Associations of PM 2.5 and its major chemical components with cognitive function: A nationwide prospective cohort study among middle-aged and older adults in China

Abstract

Objective

This study aimed to investigate the associations between fine particulate matter and its major chemical components and cognitive function among middle-aged and older adults in China.

Methods

We conducted a nationwide prospective cohort study using data from the China Health and Retirement Longitudinal Study (2011–2018). Cognitive function was repeatedly assessed through standardized tests of memory and mental status. Annual average concentrations of fine particulate matter and its five major components (sulfate, nitrate, ammonium, black carbon, and organic matter) were estimated at the city level. Fixed-effects models and restricted cubic spline analyses were used to evaluate associations, and random forest models were used to rank the relative importance of components.

Results

Higher exposure to fine particulate matter and several of its major components was significantly associated with lower cognitive scores. Among these components, sulfate exhibited the strongest adverse association with cognitive function. The findings were consistent across multiple sensitivity analyses, including those restricted to provincial capitals and those adjusting for potential confounders.

Conclusions

Exposure to fine particulate matter and its chemical components may contribute to cognitive impairment among middle-aged and older adults in China. Sulfate appears to be particularly detrimental. These results highlight the need for targeted air pollution control policies that address specific fine particulate matter components to mitigate the burden of cognitive impairment.

Introduction

Dementia, a group of disorders characterized by a significant decline in cognitive function, poses a substantial burden on individuals and society, making it one of the most significant healthcare challenges of the 21st century. ¹ Advancing age is the primary risk factor associated with the development of cognitive decline, a key feature of dementia. ² Delaying cognitive decline could potentially prevent dementia or improve the quality of life of affected individuals, thereby exerting a significant impact on social healthcare. Although curative treatments for dementia are currently unavailable, it is possible to prevent cognitive decline by addressing modifiable risk factors, including air pollution. ³

Fine particulate matter ( ${\mathrm{PM}}_{2.5}$ ), which comprises five major chemical components including sulfate ( ${\mathrm{SO}}_4^{2-}$ ), nitrate ( ${\mathrm{NO}}_3^{-}$ ), ammonium ( ${\mathrm{NH}}_4^{+}$ ), black carbon ( $\mathrm{BC}$ ), organic matter ( $\mathrm{OM}$ ), and crustal elements, represents the fourth leading risk factor for deaths and disability-adjusted life years (DALYs) in China. ⁴ ${\mathrm{PM}}_{2.5}$ plays a critical role in influencing the global climate,^5,6 deteriorating air quality,^7,8 and contributing to adverse health outcomes.^9,10 Although the effects of total ${\mathrm{PM}}_{2.5}$ are multifaceted, the impacts of its individual chemical components vary significantly because of their distinct physicochemical properties. For example, ${\mathrm{SO}}_4^{2-}$ predominantly scatters solar radiation, whereas $\mathrm{BC}$ primarily absorbs it, resulting in contrasting climate effects. BC is potentially more toxic than ${\mathrm{SO}}_4^{2-}$ and may therefore exert varying effects on the health of human beings. Considering that the chemical components of ${\mathrm{PM}}_{2.5}$ are more directly associated with changes in emissions compared to those of total ${\mathrm{PM}}_{2.5}$ , understanding the association between the chemical composition of ${\mathrm{PM}}_{2.5}$ and cognitive decline can provide more targeted guidance for policymaking.^11–13 Moreover, although the World Health Organization (WHO) has elaborated on the concentration limits of ${\mathrm{PM}}_{2.5}$ in the latest air quality guidelines released in 2021, it did not provide specific descriptions and regulations on its chemical composition. ¹⁴ Therefore, investigating the public health implications of ${\mathrm{PM}}_{2.5}$ ’s chemical composition is essential for generating evidence to guide environmental management and decision-making.

China is currently struggling with an aging population. ¹⁵ Consequently, the burden of age-related health conditions, including cognitive decline, is expected to increase. ¹⁶ Epidemiological studies have demonstrated that ${\mathrm{PM}}_{2.5}$ exposure is associated with an increased risk of cognitive impairment.^3,17 However, there remains limited understanding of the effects of the chemical components of ${\mathrm{PM}}_{2.5}$ on cognitive decline; improved understanding of these relationships would help clarify which components contribute to this association. Emerging evidence suggests that specific chemical components of PM_2.5, such as ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NO}}_3^{-}$ , ${\mathrm{NH}}_4^{+}$ , BC, and OM, adversely affect cognitive function through biological pathways involving neuroinflammation, oxidative stress, and vascular impairment. Prior nationwide studies in China have rarely examined these components individually.^18–20 Furthermore, such insights can offer new avenues for addressing the burden of ${\mathrm{PM}}_{2.5}$ -related cognitive decline. Therefore, it is essential to investigate the association between ${\mathrm{PM}}_{2.5}$ and its chemical components with cognitive function among middle-aged and older individuals in China. Such research may inform policy formulation not only within China but also in other low- or middle-income countries (LMICs). Additionally, it hold considerable social importance in addressing the growing health challenges associated with aging populations.

This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), a nationwide survey of middle-aged and older individuals in China. It investigated the relationship between individual-level exposure to ambient ${\mathrm{PM}}_{2.5}$ and its major chemical components, including $\mathrm{BC}$ , $\mathrm{OM}$ , ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NO}}_3^{-}$ , ${\mathrm{NH}}_4^{+}$ , at the participant’s residential locations. By conducting a longitudinal analysis, the study aimed to examine the association between ${\mathrm{PM}}_{2.5}$ and its chemical components and cognitive function among middle-aged and older Chinese adults. The findings of this study may help elucidate the link between ${\mathrm{PM}}_{2.5}$ and its chemical components and cognitive impairment, thereby providing a basis for targeted prevention and control measures for ${\mathrm{PM}}_{2.5}$ components.

Methods

Study population

CHARLS is an ongoing cohort survey designed to collect high-quality data from a nationally representative sample in China. The survey covers a wide range of information, including demographics, lifestyle factors, and health-related details. Details on the CHARLS study design, including sampling procedures, data acquisition methods, and quality assessment, are provided in section 1.1 of the Supplementary Material.

For this study, a prospective analysis was conducted using data from the 2011 to 2018 waves of the survey. The inclusion criteria were as follows: (a) individuals aged ≥45 years at the time of the survey and (b) individuals with available cognitive function data. The exclusion criteria were as follows: (a) individuals with missing age information; (b) individuals with incomplete urbanicity data; and (c) individuals with fewer than two cognitive function test scores. After applying these criteria, the study included a final analytic sample of 16,332 individuals, comprising 54,615 observations. All included participants had a minimum of two cognitive function measurements. The sample selection process is detailed in Figure S1 of the Supplementary Material. Figure 1 illustrates the geographical distribution of the participants across 126 survey cities. The CHARLS protocol was approved by the Ethical Review Committee of Peking University (Approval Number: IRB00001052–11015), and written informed consent was obtained from all participants. This survey adhered to the most recent Declaration of Helsinki (2013 revision). The reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. ²¹

Figure 1.

Sample distribution.

The present study was conducted in accordance with the ethical principles of the Declaration of Helsinki (1975), as revised in 2024. All patient information in the CHARLS dataset was fully deidentified by the data provider prior to analysis, ensuring participant anonymity. This prospective cohort study used the CHARLS dataset (2011–2018) to examine the association between PM_2.5 components and cognitive function. As this study involved secondary data analysis, individual written informed consent was not required; however, all participants in the CHARLS cohort had provided written informed consent at baseline.

Assessment of cognitive function

The following five components were used to assess the cognitive function: (a) orientation ability (0–4 points), in which participants identified the current day, week, month, and year; (b) immediate word recall ability (0–10 points), in which participants immediately repeated the 10 Chinese nouns they had just seen, in any order; (c) delayed word recall ability (0–10 points), in which participants repeated the same list of words after a 4 min interval; (d) numeric ability (0–5 points), in which participants completed five consecutive subtractions of 7 from 100 without any aids; and (e) visuoconstruction ability (0–1 point), in which participants redrew a previously shown geometric figure. Better cognitive performance was indicated by higher scores on each test. The primary outcome was the global cognitive score, which ranged from 0 to 30 points and was calculated by summing the scores of five cognitive test components. According to the recommendations of CHARLS data usage from previous studies,^22,23 cognitive function was categorized into two separate dimensions. The mental status dimension (0–10 points), which includes orientation, numeric ability, and visuoconstruction, represents the ability to acquire and retain knowledge and skills over time. Episodic memory, the second dimension (0–20 points), assesses immediate and delayed word recall. It focuses on learning ability performance and processing of unfamiliar materials, and it declines significantly with age. Additional information on the cognitive function test’s design can be found in section 1.2 of the Supplementary Material.

Exposure of assessment of ${\mathrm{PM}}_{2.5}$ and its components

Data on the major chemical components of ${\mathrm{PM}}_{2.5}$ were obtained from the Tracking Air Pollution (TAP) China database, ²⁴ which aims to develop a high spatial resolution dataset of major chemical components of air pollutants across China. To ensure privacy, residential addresses of CHARLS participants were geocoded at their municipal administrative unit levels. Long-term exposure to ${\mathrm{PM}}_{2.5}$ and its major chemical components was assessed at the municipal level using well-validated spatiotemporal datasets.^25–27 Details of the spatiotemporal model method of ${\mathrm{PM}}_{2.5}$ and its major chemical components are provided in section 1.3 of the Supplementary Material. ${\mathrm{PM}}_{2.5}$ chemical components showed good agreement with the existing actual observation results (daily-scale correlation coefficients ranged from 0.64 to 0.80, with most normalized mean biases within ±20%), and the model cross-validation R values were 0.64 for $\mathrm{BC}$ , 0.72 for $\mathrm{OM}$ , 0.70 for ${\mathrm{SO}}_4^{2-}$ , 0.75 for ${\mathrm{NO}}_3^{-}$ , and 0.75 for ${\mathrm{NH}}_4^{+}$ .

This study measured the average of the major chemical components of ${\mathrm{PM}}_{2.5}$ concentrations in all the cities where the participants lived based on the longitude and latitude. Gridded estimates were aggregated within the 126 city-level administrative area polygons to calculate the monthly concentrations of air pollutants from 2001 to 2018. These gridded estimates had a resolution of approximately 0.1° × 0.1° (equivalent to 10 × 10 km). Exposure windows of 1, 2, 3, 5, and 10 years before the interview date were used. Annual averages of ${\mathrm{PM}}_{2.5}$ and its major chemical components were calculated for the surveyed month and for the 11, 23, 35, 59, or 119 preceding months (i.e. the 12-, 24-, 36-, 60-, or 120-month moving averages) and then assigned to each participant using redacted geocodes.

Covariates

Covariates were selected based on prior epidemiological evidence, biological plausibility, and the availability of corresponding data within the CHARLS dataset. The final models were adjusted for age, sex, education, marital status, smoking status, alcohol consumption, body mass index (BMI), hypertension, diabetes, and region. According to existing literature, several covariates related to air pollution and cognitive function were selected.^28,29 They included the following: (a) Demographic variables. Age, sex, urbanicity (rural and urban), solitary (living alone and with a partner), household income, education (0 years, 1–6 years, and >6 years), and region (western, central, and eastern China); (b) Health behavior factors. Smoking, drinking, sleep duration at night, and social activity; and (c) Comorbidities. History of diabetes, hypertension, dyslipidemia, heart disease, stroke, depressive symptoms, and subjective memory. Additional details on covariates are provided in Table S1 of the Supplementary Material.

Statistical analysis

Participant characteristics were summarized using mean ± SD for continuous variables and frequencies with percentages for categorical variables. Linear mixed models were employed to examine the impact of ${\mathrm{PM}}_{2.5}$ and its major chemical components ( $\mathrm{BC}$ , $\mathrm{OM}$ , ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NO}}_3^{-}$ , and ${\mathrm{NH}}_4^{+}$ ) on cognitive function. The region was treated as a random effect (RE) to account for potential geographic variation.

In order to analyze the impact of prolonged exposure to diverse chemical constituents, we computed the average levels for different time frames of exposure, such as 1-year (12-month), 2-year (24-month), 3-year (36-month), 5-year (60-month), and 10-year (120-month) rolling averages.

To assess the reliability of the findings, three different models were constructed to examine the connections, with additional variables included progressively. Fixed-effects (FE) models were applied to control for unobserved time-invariant characteristics at the individual level. As time-invariant covariates such as sex and education cannot be estimated in FE models, we also conducted RE and multivariable mixed-effects models as sensitivity analyses. In these analyses, Model 1 was adjusted for age and marital status, whereas Model 2 further included education and other sociodemographic covariates. The FE terms in Model 1 (unadjusted) consisted of ${\mathrm{PM}}_{2.5}$ concentrations and their chemical constituents, whereas study region was included as a RE. Model 2 included age (continuous) and sex. In addition to age and sex, Model 3 further adjusted for education, solitary status, urbanicity, BMI, smoking, drinking, diabetes, hypertension, dyslipidemia, heart disease, and stroke (multivariable adjusted). The impacts were evaluated as changes in cognitive performance rating for each 1 SD increase in air pollution exposure, with corresponding 95% confidence intervals (CIs). In the regression model, interaction terms between exposure and stratified variables were included to conduct subgroup analyses, considering factors such as cognitive function score, chronological age, sex, and personal recollection. Each PM_2.5 component was initially analyzed in a single-pollutant model to avoid multicollinearity. Sensitivity analyses were further conducted using multipollutant models that adjusted for coexposure to other components. To enhance comparability across PM_2.5 components measured on different scales, all exposure variables were standardized using a z-score transformation (subtracting the mean and dividing by the SD). Thus, regression coefficients represent the change in cognitive score per 1 SD increase in exposure. All analyses accounted for the complex survey design of CHARLS (stratification, clustering, and multistage sampling) by applying longitudinal sampling weights and specifying stratification and primary sampling unit (PSU) variables to ensure national representativeness and accurate variance estimation.

To investigate the possible associations between air pollution exposure and cognitive function, a linear regression analysis using robust variance estimates was conducted. This analysis included the utilization of restricted cubic spline functions to assess the impact of air pollution. The splines were automatically set with three knots at the 5th, 50th, and 95th percentiles. The limited missing data were excluded, and the models were constructed using the available data. Restricted cubic splines were used to flexibly model potential nonlinear relationships between PM_2.5 exposure and cognitive outcomes, as they provide greater flexibility than quadratic functions and avoid unrealistic shapes at the tails. In addition, random forest models were applied to evaluate the relative importance of individual PM_2.5 components, as this method is robust to multicollinearity and capable of capturing nonlinear effects. These results were compared with regression-based models to ensure robustness.

The random forest method ³⁰ is a robust statistical approach utilized for evaluating the significance of variables in forecasting an outcome. To assess the impact of various pollutants on cognitive function scores, the random forest technique was utilized for prioritizing their significance. The random forest algorithm inherently measures the variable importance by evaluating the decrease in the model’s predictive performance when a given variable is randomly permuted. This metric is referred to as the decrease in node impurity or the Gini importance index. In order to confirm the robustness of the findings, ridge regression, LASSO, and mixed-effect models were additionally employed for verification.

To ensure the robustness of the findings, sensitivity analyses were conducted, taking into account additional variables such as depressive symptoms and the history of hypertension, diabetes, dyslipidemia, or stroke. The E-value ³¹ is a statistical measure that calculates the minimum level of correlation that an unmeasured confounding factor would need to have with the exposure (pollutant) and the outcome (cognitive function) on the risk ratio scale to completely explain the observed association. Larger E-values imply that a stronger unmeasured confounder would be necessary to invalidate the observed correlation, indicating increased resilience of the results to unmeasured confounding. We performed sensitivity analyses restricted to participants residing in provincial capital cities, where air quality monitoring networks are denser and provide higher spatial resolution. The results were consistent with the primary analyses, suggesting robustness of the findings.

Results

Characteristics of participants

The study involved 47,078 observations in total, collected from 15,807 participants, of whom 47.09% were female. The average age of participants was 58.99 ± 8.96 years. The mean (SD) scores for overall cognitive function, episodic memory, and mental state were 15.44 (4.74), 7.78 (3.54), and 7.64 (2.17) points, respectively. Table 1 presents additional selected covariates.

Table 1.

Summary of variable characteristics in general and by region during the study period (2011–2018).

Characteristics	Total (n = 47,078)	Western China (n = 14,800)	Central China (n = 15,714)	Eastern China (n = 16,564)	p
Demography
Age, years	58.99 ± 8.96	58.98 ± 9.08	58.91 ± 8.84	59.09 ± 8.96	0.1651
Female	22,786 (47.09)	7058 (46.27)	7623 (47.36)	8105 (47.56)	0.0469
Urban	20,457 (42.28)	6179 (40.51)	6958 (43.23)	7320 (42.96)	<0.001
Solitary	40,898 (84.53)	12,728 (83.44)	13,611 (84.58)	14,559 (85.45)	<0.001
Income, CNY	5538.63 ± 18,698.58	3925.53 ± 12,343.14	5091.55 ± 22,265.01	7409.56 ± 19,559.01	<0.001
Education					<0.001
0 year	6932 (14.33)	2293 (15.03)	2220 (13.79)	2419 (14.20)
0–6 years	22,079 (45.63)	7646 (50.12)	6694 (41.59)	7739 (45.41)
>6 years	19,379 (40.05)	5315 (34.84)	7181 (44.62)	6883 (40.39)
Clean Fuel	17,882 (39.65)	6497 (48.00)	6263 (40.37)	5122 (31.91)
Health behaviors
Smoking	19,332 (39.95)	6202 (40.66)	6555 (40.73)	6575 (38.58)	<0.001
Drinking	23,475 (48.51)	7605 (49.86)	7869 (48.89)	8001 (46.95)	<0.001
Night sleep, h	6.31 ± 1.87	6.20 ± 1.99	6.25 ± 1.88	6.48 ± 1.74	<0.001
Sociable	25,302 (52.30)	7428 (48.71)	8890 (55.24)	8984 (52.74)	<0.001
Comorbidities
Diabetes	5544 (11.97)	1547 (10.62)	1937 (12.52)	2060 (12.64)	<0.001
Hypertension	19,136 (45.11)	5941 (44.32)	6241 (44.09)	6954 (46.79)	<0.001
Dyslipidemia	8272 (18.07)	2271 (15.87)	3071 (19.98)	2930 (18.20)	<0.001
Heart disease	7972 (17.18)	2578 (17.67)	2915 (18.80)	2479 (15.20)	<0.001
Stroke	1785 (3.84)	522 (3.57)	748 (4.83)	515 (3.15)	<0.001
CESD-10	7.61 ± 5.95	8.51 ± 6.19	7.78 ± 6.00	6.65 ± 5.52	<0.001
Subjective memory	4.04 ± 0.82	4.10 ± 0.79	4.09 ± 0.80	3.95 ± 0.85	<0.001
Cognitive function
Cognitive function	15.44 ± 4.74	14.85 ± 4.79	15.59 ± 4.75	15.84 ± 4.64	<0.001
Episodic memory	7.78 ± 3.54	7.48 ± 3.55	7.86 ± 3.54	7.97 ± 3.50	<0.001
Mental status	7.64 ± 2.17	7.34 ± 2.26	7.70 ± 2.15	7.85 ± 2.09	<0.001

Values were described in mean ± SD for continuous variables and frequency (percentages) for categorical variables. The mean values and percentages in different quintiles were compared using analysis of variance analysis and χ² tests.

CESD-10: centre of epidemiologic studies depression scale, 10-item version; CNY: China Yuan.

Among all observations, 14,800 (31.44%) were from western China, 15,714 (33.38%) from central China, and 19,230 (35.18%) from eastern China. Table S4 shows the monthly moving means (95% CIs) of ${\mathrm{PM}}_{2.5}$ , $\mathrm{BC}$ , $\mathrm{OM}$ , ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NO}}_3^{-}$ , and ${\mathrm{NH}}_4^{+}$ concentrations over the preceeding 1- to 10-year periods along with cognitive function scores. The mean (SD) cognitive function scores were 15.38 (4.49), 15.55 (4.55), 15.01 (4.69), and 15.93 (5.21) during waves 1, 2, 3, and 4 of CHARLS, respectively (Table S1).

Although air quality had been improving, nearly all enrolled participants lived in areas where ${\mathrm{PM}}_{2.5}$ , concentrations exceeded the WHO guidelines (Figure S2). Hotspots were widely distributed across central and eastern China (Figure S2).

Associations of ${\mathrm{PM}}_{2.5}$ and its major chemical components with cognitive function

Figure 2 and Table S2 show differences in global cognitive scores associated with each 10 μg/m³ increase in air pollutant exposure across multiple models. After adjusting for selected covariates, negative associations between cognition function scores and ${\mathrm{PM}}_{2.5}$ and its major chemical components were robustly observed. In the multivariable-adjusted model, each 1 SD increase in the concentrations of ${\mathrm{PM}}_{2.5}$ , $\mathrm{OM}$ , ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NO}}_3^{-}$ , ${\mathrm{NH}}_4^{+}$ , and $\mathrm{BC}$ was significantly associated with a decrease in the global cognitive score: β = −0.047 (95% CI: −0.064, −0.028) for ${\mathrm{PM}}_{2.5}$ ; −0.045 (95% CI: −0.059, −0.030) for BC; −0.052 (95% CI: −0.068, −0.036) for ${\mathrm{SO}}_4^{2-}$ ; −0.046 (95% CI: −0.067, −0.025) for ${\mathrm{NO}}_3^{-}$ ; −0.053 (95% CI: −0.071, −0.034) for ${\mathrm{NH}}_4^{+};$ and −0.046 (95% CI: −0.067, −0.025) for OM in the 1-year exposure window. To validate the robustness of the association between long-term exposure to ${\mathrm{PM}}_{2.5}$ and its major chemical components with cognitive function, 2-, 3-, 5-, and 10-year pollutant levels were considered. The results demonstrated that pollutant exposure was negatively correlated with cognitive function scores from 1- to 5-year windows, with an increasing trend (Table S4). Results from the RE models were consistent with those from the FE models. Compared with Model 1, Model 2 additionally adjusted for education and other sociodemographic variables, and the effect estimates remained robust. Results from multipollutant models remained largely consistent with those from single-pollutant models, suggesting that the observed associations were robust to coexposure adjustment.

Figure 2.

Estimated differences in global cognitive score associated with long-term exposure to PM_2.5 and its major chemical components. Colors represent different exposure timescales ranging from 6 months to 5 years. Model 1 was unadjusted; Model 2 was adjusted for age (continuous) and sex; Model 3 was adjusted for age, sex, urbanicity, education, solitary, smoking, drinking, social activity, night sleep duration, household income, history of diabetes, hypertension, dyslipidemia, heart disease and stroke, depressive symptoms, and subjective memory. *p < 0.05, **p < 0.01, ***p < 0.001.

Figure 3 demonstrates the nonlinear concentration–response (C–R) relationships between ${\mathrm{PM}}_{2.5}$ and its major chemical components and the difference in global cognitive function scores. The C–R curves for each major chemical component pollutant demonstrated similar shapes. These findings align with the results obtained from the linear mixed-effects model analysis, indicating notable differences in cognitive function scores associated with increases in pollutant concentrations. Higher pollutant concentrations are associated with greater reductions in cognitive function scores. This observation is consistent with the findings from the linear mixed-effects model analysis, confirming the detrimental impact of higher pollutant concentration on cognitive function.

Figure 3.

Differences in global cognitive score are associated with concentration–response associations with long-term exposure to PM_2.5 and its major chemical components. Solid lines in the upper panels indicate changes in global cognitive scores with 95% confidence intervals between them. Lower panels display the kernel density curves and boxplots of PM_2.5 and its major chemical components distribution.

The nonlinear C–R relationships depicted in Figure 3 provide a valuable visualization of the associations between ${\mathrm{PM}}_{2.5}$ and its major chemical components with cognitive function. They further support the conclusion that higher concentrations of these pollutants are associated with adverse effects on cognitive function.

In the exposure–weight analyses, the estimated weight ranking was ${\mathrm{SO}}_4^{2-}$  >  ${\mathrm{NH}}_4^{+}$  >  ${\mathrm{NO}}_3^{-}$  >  $\mathrm{BC}$  >  $\mathrm{OM}$ among PM_2.5 and its components by the random forest method, indicating that ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NH}}_4^{+}$ , and ${\mathrm{NO}}_3^{-}$ were the most critical components (Figure 4).

Figure 4.

Relative dominance effect of PM_2.5 and its major chemical components on cognitive function scores. The model was additionally adjusted for age, sex, urbanicity, education, solitary, smoking, drinking, social activity, night sleep duration, household income, history of diabetes, hypertension, dyslipidemia, heart disease and stroke, depressive symptoms, and subjective memory.

Figure 4 presents the results of the exposure–weight analyses, which used the random forest method to estimate the weight ranking of PM_2.5 and its major chemical components. The analysis revealed that ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NH}}_4^{+}$ , ${\mathrm{NO}}_3^{-}$ , $\mathrm{BC}$ , and $\mathrm{OM}$ were ranked in order of importance, with ${\mathrm{SO}}_4^{2-}$ receiving the greatest weight, followed by ${\mathrm{NO}}_3^{-}$ , ${\mathrm{ }\mathrm{NH}}_4^{+}$ , $\mathrm{BC}$ , and $\mathrm{OM}$ . The estimated weight ranking offers valuable insights into the relative contribution of each major chemical component to cognitive function. According to the results, ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NH}}_4^{+}$ , and ${\mathrm{NO}}_3^{-}$ were identified as the most critical components, suggesting that they may exert stronger influence on cognitive function than other pollutants.

Subgroup analyses

The regression model included interaction terms between exposure and stratified variables to conduct subgroup analyses. The variables considered for stratification were cognitive function dimension, age, sex, and subjective memory. The findings revealed an adverse effect on episodic memory function, primarily driven by negative associations on cognitive function and its major chemical components (Table S5).

Moreover, the findings indicated that the primary chemical constituents had adverse impact on cognitive abilities, particularly among individuals aged 45–60 years (Table S6). Table S7 displayed sex disparities, with females exhibiting greater susceptibility to exposure to ${\mathrm{PM}}_{2.5}$ and its major chemical components than males, indicating heightened cognitive vulnerability.

Furthermore, individuals with inadequate subjective memory appeared to exhibit greater vulnerability to the impacts of the major chemical components on cognitive performance, as evidenced in Table S8. The subgroup analyses offered valuable information regarding the possible variations in the impact of air pollution on cognitive function, depending on different demographic and personal factors.

Sensitivity analyses

To assess the reliability of the primary analysis findings, sensitivity analyses were performed, categorizing the data according to underlying health conditions that could potentially affect cognitive abilities. The stratification factors included the presence of depressive symptoms, high blood pressure, diabetes, abnormal lipid levels, and stroke.

In sensitivity analyses, we examined episodic memory and mental status scores separately; the results (Table S3) were consistent with our main findings. The findings from these sensitivity analyses, presented in Tables S9–S13, demonstrated that despite excluding individuals with cognitive impairment potentially associated with these coexisting conditions, the inverse association between exposure to PM_2.5 and its major chemical components and cognitive performance remained statistically significant. These findings imply that the association between exposure to air pollution and cognitive function was not exclusively influenced by the presence of these comorbidities.

The selection process of the penalty term λ for ridge regression and LASSO is shown in Figure S3. The ridge regression revealed that ${\mathrm{SO}}_4^{2-}$ , $\mathrm{ }\mathrm{BC}$ , ${\mathrm{NO}}_3^{-}$ , ${\mathrm{NH}}_4^{+}$ , and $\mathrm{OM}$ were ranked in order of importance. The LASSO regression revealed that ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NO}}_3^{-}$ , and BC were ranked in order of importance. The linear mixed-effects model regression revealed that $\mathrm{BC}$ , ${\mathrm{NO}}_3^{-}$ , ${\mathrm{NH}}_4^{+}$ , ${\mathrm{SO}}_4^{2-}$ , and $\mathrm{OM}$ were ranked in order of standardized coefficients (Figure S4). The statistical techniques employed in the sensitivity analyses arrived at comparable findings to those in the primary text, consistently indicating that ${\mathrm{SO}}_4^{2-}$ can exert the greatest impact on the relationship between PM_2.5 and its major components and cognitive function.

The E-values of Model 3 ranged from 37.9 to 43.8, with a minimum value of 37.9. This indicates that there is a small risk that our conclusions might be overturned due to unmeasured confounding. The E-values for ${\mathrm{PM}}_{2.5}$ , ${\mathrm{NH}}_4^{+}$ , ${\mathrm{NO}}_3^{-}$ , ${\mathrm{SO}}_4^{2-}$ , $\mathrm{BC}$ , and $\mathrm{OM}$ were 42.2, 43.8, 37.5, 37.5, 37.9, and 43.0, respectively. The C–R relationships of the 1-year average pollutant exposure and cognitive function are presented in Figure 3, and the associations appear to be approximately linear.

The sensitivity analysis findings further reinforced the strength of the main results and offered additional evidence for the detrimental influence of ${\mathrm{PM}}_{2.5}$ and its primary chemical constituents on cognitive function in middle-aged and older Chinese individuals.

Discussion

Improving air quality and creating a healthier living environment are widely recognized as important goals for society. However, achieving these goals while balancing economic and social considerations remains challenging. This study’s findings, which highlight the specific effects of major chemical composition of ${\mathrm{PM}}_{2.5}$ on health and their direct association with emission changes, may offer valuable insights into policymaking related to pollutant emission restrictions.

By identifying the specific chemical components of ${\mathrm{PM}}_{2.5}$ that have the greatest association with cognitive impairment, policymakers can develop more targeted strategies to reduce them. Such an approach can optimize resource allocation and minimize the environmental burden and economic costs associated with pollution control measures. By focusing on reducing the most hazardous components, policymakers can implement interventions that are likely to yield the greatest benefits in cognitive health and overall well-being.

Consistent with our findings, several epidemiological studies from Europe and North America have reported that ${\mathrm{SO}}_4^{2-}$ , a secondary inorganic aerosol, may exert stronger adverse effects on cognitive health than other PM_2.5 components. For example, higher ambient ${\mathrm{SO}}_4^{2-}$ concentrations have been associated with accelerated cognitive decline and an increased risk of dementia in older adults.^32,33 Mechanistic studies have further suggested that ${\mathrm{SO}}_4^{2-}$ particles can penetrate deeply into the respiratory tract, induce systemic oxidative stress and neuroinflammation, and impair cerebrovascular function, thereby contributing to neurodegeneration. These findings provide cross-validation for our observation that ${\mathrm{SO}}_4^{2-}$ exhibited the strongest association with reduced cognitive function in this nationwide Chinese cohort.

Incorporating the results of this study in policymaking may facilitate the development of more effective and efficient strategies for improving air quality and protecting public health. These findings emphasize the importance of regulating and reducing specific chemical components of ${\mathrm{PM}}_{2.5}$ to prevent cognitive decline, contributing to the formulation of evidence-based and targeted policies aimed at creating a healthier environment for all.

Comparison with other studies

To the best of our knowledge, this study is the first nationwide investigation to examine the associations between major chemical components ( $\mathrm{BC}$ , $\mathrm{OM}$ , ${\mathrm{SO}}_4^{2-}$ , ${\mathrm{NO}}_3^{-}$ , and ${\mathrm{NH}}_4^{+}$ ) of ${\mathrm{PM}}_{2.5}$ and cognitive function among middle-aged and older individuals in China. Although Liu et al. ³⁴ previously evaluated similar associations using the China Family Panel Studies (CFPS) cohort and highlighted the role of leisure-time physical activity as a modifier, our study advances the literature in several important ways. First, we used the CHARLS dataset, which provides more detailed cognitive assessments, repeated longitudinal measures, and a nationally representative sample of China’s middle-aged and older population. Second, we examined a broader range of PM_2.5 chemical components and applied FE models and machine-learning approaches, thereby providing complementary methodological insights. Together, these features highlight the added value and novelty of our study in understanding the relationship between PM_2.5 components and poor cognitive function. Numerous studies have reported strong correlations between cognitive function^35–37 and long-term pollutant exposure, potentially contributing to mild cognitive impairment ³⁸ or dementia. ³⁹ Our findings were consistent with the aforementioned studies, indicating that cognitive function in middle-aged and older individuals was inversely associated with pollutant concentrations. Furthermore, the findings indicated that major chemical components in air pollution exerted negative impacts on cognitive function in this longitudinal analysis. Even after adjusting for potential confounding factors, the detrimental impact of the major chemical components on cognitive function remained statistically significant. A growing body of epidemiological research has focused on investigating the impact of being exposed to major chemical components on cognitive function. For instance, a regional-level study conducted in north-central China reported significant negative correlations between cognitive function and concentrations of potassium and iron among adults aged 40 years and above. ⁴⁰ In Shanghai, a cohort study revealed adverse associations between the pollutant components and the infant’s cognitive and motor abilities, particularly in the problem-solving domains. ⁴¹ This study indicated that cognitive function is negatively affected by the chemical components of ${\mathrm{PM}}_{2.5}$ , albeit to varying degrees of significance. Taken together, these comparisons underscore that although prior studies such as Liu et al. ³⁴ offered important insights, our study provides additional value through the nationally representative CHARLS cohort, repeated cognitive assessments, and the integration of advanced modeling strategies, thereby strengthening the evidence base on PM_2.5 components and cognition in low- and middle-income country settings.

Potential mechanism

A comprehensive understanding of the mechanisms linking major chemical components of PM_2.5 and cognitive function remains incomplete. Consistent with our findings, several epidemiological studies from Europe and North America have reported that ${\mathrm{SO}}_4^{2-}$ may exert stronger adverse effects on cognitive health than other PM_2.5 components. For example, higher ambient ${\mathrm{SO}}_4^{2-}$ concentrations have been associated with accelerated cognitive decline and an increased risk of dementia in older adults.^32,33 Mechanistic studies further suggest that ${\mathrm{SO}}_4^{2-}$ particles can penetrate deeply into the respiratory tract, induce systemic oxidative stress and neuroinflammation, and impair cerebrovascular function, thereby contributing to neurodegeneration. These findings provide cross-validation for our observation that ${\mathrm{SO}}_4^{2-}$ exhibited the strongest association with impaired cognitive function in this nationwide Chinese cohort. BC and OM primarily originate from sources associated with combustion, including emissions from vehicles and combustion in industrial and agricultural sectors (Table S2). Anthropogenic fossil fuel combustion and natural sources produce secondary inorganic ions. ⁴² Once fine ambient particles infiltrate lung tissues and enter the bloodstream, they can trigger oxidative stress, provoke vascular inflammation responses, and disrupt various physiological processes. ⁴³ Ammonia plays a key role in neutralizing atmospheric nitric and sulfuric acids. ${\mathrm{NH}}_4^{+}$ , a neurotoxic agent, ⁴⁴ was reported to have a notable correlation with cognitive impairment. The presence of Aβ in the bloodstream can stimulate Aβ synthesis in the endoplasmic reticulum of astrocytes, particularly Aβ42, a major component of senile plaques observed in individuals with Alzheimer’s disease. ⁴⁵ Mechanisms ⁴⁶ may involve inflammation, oxidative stress, and mitochondrial dysfunction. This component is primarily derived from the photochemical transformation of precursor pollutants. Inhalation of airborne particles can induce oxidative stress and inflammation. ⁴⁷ A complex mixture of organic substances originates from primary organic carbon emitted directly into the atmosphere or from secondary organic carbon generated through gas-to-particle reactions. Exposure to organic components may increase the risk of cognitive impairment through pathways involving oxidative stress, endothelial cell damage, systemic inflammation, and abnormal lipid levels. ⁴⁸ The results support the need for strengthened control of transport-related emission sources. Future investigations are required to validate the impacts of various chemical constituents on cognitive function.

Explanation for the subgroup analysis results

Subgroup analyses suggested that age and sex can modify the relationship between cognitive function and major chemical components. Females and individuals aged 45–60 years experienced more pronounced effects. A clear sex disparity was observed, with women exhibiting greater vulnerability to pollutant-related cognitive impairment, consistent with previous evidence. ⁴⁹ Middle-aged adults also appeared more susceptible compared to older adults. One possible explanation is that individuals in this age group have higher exposure due to greater time spent in outdoor activities.

The stronger associations observed among females may be attributable to multiple mechanisms. Biologically, females may exhibit greater vulnerability to the neurotoxic effects of PM_2.5 due to hormonal influences and heightened inflammatory responses. Sociodemographic factors, including lower average education levels and reduced occupational outdoor exposure in the CHARLS female cohort, may also contribute to this disparity. Furthermore, the higher prevalence of chronic conditions, including anemia and depression, among females may further increase their susceptibility. These findings are consistent with those of prior studies reporting greater cognitive impacts of air pollution among females.

Clinical implications

The findings of this study have significant implications for public health and policymaking. First, identifying ${\mathrm{NH}}_4^{+}$ , ${\mathrm{NO}}_3^{-}$ , and ${\mathrm{SO}}_4^{2-}$ as the most critical components of ${\mathrm{PM}}_{2.5}$ associated with cognitive function provides valuable insights for the development of air pollution guidelines and regulations. Focusing on these specific components may allow the policymakers to prioritize strategies to reduce their emissions and mitigate their adverse effects on cognitive health.

Second, although the individual effect size of ${\mathrm{PM}}_{2.5}$ and its major chemical components on cognitive function may be relatively modest, the widespread nature of air pollution exposure makes this a significant public health concern. Given that the entire population is exposed to air pollutants to some extent, even a small effect can result in a substantial disease burden at the population level. Therefore, these findings highlight the need for comprehensive efforts to control and reduce harmful pollutants, including ${\mathrm{NH}}_4^{+}$ , ${\mathrm{NO}}_3^{-}$ , and ${\mathrm{SO}}_4^{2-}$ , to protect cognitive health at the population level.

Finally, the study emphasizes the importance of personal awareness and protection, particularly among middle-aged adults and females. Middle-aged individuals, who may be more vulnerable to the effects of air pollution, should monitor local air quality in their living environments and take necessary precautions to minimize exposure. Such measures may include choosing residences with lower levels of vehicle emissions and photochemical smog as well as adopting personal protective measures to mitigate the detrimental effects of air pollution on cognitive function.

Overall, this study highlights the need for continued efforts to reduce air pollution and mitigate its impact on cognitive health, with particular emphasis on ${\mathrm{NH}}_4^{+}$ , ${\mathrm{NO}}_3^{-}$ , and ${\mathrm{SO}}_4^{2-}$ as key components of ${\mathrm{PM}}_{2.5}$ . Implementing targeted interventions and policies may help reduce the risk of cognitive impairment and promote healthier aging among middle-aged adults and the older population.

Limitations and strengths

This study should be interpreted in light of several limitations. Furthermore, as an observational study, causal inference cannot be established, and residual confounding may persist despite extensive covariate adjustment; therefore, the results should be interpreted with caution. (1) Exposure assessment. The study relied on registered residential addresses to estimate exposure, which may not fully capture individual exposure patterns such as time spent outdoors or variations in breathing rates. One limitation of this study is that exposure assessment was conducted at the broad city level, which may introduce nondifferential misclassification. Due to ethical constraints, we were unable to obtain more detailed individual location data. However, given that air pollutants typically diffuse across urban areas and that relatively long exposure window was used, the long-term average exposure among residents within the same city is likely similar, limiting the risk of substantial misclassification. Furthermore, numerous validation studies have supported the feasibility of using city-level estimates to represent long-term population exposure levels. Individual-level exposure assessment would provide a more accurate estimation of personal exposure to air pollutants. Although PM_2.5 components are correlated with total PM_2.5 mass and co-pollutants, sensitivity analyses adjusting for these factors yielded consistent findings, suggesting that the observed associations are unlikely to be fully explained by such correlations (2) Spatial variation. Although satellite-based models were used to estimate air pollution levels, differences in exposure within the same city may still exist and were not completely captured. Fine-scale variations in pollutant concentrations driven by local emission sources and geographical factors may reduce the accuracy of exposure estimates. (3) Colocalization of noise and air pollution. Noise pollution often shares sources with air pollution and co-occurs in urban areas. However, this study did not include noise exposure data in its models, which could potentially confound the association between air pollution and cognitive function. (4) Biological mechanisms. The study acknowledges the lack of understanding of the specific biological mechanisms through which major chemical components of PM_2.5 affect cognitive impairment. Although animal-based studies provide insights into the general mechanisms of PM_2.5, further research is needed to elucidate the specific mechanisms associated with the major chemical components.

Despite these limitations, the study also has notable strengths. The use of a nationally representative cohort spanning multiple provinces and cities in China enhances the generalizability of the findings to populations experiencing high levels of air pollution. Furthermore, the high temporal (monthly) and spatial (10 × 10 km grids) resolution of the exposure data provides more accurate and detailed exposure measurements, strengthening the study’s validity. By acknowledging the limitations and the strengths, the study provides a balanced interpretation of the findings and identifies areas for further research and improvement in future studies.

Conclusion

Long-term exposure to PM_2.5 and certain chemical components, particularly ${\mathrm{SO}}_4^{2-}$ , may be associated with poor cognitive function in middle-aged and older adults in China. These findings highlight the potential importance of regulating specific PM_2.5 components to reduce the burden of cognitive impairment; however, further research is warranted to confirm these associations.

Supplemental Material

sj-pdf-1-imr-10.1177_03000605251406802 - Supplemental material for Associations of PM_2.5 and its major chemical components with cognitive function: A nationwide prospective cohort study among middle-aged and older adults in China

Supplemental material, sj-pdf-1-imr-10.1177_03000605251406802 for Associations of PM_2.5 and its major chemical components with cognitive function: A nationwide prospective cohort study among middle-aged and older adults in China by Shaomin Diao and Xiaoming Shen in Journal of International Medical Research