Heterogeneous Risks of Chemical Components and Sources in Size-Resolved Particulate Matter on Ischemic and Hemorrhagic Stroke

Abstract

Atmospheric particulate matter (PM) from different emission sources exhibits distinct size distributions and chemical compositions, which may differentially influence stroke risk. However, how particle size, chemical composition, and emission source jointly shape subtype-specific stroke risks remains unclear. In this study, 58,763 ischemic stroke (IS) and 2,588 hemorrhagic stroke (HS) hospitalizations were analyzed alongside 41 PM chemical components and 6 major emission sources identified across 9 aerodynamic size fractions below 10 μm. Associations between PM physicochemical properties and stroke were assessed in a time-series framework. Stroke risks exhibited pronounced heterogeneity across particle size, chemical composition, emission source, stroke subtype, and demographic groups. IS was associated with a broad range of components below 1.1 μm and above 3.3 μm, whereas HS showed stronger links to specific metals (As, Cu, Fe, Mn, Ni, Pb) and polycyclic aromatic hydrocarbons. Coal combustion (RR per IQR = 1.05; 95% CI: 1.01–1.10) and industrial emissions (RR per IQR = 1.18; 95% CI: 1.00–1.41) were the principal contributors to IS and HS, respectively. Individuals younger than 60 years showed increased HS risks related to most components in size <0.43 μm and industrial emissions. Females showed stronger susceptibility to Cu, traffic-related sources, and resuspended dust. Overall, our findings show that stroke risk varies according to particle size, chemical composition, and emission sources, with distinct patterns across stroke subtypes and population groups. This integrative understanding helps clarify the priority drivers of stroke risk and supports more strategically targeted air quality interventions.

1. Introduction

Stroke is one of the leading causes of death and disability worldwide, accounting for one in ten deaths in 2021. Environmental exposures play a critical role in shaping the global stroke burden. Among these, atmospheric particulate matter (PM) is a key environmental risk factor, responsible for nearly two million stroke deaths worldwide, highlighting a substantial and pervasive stroke burden linked to PM.

However, PM is a complex and heterogeneous mixture whose chemical composition varies substantially across particle size fractions. , Because physicochemical properties and toxic potentials are inherently size-dependent, the health effects of PM likely arise from the integrated influence of particle size and chemical constituents rather than from mass alone. Conventional PM mass concentration metrics fail to capture this critical heterogeneity as they do not reflect the joint variation in size distribution and composition that may differentially shape health risks. Previous studies have reported associations between specific PM components and stroke onset, and particle size has been shown to determine the location and efficiency of pulmonary deposition. Given that chemical constituents are not uniformly distributed across size fractions, the combined effects of composition and size, characterized through size-resolved components, are likely to be critical determinants of PM-related health outcomes. Nevertheless, establishing causal relationships between PM exposure and its size- and composition-specific health effects remains challenging.

Emission sources play a central role in shaping the physicochemical properties of PM and represent a primary lever for environmental management. Recent studies have examined wildfire-related PM_2.5 and its health effects. , Routine urban sources, including coal combustion, industrial emissions, and traffic, are the dominant contributors to air pollution in most cities. These sources generate PM with distinct characteristics: agriculture, mining, construction and road dust typically produce coarse particles rich in crustal components, whereas vehicle exhaust and the combustion of coal, fuel oil and wood primarily emit fine particles dominated by combustion-derived components. Although substantial evidence has documented the adverse health effects of PM_2.5, the links between size-dependent sources or constituents and stroke risks remain unclear.

To address this critical knowledge gap, we demonstrate pronounced heterogeneity in stroke risks across 9 aerodynamic size fractions, 41 chemical components, and 6 major emission sources in a Chinese megacity. The associations between particulate matter and stroke were not uniform but varied systematically according to the particle size, chemical composition, and emission source. Distinct patterns emerged between ischemic stroke (IS) and hemorrhagic stroke (HS), with specific components and source categories showing disproportionately elevated risks within particular size ranges. These findings demonstrate that particulate matter toxicity is governed by its size-specific physicochemical properties and source contributions, giving rise to heterogeneous impacts across stroke subtypes and susceptible populations. Together, these insights offer a scientific rationale for more targeted and health-centered air pollution mitigation strategies.

2. Methods

2.1. Study Population

This study included adult patients (aged ≥ 18 years) who were permanent residents of Tianjin, a Chinese megacity, and were admitted to one of five tertiary (Class III grade A) hospitals equipped with dedicated stroke units from January 1, 2018, to December 31,2020. Eligible patients presented with acute stroke symptoms and were hospitalized within 7 days of symptom onset. Diagnosis was confirmed through clinical assessment and neuroimaging, including computed tomography (CT) and/or magnetic resonance imaging (MRI). Patients were included if stroke was recorded as the primary cause of admission. Stroke types were classified into ischemic stroke (ICD-10 codes I63 and G45) and hemorrhagic stroke (ICD-10 codes I60–I61). The study population was further stratified by sex (male or female) and age (<60 years or ≥60 years).

2.2. Exposure Data

Size-resolved PM samples were collected across 9 aerodynamic diameter bins (<0.43 μm, 0.43–0.65 μm, 0.65–1.1 μm, 1.1–2.1 μm, 2.1–3.3 μm, 3.3–4.7 μm, 4.7–5.8 μm, 5.8–9.0 μm, and 9.0–10 μm) using a filter-based cascade impactor (Andersen Series, Thermo Fisher Scientific, USA) from May 2018 to December 2020. Sampling was conducted at the Comprehensive Atmospheric Environment Observation Station of Nankai University (Figure S1). The sampling period encompassed 270 effective days, capturing seasonal variations in atmospheric conditions, as well as episodes of heavy pollution. The chemical analyses were performed following the procedures described in our previous studies. , QA/QC procedures and details of instruments and analytical methods are described in our previous study , and summarized in Text S1–S2 and Table S1. Concentrations of four gaseous air pollutants (O₃, SO₂, NO₂, CO) were obtained from Tianjin Eco-Environmental Monitoring Center (https://air.cnemc.cn:18007/). The locations of ground-based air quality monitoring stations are shown in Figure S1. Meteorological data including the daily mean temperature, mean relative humidity and pressure were obtained from the fifth generation ECMWF (European Centre for Medium-Range Weather Forecasts) atmospheric reanalysis hourly climate data set (https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5). Details of the aggregation and processing of exposure data are provided in Text S3.

2.3. Source Contribution to Fine and Coarse PM

The Positive Matrix Factorization (PMF) model estimates the contributions of different sources to particulate matter mass concentrations by exploiting differences in their chemical composition profiles, and is widely recognized as an efficient source apportionment tool. , In our study, based on prior knowledge, , different source profiles were identified for fine and coarse PM. Fine PM was defined as <0.43 μm, 0.43–0.65 μm, 0.65–1.1 μm, and 1.1–2.1 μm, whereas coarse PM was categorized as 2.1–3.3 μm, 3.3–4.7 μm, 4.7–5.8 μm, 5.8–9.0 μm, and 9.0–10 μm. The PMF analysis was conducted separately for fine and coarse PM, with the number of sources set to six and identified as follows: coal combustion, gasoline vehicles, diesel vehicles, secondary aerosol, industrial emissions, and fugitive dust. The basic model in our study is specified as

$$X_{i,j,k}=\sum _{s=1}^F{g}_{i,s,k}{f}_{s,j,k}+e_{i,j,k}$$ 1

In eq , X _i,j,k represents the total concentration of i-th PM sample, j-th species and k size bins; g _i,s,k also referred to as the G matrix, denotes the concentration of the s-th source for the i-th sample and k-th PM size bin (unit: μg/m³) ; f _s,j,k also referred to as the F matrix, represents the fraction of the j-th species in the s-th source for the k-th PM size bin, and is constrained to be non-negative. where F denotes the total number of sources. The optimal number of factors (F) for both fine and coarse particles was determined by evaluating Q/Qexp values, the interpretability of factor profiles, Bootstrap (BS) stability, and displacement (DISP) analysis. PMF results with F = 6 for both fine and coarse particles showed stable BS mapping (≥80% factor matching) and the absence of significant factor swaps in DISP diagnostics. Detailed procedures and results for factor number selection are provided in Text S4.

The PMF model input includes two files: chemical species concentration data and corresponding uncertainty data. The uncertainty was calculated by using the following formula:

$$u_{i,j,k}=\{\begin{array}{ll}\frac{5}{6}\times \mathrm{MDL},& c_{i,j,k}\le \mathrm{MDL}\\ \sqrt{{(\mathrm{Error}\mathrm{Fraction}\times {\mathit{c}}_{\mathrm{i},\mathrm{j},\mathrm{k}})}^2+{(0.5\times \mathrm{MDL})}^2},& c_{i,j,k}>\mathrm{MDL}\end{array}$$ 2

In eq , u _i,j k represents the uncertainty of i-th sample and j-th species for a k size bin; c _i,j,k is the concentration of chemical species, and MDL is the method detection limit for each species. All 41 components with the total PM concentration were input in the PMF model. Both the fine PM PMF model and the coarse PM PMF model were run 20 times, and the final results converged in both cases.

2.4. Modeling Size-Resolved Health Effects of Particulate Matter Components

A two-stage statistical framework , was used to assess the association between size-specific PM components and stroke hospital admissions. In this time series study, each component was modeled separately using the same analytical procedure.

In Stage 1, a time-series linear regression was conducted to model the relationship between the total PM mass concentration and the corresponding component concentration. This step aims to isolate the component-specific variation that is independent of the overall PM mass. The formula is as follows:

$$\mathrm{P}{\mathrm{M}}_t={\epsilon }_0+a_{i,j}\mathrm{com}{\mathrm{p}}_{i,j,t}+\mathrm{residua}{\mathrm{l}}_{i,j,t}$$ 3

In eq , PM _t represents the daily total PM mass concentration at day t; ε ₀ is the intercept of the model; a _i, j is the linear regression coefficient of j-th component within the i-th size fraction; comp _i, j, t refers to the mass concentration of the j-th component within the i-th PM size fraction at day t; residual _i, j, t denotes the residuals obtained from the regression of PM on the component. The residual term captures the portion of temporal variation in total PM mass that cannot be explained by the specific size-resolved component.

In Stage 2, a quasi-Poisson generalized additive model was fitted with daily stroke admission counts as the dependent variable. Based on a time-series study design, this model quantifies the short-term associations between size-resolved PM components and daily stroke admissions while controlling for temporal trends and potential confounders, including long-term time trends, temperature, relative humidity, day-of-week effects, and lockdown periods. Size-specific component concentrations were included as exposure variables through a linear term, while the corresponding residuals were added as an additional linear term to account for potential confounding by other PM components. Daily stroke admissions were aligned one-to-one with days covered by size-resolved PM data based on the calendar date. For days without PM sampling data, exposure concentrations were recorded as missing (NA). Stage 2 allows for the removal of extraneous variation attributable to other PM constituents, thereby isolating the independent effect of the component of interest. The formula is as follows:

$$\log (\mathit{Y})={\beta }_{int}+{\beta }_{i,j}\mathrm{com}{\mathrm{p}}_{i,j}+\mathrm{residua}{\mathrm{l}}_{i,j}+s(\mathrm{temp})+ns(\mathrm{RH},df=3)+ns(\mathrm{time})+\mathrm{DOW}+\mathrm{LOCKDOWN}$$ 4

In eq , the Y represents the daily counts of hospital admissions. The intercept term is denoted by β _int and comp _i, j refers to the mass concentration of the j-th component in the i-th size fraction of particulate matter. The term residual _i, j represents the residuals from the regression of each component against the total PM mass concentration in eq . Temperature, relative humidity, long-term time trends, day-of-week effects, and city lockdown status were included in the model as potential confounding factors. s(temp) is a smooth spline function of temperature, allowing for nonlinear effects, while ns­(RH,df = 3) represents a natural spline for relative humidity with 3 degrees of freedom. Additionally, ns (time) controls for temporal trends, using natural cubic splines with 7 degrees of freedom per year. DOW (day of the week) was included to account for weekly variation in hospital admission patterns and time-varying activity patterns. LOCKDOWN was defined as a binary indicator (0 = nonlockdown period; 1 = lockdown period) to adjust for potential changes in air pollution levels, population mobility, and healthcare-seeking behavior during the COVID-19 pandemic. ,

This two-stage modeling framework minimizes the influence of other particulate matter components while retaining the original mass concentration units of each component. Compared with alternative metrics such as component proportions, it allows for more straightforward and interpretable estimation of component–health associations.

2.5. PM Source Health Effects Modeling

Source-specific contributions to PM were estimated using positive matrix factorization (PMF) based on chemical composition profiles. The apportioned concentrations of each source were introduced into a GAM model as exposure variables. The health effects of PM sources are estimated as follows:

$$\log (\mathit{Y})={\beta }_{int}+{\beta }_{i,s}\mathrm{sourc}{\mathrm{e}}_{i,s}+s(\mathrm{temp})+ns(\mathrm{RH},df=3)+ns(\mathrm{time})+\mathrm{DOW}+\mathrm{LOCKDOWN}$$ 5

In eq , the Y represents the daily counts of hospital admissions. source _i, s refers to the mass concentration of source s in the i-th size fraction. The intercept term, confounding factors, and long-term trends, along with day of the week and lockdown status, remain the same as those in eq . Results are expressed as RRs of stroke admissions per IQR increase in PM source concentration.

Results were expressed as relative risks (RRs) of stroke admissions per interquartile range (IQR) increase in the exposure concentration. Relative risk was calculated as follows:

$$\mathrm{RR}=\exp (\beta •c)$$ 6

In eq , β was the log–linear coefficient fitted by eq and eq , c denotes the interquartile range of the specific exposure concentration, including size-resolved components or size-resolved sources. An RR > 1 indicates an increased risk of stroke admissions associated with an IQR increase in the concentration of specific components or sources, whereas an RR < 1 indicates a decreased risk.

2.6. Sensitivity Analysis

Sensitivity analyses were performed for both component- and source-based models. These included further adjustment for gaseous pollutants (SO₂, NO₂, CO, and O₃) and changing the degrees of freedom for confounding variables. These analyses were conducted to assess the robustness of the main findings. Statistical analyses were performed in R (version 4.1.2), and source apportionment was conducted using EPA PMF 5.0.14.

3. Results

3.1. Health and Exposure Data for PM, Components, and Sources

During 2018–2020, we collected 58,763 records of IS and 2,588 records of HS. The majority of IS and HS patients were male and aged over 60 years (Table ). Summary statistics for PM mass concentrations across 9 size fractions and meteorological variables are provided in Table S2. Concentrations of 41 components spanning seven chemical groups in total PM10 are summarized in Table . Carbonaceous components and secondary inorganic ions dominated the PM mass, followed by natural components and halogen ions. Potentially toxic elements (e.g., As and Cr) and PAHs occurred at relatively low concentrations. Source apportionment identified six sources across different particle size fractions. Profiles of source components for fine (PM_2.1) and coarse (PM_2.1–10) particles are shown in Figure S2. Secondary aerosol contributed the largest proportion to fine PM, whereas coarse PM was mainly influenced by fugitive dust and industrial emissions (Figure S3). Average source contributions across size fractions are presented in Table S3. Performance indicators confirmed the robustness of the PMF models (Table S4). Spearman correlation analysis (figure S4) showed generally weak linear associations between environmental variables and stroke admissions (|r| < 0.2). Moderate correlations were observed among meteorological factors and air pollutants, supporting the necessity of adjusting for meteorological conditions in subsequent analysis. Coal combustion showed negative correlations with other sources, as determined by (Figure S5).

1. Daily and Total Ischemic (IS) and Hemorrhagic Stroke (HS) Admissions during the Study Period.

Stroke subtype	Subgroup	Mean ± SD (per day)	Median (P25, P75) (per day)	Total admissions (N)
IS	Age > 60	36 ± 11	37 (28, 44)	39,448
	Age ≤ 60	18 ± 6	18 (13, 22)	19,315
	Male	34 ± 11	35 (26, 42)	37,393
	Female	19 ± 7	20 (15, 25)	21,370
HS	Age > 60	1 ± 1	1 (0, 2)	1,329
	Age ≤ 60	1 ± 1	1 (0, 2)	1,259
	Male	2 ± 1	1 (1, 2)	1,686
	Female	1 ± 1	1 (0, 1)	902

2. Daily Average Concentrations of 41 PM Chemical Components Aggregated across 9 Particle Size Fractions during the Study Period.

Component type	Chemicals(abbreviation)	Mean concentration in PM10 (μg/m3)
carbonaceous components	Organic carbon(OC)	38.3
carbonaceous components	Elemental carbon(EC)	12.6
secondary inorganic ions	Nitrate(NO3 –)	17.7
secondary inorganic ions	Sulfate(SO4 2–)	10.4
secondary inorganic ions	Ammonium(NH4 +)	11.4
halogen ions	Chloride(Cl)	4.25
halogen ions	Fluoride(F)	0.715
halogen ions	Bromine(Br)	0.127
natural components	Aluminum (Al)	10.8
natural components	Titanium(Ti)	0.156
natural components	Sodium(Na)	2.41
natural components	Potassium(K)	0.809
natural components	Magnesium(Mg)	1.16
natural components	Calcium(Ca)	8.13
potentially toxic elements	Arsenic(As)	6.08 × 10–2
potentially toxic elements	Cadmium(Cd)	1.84 × 10–3
potentially toxic elements	Cobalt(Co)	7.39 × 10–3
potentially toxic elements	Chromium(Cr)	0.197
potentially toxic elements	Copper(Cu)	0.334
potentially toxic elements	Iron(Fe)	6.01
potentially toxic elements	Manganese(Mn)	0.139
potentially toxic elements	Nickel(Ni)	0.134
potentially toxic elements	Lead(Pb)	0.100
potentially toxic elements	Vanadium(V)	1.28 × 10–2
potentially toxic elements	Zinc(Zn)	0.544
low molecular weight PAHs	Naphthalene(Nap)	8.85 × 10–4
low molecular weight PAHs	Acenaphthylene(Acy)	6.41 × 10–4
low molecular weight PAHs	Acenaphthene(Ace)	1.44 × 10–3
low molecular weight PAHs	Fluorene(Flu)	1.27 × 10–3
low molecular weight PAHs	Phenanthrene(Phe)	2.54 × 10–3
low molecular weight PAHs	Anthracene(Ant)	1.14 × 10–3
medium molecular weight PAHs	Fluoranthene(Fla)	2.24 × 10–3
medium molecular weight PAHs	Pyrene(Pyr)	2.08 × 10–3
medium molecular weight PAHs	Benzo[a]anthracene(BaA)	1.69 × 10–3
medium molecular weight PAHs	Chrysene(Chr)	2.07 × 10–3
high molecular weight PAHs	Benzo[b]fluoranthene(BbF)	2.23 × 10–3
high molecular weight PAHs	Benzo[k]fluoranthene(BkF)	1.41 × 10–3
high molecular weight PAHs	Benzo[a]pyrene(BaP)	1.68 × 10–3
high molecular weight PAHs	Indeno[1,2,3-cd]pyrene(IcdP)	3.91 × 10–3
high molecular weight PAHs	Dibenzo[a,h]anthracene(DahA)	3.78 × 10–3
high molecular weight PAHs	Benzo[g,h,i]perylene(BghiP)	2.69 × 10–3

3.2. Effects of Size-Resolved PM Components on Ischemic and Hemorrhagic Stroke

Figure shows the RRs of components in 9 sizes for IS and HS hospitalizations. The maximum RR was observed for Cr in PM_9–10 (1.02, 95% CI: 0.98,1.07; per IQR increase) in IS and for BaP in PM_0.43 (1.19, 95% CI: 1.01, 1.42; per IQR increase) in HS. For IS, some inorganic ions, natural components, OC, EC and potentially toxic elements showed slightly high RRs at sizes <1.1 and >3.3 μm, and medium/high-molecular-weight PAHs showed relatively high RRs in most sizes. For HS, the As, Cu, Fe, Mn, Ni, Pb (at sizes <1.1 μm), and all measured PAHs showed high RRs in most sizes, and the risks of more components (including many halogen ions, potentially toxic elements, and all PAHs) were stronger at sizes <0.65 μm than in other sizes. The sensitivity analysis results for both gas pollutants adjusted and the changing degree of freedom of confounders are in line with the main results (Tables S5–S16).

1.

(A) Average proportions of 41 chemical components across 9 PM size fractions. (B–C) Relative risks of PM components across nine size fractions for ischemic (B) and hemorrhagic stroke (C). Relative risks (RRs) represent the change in stroke risk associated with an interquartile range (IQR) increase in each component. Size fractions 1–9 correspond to aerodynamic diameters of 0–0.43, 0.43–0.65, 0.65–1.1, 1.1–2.1, 2.1–3.3, 3.3–4.7, 4.7–5.8, 5.8–9.0, and 9.0–10 μm. Bubble size reflects the absolute excess risk, calculated as the absolute percent change in relative risk (|RR – 1| × 100%). Larger bubbles indicate greater effect magnitude, independent of whether the association is positive or negative.

The age- and sex-stratified RRs of components for IS are shown in Figure A and B. In age-stratified analyses, individuals younger than 60 years exhibited higher IS risks associated with exposure to secondary inorganic ions at sizes <1.1 μm and >4.7 μm, several natural components and potentially toxic elements across most sizes, and some medium/high-molecular-weight PAHs at sizes <1.1 μm and 4.7–5.8 μm. The highest RR in this group was observed for Mg at 9.0–10 μm (1.07, 95% CI: 1.00,1.14; per IQR increase). For individuals aged 60 years or older, stronger associations were limited to high-molecular-weight PAHs at sizes >0.65 μm, with the highest RR for BghiP at 0.65–1.1 μm (1.03, 95% CI: 0.98,1.08; per IQR increase). Overall, younger individuals appeared sensitive to a broader range of components and smaller particle sizes (<0.65 μm). In sex-stratified analyses, SO₄ ^2– at sizes <0.65 μm, Ca²⁺ at <0.65 μm and 4.7–5.8 μm, and high-molecular-weight PAHs at <5.8 μm were associated with increased IS risk in males, with the highest RR for IcdP at 0.65–1.1 μm (RR 1.03, 95% CI:0.97,1.09; per IQR increase). Females showed heightened sensitivity to a wider range of components, including natural components and potentially toxic elements at sizes >2.1 μm, and OC and EC across most sizes, suggesting that fugitive dust in larger particles (>2.1 μm) and combustion-related components may have stronger impacts in females than in males.

2.

Age- and sex-specific relative risks of hospitalization for ischemic and hemorrhagic stroke associated with particulate matter components. (A, B) Ischemic stroke and (C, D) hemorrhagic stroke relative risks associated with an interquartile range (IQR) increase in PM components across particle size fractions.

The RRs of components in each size for age- and sex-stratified HS are shown in Figure C and D. For individuals younger than 60 years, most PAHs (except for IcdP and DahA) in all sizes showed high associations with HS, with the maximal RR for BaP at 4.7–5.8 μm (1.52, 95% CI: 1.06, 2.20; per IQR increase). Some natural components and potentially toxic elements in specific sizes, such as Cr in 5.8–9.0 μm, also showed high RR. It is notable to find that most components at <0.43 μm showed high RRs on HS of individuals younger than 60 years. Consistently with the IS, HS for individuals aged 60 years or older showed weak links with most components, except that some PAHs at 0.43–0.65 μm as well as some secondary inorganic ions and natural components at 9–10 μm showed slightly high risks. The sex-stratified RR for HS were generally consistent for females and males that they were linked with many HPAHs in small sizes and many MPAHs in coarse sizes. Difference can be observed at <0.65 μm that males were more sensitive to the As and SO₄ ^2– which may be associated with coal combustion, while females were more sensitive to the Mg, Ti and Cu which may be associated with fugitive dust.

3.3. Effects of Size-Resolved Sources on Ischemic and Hemorrhagic Stroke

The RRs of sources for IS and HS across 9 size fractions (Figure ) indicated that the highest RR were 1.06 (95% CI: 1.01, 1.10; per IQR increase) for coal combustion in ischemic stroke and 1.18 (95% CI: 1.00, 1.41; per IQR increase) for industrial emissions in HS. It is consistent that the increase in industrial emissions exhibited relatively high effects and secondary aerosol showed low risks on most stratifications. Coal combustion showed moderate RRs (ranging from 1.01 to 1.05) for ischemic stroke in most sizes, while it showed no risk on HS at 0.65–1.1 μm, 1.1–2.1 μm, and 9.0–10.0 μm, but high risk (higher than 1.10) at 0.43–0.65 μm and 2.1–3.3 μm. Fugitive dust showed no risk for ischemic stroke, but high risk on HS at <1.1 μm. Consistently with the RRs of components, HS was sensitive to more sources at <0.43 μm. The sensitivity analysis of the associations between PM sources and stroke admissions further supported the results presented above (Tables S17–S28).

3.

Stroke risk estimates associated with particulate matter from different sources across size fractions. Relative risks for ischemic and hemorrhagic strokes are estimated for each particle size fraction, representing the risk associated with an interquartile range (IQR) increase (μg/m³) in source-specific PM concentrations.

The RRs of PM sources in each size for age- and sex-stratified IS are shown in Figure A–F. Age-stratified IS showed that, for individuals younger than 60 years, exposure to coal combustion at >2.1 μm, industrial emissions in most sizes, fugitive dust in most sizes and secondary aerosol at <0.43 μm showed higher risks for IS. For individuals aged 60 years or older, IS showed stronger associations with coal combustion in most sizes, industrial emissions at sizes <1.1 and >9.0 μm, and traffic emissions >3.3 μm. By comparison, younger individuals were more sensitive to industrial emissions and fugitive dust, while older individuals were more sensitive to traffic emission, which is consistent with the finding that younger individuals were sensitive to more potentially toxic elements while older individuals were sensitive to low-molecular-weight PAHs. When comparing sex-stratified IS, industrial emissions and coal combustion showed high RRs in both females and males, while females were more sensitive to traffic emissions and coarse fugitive dust, which is consistent with the finding that females were more sensitive to natural components and potentially toxic elements at >2.1 μm, and OC and EC in most sizes than males.

4.

Percent changes in ischemic (IS) and hemorrhagic stroke (HS) admissions are estimated for an interquartile range (IQR) increase in PM from six sources across size fractions.(A–F) IS: (A) Coal combustion, (B) Gasoline vehicles, (C) Diesel vehicles, (D) Secondary aerosol, (E) Industrial emissions, and (F) Fugitive dust. (G–L) HS: (G) Coal combustion, (H) Gasoline vehicles, (I) Diesel vehicles, (J) Secondary aerosol, (K) Industrial emissions, and (L) Fugitive dust.

Figure G–L shows the RRs of sources in each size for age- and sex-stratified HS. For individuals younger than 60 years, coal combustion at <0.43 and >2.1 μm, industrial emissions in most sizes, traffic emissions at >5.8 μm, and fugitive dust at <2.1 μm showed high risks for HS. Compared with younger individuals, HS among older individuals showed weaker associations with fugitive dust but stronger associations with traffic emissions in fine sizes, which was in agreement with the age-stratified IS. For sex-stratified HS, industrial emissions still showed a relatively strong adverse effect, and females were more sensitive to traffic emissions and fugitive dust.

4. Discussion

4.1. Main Conclusions and Biological Mechanisms

To our knowledge, this is the first comprehensive assessment of the effects of chemical components and source contributions across 9 size fractions on stroke. We found that certain potentially toxic elements and PAHs, although contributing only a small fraction of PM mass, were associated with higher health risks than some components with larger mass contributions, particularly in the case of HS. Previous studies have reported significant associations between stroke mortality and exposure to major PM_2.5 components, including black carbon (BC), organic matter (OM), and sulfate (SO₄ ^2–), for both short-term and long-term exposure. − Our findings add to the body of evidence by highlighting that the health impacts of components are not solely driven by mass concentration but also by the intrinsic toxicity.

In addition, given that the physicochemical and toxicological characteristics of PM vary by source, understanding these differences is crucial for optimizing cost-effective prevention measures. In the present study, industrial emissions generally exhibited high risks, despite contributing lower fractions of PM mass. Coal combustion, traffic emissions, and fugitive dust showed adverse impacts in some sizes on specific populations, while secondary aerosol (which contributed higher fractions of PM mass) showed relatively low risks for both IS and HS. Although certain source categories share similar labels across particle modes, they were identified independently in the fine and coarse PMF analyses and should not be interpreted as physically identical sources. Our findings are consistent with previous research. A case-crossover study conducted in New York State found that spark-ignition emissions (identified as gasoline vehicles in our study) in PM_2.5 were positively associated with IS hospitalizations, whereas secondary components showed no adverse effects. A time series study in London found that total PM_2.5 and PM_2.5 from exhaust (identified as gasoline and diesel vehicles in our study) are positively associated with total anterior circulation infarct (TACI) incidence, which is a subtype of IS.

Although the biological mechanisms behind the associations between PM component exposure and stroke are still not clear, some mechanisms have been proposed. An adverse outcome pathway (AOP)-based framework has outlined the potential cerebrovascular effects of air pollutants. It traces a sequence from molecular initiating events, including reactive oxygen species generation and sensory receptor activation, to key cellular and organ-level processes such as endothelial dysfunction, increased blood–brain barrier permeability, and atherosclerosis, which ultimately result in adverse cerebrovascular outcomes. A Panel Study in Guangzhou found potentially toxic metals such as Pb were related to decreased high-density lipoprotein cholesterol (HDL-C) levels, a known risk factor for IS incidence. Heavy metals such as cadmium have a long biological half-life in the human body and may contribute to the development of cerebrovascular disease. A time-series study conducted in 184 Chinese cities quantified exposure to seven PAHs (including BaA, Chr, BbF, BkF, BaP, DahA, and IcdP) and found that short-term exposure to these PAHs was associated with increased hospital admissions for IS. The same body of work also reported experimental evidence from animal studies showing that PAHs can induce inflammatory responses and oxidative stress, potentially via the phosphoinositide 3-kinase/Akt (PI3K/Akt) and mitogen-activated protein kinase (MAPK) signaling pathway. Such mechanisms may account for the greater stroke risks observed for potentially toxic elements and PAHs, whereas the increased risks from industrial emissions point to the potential importance of metals.

4.2. Effect Modification by Stroke Subtype, Age, and Sex

Differences were observed across stroke types and in age- and sex-stratified analyses. These disparities may arise from differences in pathophysiological mechanisms underlying specific stroke types and from heterogeneous exposure driven by population-specific time-activity patterns. In the present work, the IS was sensitive to more components (some ions, natural components, OC, EC, potentially toxic elements, and PAHs) but with lower RRs, while the HS was sensitive to fewer components (As, Cu, Fe and most PAHs) but with higher RRs. IS is more strongly influenced by processes such as atherosclerosis progression, plaque rupture, and thrombosis, whereas HS is more closely linked to elevated blood pressure and vascular fragility. PM components (such as metals and PAHs) induced chronic inflammation, reactive oxygen species generation, endothelial dysfunction, and pro-thrombotic states are known to promote atherosclerosis and thrombus formation, thereby increasing the risk of ischemic events. While the evidence for HS remains inconsistent: findings from a multicity study in 26 Chinese cities reported no associations between short-term exposure to PM_2.5 mass concentration and HS admissions; in contrast, a case-crossover study in Taiwan, China, observed a positive association between PM_2.5 exposure and HS admissions. A cohort study in China reported that higher concentrations of PM_2.5 and its chemical constituents were associated with elevated systolic blood pressure (SBP) and an increased risk of hypertension, which are risk factors for HS. These discrepancies underscore the need for further analyses focusing on PM components to elucidate the potential drivers of HS risk.

Consistent for both IS and HS, individuals younger than 60 years appeared to be more sensitive to the acute effects of a wider range of PM components and sizes and older individuals showed weaker associations with fugitive dust but stronger associations with traffic-related sources. One possible explanation relates to differences in outdoor exposure patterns. The Exposure Factors Handbook of Chinese Population reports that younger adults typically spend longer durations outdoors compared to individuals aged older than 60 years, potentially resulting in greater exposure to ambient particulate matter. Literature has demonstrated that traffic-related pollutants are more readily transported into indoor microenvironments due to their properties and the close proximity. A time series study also found that road dust was more strongly associated with nonaccidental hospital admissions in the 0–15 years age group compared to the 65–74 years group, which is partly because children are more susceptible to road dust and tend to spend more time outdoors. A double-blind, randomized crossover study reported that inhalation of PM_2.5 increased diastolic blood pressure in healthy adults aged 18–50 years. Similarly, a case-crossover study in Israel observed that young adults experienced a higher risk of IS associated with PM exposure.

Furthermore, for both IS and HS, females were more sensitive than males to traffic emissions and fugitive dust and showed greater susceptibility to copper (Cu). Existing evidence suggests that certain PM components may disrupt estrogen signaling and induce oxidative stress, thereby contributing to adverse alterations in blood lipid profiles. , A recent review on sex differences in air pollution health effects reported that, following PM exposure, females experience reduced estrogen levels and greater PM deposition in the lungs compared with males. Copper, as a redox-active transition metal, exhibits strong oxidative potential, which may underlie these sex-specific biological responses.

4.3. The Impacts of PM Physicochemical Properties on Stroke Risks

PM physicochemical properties may jointly influence the cerebrovascular effects of PM. Interestingly, in this study, we found that most components in <0.43 μm showed high RR in individuals younger than 60 years, while many showed low risks in other size ranges, suggesting a different underlying mechanism (Figure ). Previous studies have reported stronger health effects associated with smaller PM compared to larger particles, − attributing to the possibility that smaller particles may carry more toxic components, or penetrate more efficiently into the terminal airways and even the circulatory system. Our study offers a unique opportunity to explore this hypothesis using size-resolved component data. The averaged chemical composition (Figure A) showed a difference among 9 sizes. Some components (such as EC, As, Co, Low molecular weight PAHs, DahA, BghiP, etc.) showed higher abundances in PM_0.43, while some (such as Al, Cu, Fe, Mn, etc.) were lower than in coarse PM. As shown in Figure , more components in PM_0.43 showed high HS admission RR than in other size ranges, indicating that PM-related health risks may be primarily driven by particle size in PM_0.43, while in other size fractions, chemical composition may play a more dominant role in determining health effects. A fraction of PM_0.43 is likely to penetrate deep into the lungs and directly enter the bloodstream, thereby exerting toxic effects on the cerebrovascular system. Younger individuals may exhibit heightened susceptibility to ultrafine particles, potentially due to their higher capacity for pulmonary gas exchange and faster blood circulation.

5.

(A) Normalized contributions of chemical components across nine size-resolved fractions. (B) Comparison of cumulative risks of components across the nine fractions with direct risk estimates based on total PM10 concentrations. RRs represent the risk of hemorrhagic stroke in individuals aged <60 years associated with an increase in each component’s concentration, assuming PM₁0 increases by 10 μg/m³ and the component rises proportionally to its mean contribution to PM₁₀.

To evaluate the overall stroke risk attributable to 9 size fractions and to highlight the necessity of distinguishing particle size, we calculated the RR for HS using two approaches: (1) direct estimation based on the total concentration of each component in PM₁₀, obtained by summing concentrations across all 9 size fractions, and (2) a cumulative risk estimation, derived by summing the risks from each individual fraction. Details of the calculation are provided in Text S5. The results are shown in Figure B. Notably, for certain potentially toxic elements such as As, Cr, and Pb, as well as for nearly all of the PAHs, the cumulative risks exceeded the directly estimated risks, underscoring the critical role of size-resolved components in PM risk assessment.

The health effects of individual PM components remain insufficiently understood due to limited evidence, and existing studies often focus on components with high mass fractions. In our study, we estimated stroke admission risks for a total of 9 size bins and 41 components representing different chemical types, enabling a direct comparison of health effects across components and sizes. Furthermore, we conducted an additional analysis to assess how species composition modifies PM-related health risks. Using k-means clustering, three distinct chemical composition profiles were identified for both PM_0.43 and PM_4.7–5.8 (details in Text S6). The normalized chemical composition profiles are shown in Figure S6A, with cluster 3 characterized by higher proportions of potentially toxic elements and high-molecular-weight PAHs. Figure S6B shows that the proportions of the three composition clusters in PM_0.43 and PM_4.7–5.8 were comparable. Figure S6C presents the PM-induced risk of HS across three composition clusters and two size fractions. Despite having a similar chemical composition, smaller particles were linked to an elevated risk of HS, further underscoring the critical role of particle size in the risk assessment of PM and its chemical constituents.

We also observed clear differences in PM composition and source contributions between clean and haze conditions. Following the national air quality standard, we stratified the study period using 75 μg/m³ as the threshold for daily PM_2.5 concentrations, classifying days into clean (<75 μg/m³) and haze (≥75 μg/m³) categories. As shown in Figure S7, haze days were characterized by a higher relative proportion of water-soluble ions (SO₄ ^2–, NO₃ ^–, and NH₄ ⁺), indicating enhanced secondary formation processes. Consistently, source apportionment analysis demonstrated increased contributions from secondary sources during haze episodes­(Figure S8).

We then conducted stratified GAM analyses to estimate associations between IS admissions and PM components and sources under clean and haze conditions. The results (Table S29) indicate that effect estimates for most components were elevated during haze days, suggesting stronger health impacts under high-pollution scenarios. Interestingly, although secondary sources contributed a larger fraction of PM mass during haze days, their associated health risks did not increase proportionally. In contrast, coal combustion–related sources exhibited amplified health effects during haze conditions­(Table S30). This finding highlights that increased mass contribution does not necessarily translate into proportionally higher toxicity and underscores the importance of source-specific control strategies in health warning systems.

4.4. Strengths and Limitations

Our study has several strengths. First, to our knowledge, this is the first investigation to examine the associations between PM components and sources with stroke across 9 size fractions. Although many studies have attributed the elevated health risks of smaller particles to their larger specific surface area, which may enable them to carry a higher relative load of toxic components, our findings suggest a more dynamic framework. In particular, even for the same type of chemical component, PM_0.43 exhibited higher stroke risks in our cluster-based modification analysis, indicating a transition in relevance from particle size alone toward the interplay between size and chemical composition, which may be more critical in determining cerebrovascular risks. Second, we explored the role of population-level effect modifiers in shaping the association between PM exposure and stroke risk, which supports the development of targeted early warning systems and precision interventions for vulnerable groups. These ecological findings underscore the need for further experimental research to elucidate the biological mechanisms by which size-resolved PM components affect human health. This study has several limitations. First, limited assessment of individual-level exposure may have occurred due to the time-series design, which could introduce a degree of exposure misclassification. PM component concentrations were obtained from a single central monitoring site. Although we discussed the ability of the single site to capture city-level temporal variation (Text S7), there may be potential exposure misclassification. However, because the spatial distribution of air pollutants is generally more stable than their temporal variation, the estimated associations are likely to remain informative and provide a credible reference for population-level inference. Second, although major covariates were adjusted for, residual confounding remains possible. Individual-level factors, such as hypertension, dietary habits, genetic predisposition, physical activity, and smoking status, were not available and may have influenced the results. Third, stroke hospital admissions were identified from five tertiary hospitals in Tianjin, which may not fully capture all stroke cases in the general population, potentially leading to selection bias. Fourth, although the two-stage modeling framework minimizes the influence of other particulate matter components, the residual-based adjustment may still introduce measurement uncertainty arising from filter-based quantification including weighing variability, conditioning, humidity effects, and laboratory analytical errors. Furthermore, the residual term may still reflect the influence of unmeasured coexposures. Therefore, the estimated associations should be interpreted as conditional relationships rather than as definitive evidence of component-specific causality. In addition, the relatively smaller number of hemorrhagic stroke cases may have limited statistical power and reduced estimate stability; thus, findings related to hemorrhagic stroke warrant further confirmation in larger studies. Fifth, although the sampling period spanned all seasons and months­(Figure S9), the noncontinuous exposure measurements may introduce additional statistical uncertainty, which could reduce statistical power and increase the variability of the estimated effects.

Finally, although the estimated relative risks were modest, they are consistent with previously reported effect sizes in component-specific air pollution studies , and may translate into substantial population-level impacts given the high incidence of stroke and widespread exposure, reinforcing the broader public health relevance of our findings.

5. Conclusion

By integrating source apportionment with a two-stage epidemiological modeling framework, we quantified exposure–response relationships across 6 emission sources, 41 chemical components, 9 particle size fractions, and 2 stroke subtypes, thereby establishing a structured source–particle property–susceptible population–stroke subtype risk pathway. Coal combustion and its associated water-soluble ions and polycyclic aromatic hydrocarbons were primarily linked to increased IS risk among males, whereas industrial emissions, together with related potentially toxic metals and PAHs, were associated with elevated HS risk among individuals younger than 60 years. These findings underscore the critical role of particulate physicochemical characteristics in shaping stroke risk and highlight the differential impacts of emission sources across stroke subtypes and vulnerable populations, thereby providing a more precise framework for stroke prevention and targeted environmental health protection.

Glossary

Abbreviations

PM

Particulate matter

IS

Ischemic stroke

HS

Hemorrhagic stroke

GAM

Generalized additive model

CI

Confidence interval

IQR

Interquartile range

CT

Computed tomography

MRI

Magnetic resonance imaging

ICD-10

International Classification of Diseases, 10th Revision

QA/QC

Quality assurance/quality control

ECMWF

European Centre for Medium-Range Weather Forecasts

DOW

Day of week

PMF

Positive matrix factorization

RR

Relative risk

PAHs

Polycyclic aromatic hydrocarbons

BC

Black carbon

OM

Organic matter

AOP

Adverse outcome pathway

PI3K

Phosphoinositide 3-kinase

MAPK

Mitogen-activated protein kinase

HDL-C

High-density lipoprotein cholesterol

SBP

Systolic blood pressure

TACI

Total anterior circulation infarct

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/envhealth.6c00019.

• Smmary of the environmental factors included in the analysis, the quality assurance and quality control procedures applied to particulate matter sampling and chemical analysis, and the results of sensitivity analyses, including two-pollutant models and models fitted with alternative degrees of freedom for confounding adjustment. Additional methodological details are also presented, including component-specific cumulative risk assessments and evaluations of composition-based modification effects (PDF)

Conceptualization: YT, XL, LW, YF; Methodology: YT, XL; Data curation: KH, MMW, MYW, FW, QD, XF; Formal analysis: KH; Investigation: KH, MMW, YT, XL; Visualization: KH; Funding acquisition: YT, XL; Resources: YT, XL, LW, YF; Validation: YT; Supervision: YT, XL; Writing – original draft: KH, MMW, YT, XL; Writing – review and editing: KH, YT, XL

China Meteorological Administration–Nankai University Cooperation Project for Environment-Health-Meteorology Research grant CMANKUA202402 (YT). National Natural Science Foundation of China grant 42275197 (XL). Fundamental Research Funds for the Central Universities, Nankai University grants 040–63253206 (YT). Fundamental Research Funds for the Central Universities, Nankai University grants 040–63241557 (YT). Fundamental Research Funds for the Central Universities, Nankai University grants 040–9242000720 (YT).

The authors declare no competing financial interest.