when interpreting its absolute score. Thirdly, in terms of design and confounding control, although many variables have been adjusted, it is still impossible to completely eliminate the influence of residual confounding (genetics, fine socio-economic factors, etc.). The study used the average exposure from 2012 to 2015 to assess the outcome in 2015 and adjusted the baseline score in 2011. Although this design has longitudinal logic, the exposure did not completely precede the outcome, and room should be left for causal inference. Meanwhile, when conducting stratified analysis by age in this study, there was a significant imbalance in the sample size of subgroupsTherefore, we mainly regard the age stratification results as exploratory findings to hint at potential vulnerable Windows rather than conclusive conclusions. Future research needs to be verified in larger-scale cohorts with a more balanced distribution across age groups. Furthermore, the effector modification effect of inflammation is based only on a single baseline measurement and fails to reflect its changes over time, which may affect the sufficiency of the analysis. Finally, this study is based on the Chinese CHARLS cohort, which has specific environmental and population characteristics. Extra caution is needed when extrapolating the conclusions to other regions or populations.

Conclusion

Based on the large-scale longitudinal data from the China Longitudinal Study of Health and Retirement (CHARLS), this study systematically evaluated the relationship between environmental air pollutants and the composite risk score of neurodegeneration. The results showed that there was a significant age-specific association between long-term PM_2.5 exposure and the composite risk score of neurodegeneration: for every 10 µg/m³ increase in PM_2.5 concentration (average from 2012 to 2015), the probability of an increase in the composite risk score of neurodegeneration among people aged 60–74 increased by 52% (OR = 1.52). Among the four pollutants evaluated, PM_2.5 demonstrated the most consistent and specific association patterns. Furthermore, the use of multi-year average exposure metrics is consistent with the hypothesis of chronic cumulative pathological processes. This study did not find that baseline systemic inflammation levels had a modifying effect on the association between air pollutant exposure and the composite risk score of neurodegeneration. Overall, these findings suggest that long-term exposure to PM_2.5 is associated with adverse conditions of multi-dimensional neurodegenerative risk indicators, especially among young elderly populations. This highlights the possibility of a period that focuses on protecting environmental health. The research results also emphasize the importance of strengthening targeted protection for vulnerable subgroups, especially those with lower educational attainment, providing crucial epidemiological evidence for formulating precise intervention strategies.

Abbreviations

PM_2.5

Fine Particulate Matter (particles with an aerodynamic diameter ≤ 2.5 micrometers)

PM₁₀

Inhalable Particulate Matter (particles with an aerodynamic diameter ≤ 10 micrometers)

NO₂

Nitrogen Dioxide

O₃

Ozone

CRP

High-sensitivity C-reactive protein

WBC

White blood cells

OLS

Ordinary Least Squares

CHARLS

China Health and Retirement Longitudinal Study

CESD-10

The abbreviated (10-item) Center for Epidemiologic Studies Depression Scale

ADL

Activity of Daily Living

SD

Standard Deviation

SE

Standard Error

CI

Confidence Interval

Authors’ contributions

R.G. and G.S. conceived the paper. R.G., W.Z., and G.D. collected the data. R.G. drafted the manuscript. L.S. and G.S. revised the manuscript. L.S., J.D., and W.Z. provided ideas for revising the manuscript. G.S. contributed to the critical revision of the manuscript for important intellectual content and approved the final version of the manuscript. R.G. and G.S. are the study guarantors. All authors have read and agreed to the published version of the manuscript.

Funding

The research was funded by Philosophy and Social Sciences Planning Project of Guangdong Province in 2023: Research on the mechanism of national voluntary epidemic prevention behavior based on Repast-fsQCA in the context of COVID-19 “Class B and B control”, grant number GD23CGL11; Research Project of Guangdong Institute of Modern Hospital Management, grant number GDXDYGS2025041; Special Project on Teaching Reform Research and Practice in International Education of Southern Medical University for the 2024 Academic Year, grant number 2024GJG003; 2025 Southern Medical University Research Project on “Health and Wellness in Belt and Road Countries” — From the “Health Silk Road” to the “Health Community with a Shared Future”: A Study on the Assessment of Health System Resilience and Innovative Cooperation Mechanisms in Belt and Road Countries; Southern Medical University's Higher Education Teaching Reform Project in 2025 — Reconstruction of the All-English Training System for International Medical Talents Serving the "Healthy China" Strategy: Integrated Reform of Curriculum, Faculty, and Innovation and Entrepreneurship; and 2026 Guangdong Provincial Graduate Education Innovation Program Project — "China Experience—International Standards—Innovative Practice" Tripartite Synergy: Construction and Empirical Study of a Pathway for High-Level Internationalized Medical Talent Development, grant number 2026JGXM_089.

Data availability

The original data of this research are publicly available in the CHARLS website [http://charls.pku.edu.cn].

Declarations

Ethics approval and consent to participate

All participants provided informed written consent prior to their interviews, and the present study received review and approval from the Biomedical Ethics Review Board of Peking University. The ethical approval number is IRB00001052-11015. The survey was in accordance with the ethical standards delineated in the 1964 Declaration of Helsinki and its subsequent amendments.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.