Regional heterogeneity and the role of social vulnerability in air pollution-related mortality from chronic lower respiratory diseases: a nationwide analysis of the United States (2010–2020)

Abstract

Background

Fine particulate matter (PM_2.5) is a recognized risk factor for chronic lower respiratory diseases (CLRD) mortality. While Environmental Justice frameworks posit that socioeconomically disadvantaged populations bear disproportionate health burdens, nationwide studies using pooled estimates may mask critical geographic and seasonal heterogeneities. This study aimed to evaluate these spatiotemporal variations across the contiguous US and examine effect modification by the Social Vulnerability Index (SVI).

Methods

We analyzed monthly time-series data from 3,108 counties (2010–2020). CLRD mortality data (ICD-10: J40-J47) were obtained from CDC WONDER, and monthly average concentrations of PM_2.5 and ozone were aggregated from daily measurements derived from the US EPA Air Quality System using inverse distance weighting (IDW). Generalized Linear Models (GLM) with a negative binomial distribution were employed, adjusting for ozone, temperature, seasonality, and long-term trends. Analyses were stratified by Census Region and season. We introduced an interaction term (PM_2.5×SVI) to test for effect modification and used restricted cubic splines (RCS) to visualize exposure-response associations. Sensitivity analyses, including removing ozone adjustment, altering temperature controls, and excluding 2020 data to rule out COVID-19 anomalies, were conducted to verify robustness.

Results

Nationwide, the association was borderline (RR = 1.02, 95% CI: 1.00-1.04, P = 0.075), but regional disparities were striking. The Midwest exhibited exceptionally high risk (RR = 2.61, 95% CI: 2.45–2.79, P < 0.001) with a steep exposure-response curve consistent across seasons. Conversely, the South showed a negative association (RR = 0.70, 95% CI: 0.53–0.92, P = 0.043), primarily driven by the warm season. No significant associations were observed in the Northeast or West. Importantly, SVI did not significantly modify the PM_2.5-mortality association in any region (P-interaction > 0.05), and results remained robust across sensitivity models.

Conclusions

The impact of PM_2.5 on CLRD mortality exhibits substantial regional heterogeneity. The Midwest identified as a highly susceptible hotspot, while the South’s negative association likely reflects exposure misclassification (e.g., air conditioning use). Our findings support the “Universal Susceptibility” hypothesis in the Midwest, suggesting that regional pollution haze poses a consistent threat across socioeconomic spectrums, unaffected by community vulnerability levels. Policy interventions should prioritize region-specific emission controls, particularly in the Midwest.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-26907-2.

Introduction

Chronic Lower Respiratory Diseases (CLRD), comprising chronic obstructive pulmonary disease (COPD), asthma, and bronchiectasis, rank as the third leading cause of death globally, imposing a substantial burden on public health systems [1, 2]. Epidemiological evidence indicates that even short-term exposure to fine particulate matter (PM_2.5) can induce acute exacerbations of respiratory symptoms through oxidative stress and systemic inflammation, leading to excess mortality [3, 4]. Although the United States has achieved significant reductions in average particulate concentrations over the past few decades pursuant to the Clean Air Act, recent studies suggest that air quality improvements have stagnated or even reversed in certain regions due to climate change-driven wildfires and the lagged effects of specific industrial emissions [5, 6].

In assessing the health effects of air pollution, “Environmental Justice” (EJ) remains a central theme. Traditional EJ theory posits that communities with lower socioeconomic status (SES) are not only exposed to higher concentrations of pollutants [7, 8] but also exhibit greater susceptibility due to limited medical resources, poorer baseline health (e.g., obesity, high smoking rates), and psychosocial stress [9]. However, empirical support for this hypothesis remains inconsistent. While some studies confirm that disadvantaged communities bear a disproportionate risk, others propose the “Universal Susceptibility” hypothesis, suggesting that when pollution manifests as widespread regional haze, affluent and impoverished communities alike are unable to avoid exposure [10, 11].

The root of these contradictory findings may lie in the methodological limitations of prior research. Most existing studies rely either on single-city analyses, which lack generalizability, or on nationwide pooled models that mask substantial heterogeneity across geographic regions. Given the vastness of the United States, significant differences exist in pollution sources—such as wildfires in the West, agricultural ammonia emissions in the Midwest, and traffic exhaust in the Northeast—as well as in demographic structures and climate adaptation behaviors (e.g., prevalence of air conditioning). These variations likely lead to complex disparities in actual personal exposure and PM_2.5 toxicity [8, 12].

To address these knowledge gaps, this study utilizes a comprehensive, county-level dataset covering the 48 contiguous US states from 2010 to 2020. The objectives of this study are: (1) to systematically evaluate the association between short-term PM_2.5 exposure and CLRD mortality on a nationwide scale; (2) to compare risk disparities across the four US Census Regions (Northeast, Midwest, South, and West); (3) to investigate the temporal stability of these associations across different seasons (warm vs. cold); and (4) to examine whether socioeconomic status modifies the health effects of air pollution in different regions using the multi-dimensional Social Vulnerability Index (SVI).

Methods

Study design and data sources

We employed an ecological time-series study design with the county-month as the unit of analysis. The study area covered 3,108 counties across the 48 contiguous United States over an 11-year period from January 2010 to December 2020. Data aggregation and linkage were performed using unique 5-digit Federal Information Processing Standards (FIPS) codes.

Monthly mortality counts aggregated at the county level were obtained from the Centers for Disease Control and Prevention (CDC) WONDER database. We identified cases across all age groups (0–85 + years) where the underlying cause of death was CLRD (ICD-10 codes J40-J47) [13].

Daily summary monitoring data (specifically, the daily arithmetic mean) for PM_2.5 and ozone (O₃) were acquired from the US Environmental Protection Agency (EPA) Air Quality System (AQS). We prioritized these ground-based measurements over satellite-derived products to avoid non-random missingness caused by cloud cover and high surface reflectance (e.g., snow or ice), thereby ensuring robust temporal continuity [14, 15]. To address the spatial limitation of monitoring stations, we applied Inverse Distance Weighting (IDW) interpolation to estimate county-level monthly and annual average concentrations (based on the daily means) at the geometric centroid of each county [16]. This approach was selected to capture the representative ambient air quality of the administrative unit. Given the regional homogeneity of PM_2.5 compared to traffic-related pollutants, the geometric centroid provides a reliable estimate of background exposure [17]. Furthermore, validation studies in time-series settings indicate that geometric mean concentrations exhibit high temporal correlations (> 0.9) with population-weighted metrics, ensuring that the temporal variability driving our mortality estimates remains robust [18]. Consequently, the spatial resolution of our exposure assessment is defined at the county level. The specific geographical locations of the monitoring stations are visualized in Supplementary Figure S1, which demonstrates dense coverage in highly populated regions, ensuring robust exposure assessment for the majority of the study population. The detailed descriptive statistics regarding the number of monitored counties and the corresponding population coverage are provided in Supplementary Table S2.

Monthly average temperature data were sourced from the National Oceanic and Atmospheric Administration (NOAA) “Climate at a Glance” County Mapping system (derived from the nClimDiv dataset) [19]. This dataset provides spatially averaged temperature values for each county, derived from station data using climatologically aided interpolation, thereby eliminating the need for further spatial processing. Annual average temperatures were calculated as the arithmetic mean of the 12 monthly averages for each county.

The SVI was obtained from the CDC Agency for Toxic Substances and Disease Registry (ATSDR). We utilized the specific county-level SVI dataset provided by ATSDR, which aggregates 15 census tract-level variables (e.g., poverty, education, housing crowding, minority status) into a composite ranking ranging from 0 to 1, with higher values indicating greater social vulnerability [20].

Statistical analysis

Descriptive statistics for continuous variables are expressed as mean±standard deviation (SD) or median with interquartile range (IQR) depending on their distribution, while categorical variables are presented as frequencies and percentages (%). Spearman’s rank correlation analysis was performed to assess collinearity among environmental factors. The annual spatial visualization was based on daily arithmetic means aggregated to annual averages. Given that mortality counts typically exhibit overdispersion, we employed Generalized Linear Models (GLM) with a negative binomial distribution for the time-series regression analysis [21]. The core model specification is defined as follows:

where represents the expected death count in county at month ; and denote the monthly average concentrations of fine particulate matter and ozone, respectively; represents the county-level Social Vulnerability Index; denotes the natural spline function with 3 degrees of freedom to control for the non-linear confounding effects of ambient temperature [22]; includes categorical variables for year and month to account for long-term trends and seasonality; and serves as the offset term to standardize for population size.

To investigate regional heterogeneity, analyses were stratified by the four US Census Regions: Northeast, Midwest, South, and West. To further explore the potential drivers of regional divergence and rule out seasonal confounding, we conducted a season-stratified analysis, defining the cold season as October to March and the warm season as April to September. Separate models were fitted for each season within each region.

To evaluate whether social vulnerability modifies the health effects of air pollution, we introduced an interaction term (PM_2.5×SVI) into the models. Furthermore, the non-linear exposure-response relationship between PM_2.5 and mortality was visualized using restricted cubic splines (RCS) with 3 knots [23]. Although ozone typically peaks during the warm season, we included year-round ozone concentrations in the models to account for potential confounding effects across all seasons, given recent evidence of rising winter ozone levels in certain US regions and its persistent association with respiratory mortality throughout the year [24, 25].

Sensitivity analyses were performed to verify the robustness of our findings through three specific specifications: (1) excluding ozone adjustment to assess stability against co-pollutant confounding; (2) replacing the non-linear temperature spline with a linear term to test model dependency; and (3) restricting the study period to 2010–2019 to exclude potential data anomalies and confounding associated with the COVID-19 pandemic. Data cleaning and primary model fitting were performed using Python (version 3.9) with the pandas and statsmodels libraries, while additional statistical verification and advanced visualization were conducted using R software (version 4.2.1; R Foundation for Statistical Computing, Vienna, Austria). Spatial visualization was performed using ArcGIS software (version 10.8, Esri, Redlands, CA, USA). All statistical tests were two-sided, and a P-value < 0.05 was considered statistically significant.

Results

Descriptive statistics

Table 1 presents the baseline monthly characteristics of environmental exposures, meteorological conditions, and health outcomes across US Census Regions during the study period. The nationwide average monthly PM_2.5 concentration was 8.35 ± 2.58 µg/m³. The South had the highest average PM_2.5 exposure (8.70 ± 1.88 µg/m³) and the greatest social vulnerability (SVI = 0.65). In contrast, the West recorded the highest average ozone levels (0.034 ppm), while the Northeast exhibited the highest average monthly mortality burden per county (6.39 deaths). To visualize the long-term spatial patterns distinct from the monthly temporal variations shown in Table 1, we aggregated the data into annual averages, as shown in Fig. 1. Spearman’s rank correlation analysis indicated that correlations between PM_2.5, O₃ and temperature were generally weak (< 0.4), confirming that multicollinearity was not a confounding issue in our regression models (Supplementary Table S1). Regarding exposure monitoring coverage, 1,206 of the 3,108 counties (38.8%) contained at least one active EPA monitoring station during the study period. While the raw spatial coverage varies by region, these monitored counties account for 84.3% of the total study population nationwide (ranging from 76.6% in the South to 97.5% in the West), ensuring that ground-based observations directly represent the vast majority of the population at risk (Supplementary Table S2).

Table 1.

Descriptive statistics of environmental exposures, meteorological conditions, and mortality from Chronic Lower Respiratory Diseases across US Census Regions (2010–2020)

Variable	Nationwide	Northeast	Midwest	South	West
Sample Characteristics
No. of Counties	3,108	217	1,055	1,422	414
Total Study Months	410,256	28,644	139,260	187,704	54,648
Health Outcomes
Total Deaths (Cumulative)	934,670	182,942	187,318	318,315	246,095
Average Monthly Deaths (County)	2.28 ± 9.86	6.39 ± 11.54	1.35 ± 7.12	1.70 ± 7.09	4.50 ± 18.46
Environmental Exposures (Monthly Avg)
PM2.5 (µg/m³)	8.35 ± 2.58	8.04 ± 2.40	8.22 ± 2.20	8.70 ± 1.88	7.67 ± 4.62
Ozone (ppm)	0.031 ± 0.007	0.030 ± 0.007	0.031 ± 0.006	0.030 ± 0.006	0.034 ± 0.008
Meteorology (Monthly Avg)
Temperature (℃)	12.98 ± 9.92	9.49 ± 9.48	9.91 ± 10.64	16.79 ± 8.22	9.59 ± 8.89
Socioeconomic Status
Social Vulnerability Index (0–1)	0.50 ± 0.29	0.37 ± 0.21	0.33 ± 0.23	0.65 ± 0.25	0.49 ± 0.30

Fig. 1.

Spatial Distribution of Mortality and Environmental-Social Determinants across the Contiguous United States (2010–2020)

Regional heterogeneity in mortality risk

The regression analysis revealed that national averages masked substantial spatial heterogeneity in the impact of PM_2.5 on CLRD mortality (Table 2). While the nationwide pooled model suggested a borderline positive association (RR = 1.02, 95% CI: 1.00-1.04, P = 0.075), stratified analyses uncovered striking regional disparities. The Midwest emerged as a region of exceptionally high susceptibility, with an RR of 2.61 (95% CI: 2.45–2.79, P < 0.001), significantly exceeding estimates from other regions. In stark contrast, the South exhibited a significant negative association (RR = 0.70, 95% CI: 0.53–0.92, P = 0.043). No statistically significant associations were observed in the Northeast or West, where confidence intervals crossed the null value of 1.0. These distinct spatial patterns are visually emphasized in the forest plot (Fig. 2), which highlights the unique vulnerability of the Midwest.

Table 2.

Association between PM_2.5 and CLRD mortality, and effect modification by social vulnerability index (SVI). RR and P-values were derived from the main effect models (without interaction term). P-interaction values were derived from separate interaction models adjusting for the same covariates

Region	No. of County-Months	RR (95% CI) per 10 µg/m³	P-value	P-interaction (SVI)
Nationwide	410,256	1.02 (1.00, 1.04)	0.075	0.489
Northeast	28,644	1.06 (0.98, 1.15)	0.145	0.160
Midwest	139,260	2.61 (2.45, 2.79)	< 0.001	0.645
South	187,704	0.70 (0.53, 0.92)	0.043	0.913
West	54,648	0.96 (0.93, 1.00)	0.149	0.659

Fig. 2.

Forest plot showing the relative risk (RR) of mortality from chronic lower respiratory diseases associated with a 10 µg/m³ increase in PM_2.5 concentration, stratified by US Census Region. Dots represent the point estimate of RR, and horizontal bars indicate 95% confidence intervals. Red indicates a statistically significant positive association, blue indicates a significant negative association, and gray indicates no significance (P>0.05)

Exposure-response relationship

To further examine the linearity of these associations, we modeled exposure-response curves using restricted cubic splines (Fig. 3). The curve for the Midwest (red line) displayed a steep, monotonic increase in mortality risk with rising PM_2.5 concentrations, indicating that risk is elevated even at lower exposure levels. Conversely, the curve for the South (blue line) trended downwards, consistent with the linear risk estimates and the seasonal heterogeneity described below (Table 3). The nationwide average curve (black line) remained relatively flat, further illustrating how regional signals cancel each other out in pooled analyses.

Fig. 3.

Exposure-response relationships between monthly PM_2.5 concentration and mortality from chronic lower respiratory diseases (CLRD) stratified by US Census Region (2010–2020). Curves were estimated using generalized linear models with restricted cubic splines (3 knots). The solid lines represent the relative risk (RR) of mortality compared to a reference PM2.5 concentration of 0 μg/m³. Shaded areas indicate 95% confidence intervals. Models were adjusted for ozone, ambient temperature, SVI, seasonality, and long-term trends

Table 3.

Seasonal stratification and sensitivity analyses of the association between monthly PM_2.5 concentration and mortality from chronic lower respiratory diseases (CLRD) by US Census Region (2010–2020). Values are presented as RR (95% CI) per 10 µg/m³ increase in PM_2.5. Seasonal stratification: cold season is defined as October to March; warm season is defined as April to September

Model Specification	Nationwide	Northeast	Midwest	South	West
Primary Analysis	1.02 (1.00, 1.04)	1.06 (0.98, 1.15)	2.61 (2.45, 2.79)	0.70 (0.53, 0.92)	0.96 (0.93, 1.00)
Seasonal Stratification
Cold Season (Oct–Mar)	1.32 (1.28, 1.37)	1.43 (1.30, 1.58)	2.68 (2.53, 2.82)	1.07 (0.99, 1.16)	1.05 (1.00, 1.11)
Warm Season (Apr–Sep)	0.83 (0.80, 0.86)	1.02 (0.96, 1.08)	2.52 (2.31, 2.76)	0.32 (0.29, 0.35)	0.88 (0.84, 0.92)
Sensitivity Analyses
I† Excluding ozone adjustment	1.02 (1.00, 1.05)	1.09 (1.00, 1.17)	2.62 (2.46, 2.80)	0.68 (0.64, 0.72)	0.98 (0.95, 1.01)
II‡ Linear temperature control	1.02 (1.00, 1.05)	1.19 (1.10, 1.28)	2.54 (2.38, 2.70)	0.76 (0.72, 0.80)	0.96 (0.93, 0.99)
III§ Pre-pandemic (2010–2019)	1.00 (0.98, 1.03)	1.05 (0.97, 1.14)	2.59 (2.42, 2.77)	0.68 (0.64, 0.72)	0.96 (0.92, 1.00)

Sensitivity Analysis I^†: Model excluding ozone adjustment

Sensitivity Analysis II^‡: Model using a linear term for temperature control

Sensitivity Analysis III^§: Model restricted to the pre-pandemic period (2010–2019) to exclude potential confounding from COVID-19

Seasonal stratification and regional divergence

To investigate the drivers of the divergent exposure-response shapes—specifically the steep rise in the Midwest versus the inverse trend in the South—we conducted a season-stratified analysis (Table 3). The high mortality risk in the Midwest remained consistent across seasons, with RR estimates of 2.68 (95% CI: 2.53–2.82) in the cold season and 2.52 (95% CI: 2.31–2.76) in the warm season. This consistency indicates that seasonal confounding (e.g., winter heating) is not the primary driver of the observed association. Conversely, the inverse association in the South was exclusively driven by the warm season (RR = 0.32, 95% CI: 0.29–0.35), consistent with exposure misclassification driven by high air-conditioning prevalence. During the cold season, the risk estimate in the South moved toward the null (RR = 1.07, 95% CI: 0.99–1.16), aligning with reduced behavioral adaptation during cooler months.

Universal susceptibility (effect modification)

We tested the Environmental Justice hypothesis by introducing an interaction term between PM_2.5 and SVI (Table 2). The interaction terms were not statistically significant in the nationwide model or in any of the four regional models (P for interaction > 0.05). This lack of effect modification suggests that SVI, as a composite socioeconomic indicator, did not significantly alter the lethal effects of PM_2.5. In the high-risk Midwest region, the pollution burden affected communities broadly, supporting the hypothesis of “Universal Susceptibility” rather than risks being concentrated solely in socially vulnerable neighborhoods.

Sensitivity analysis

Sensitivity analyses confirmed the robustness of our primary findings (Table 3). The distinct high risk in the Midwest (RR remaining > 2.5) and the negative association in the South (RR remaining < 0.8) persisted across all model specifications, including models without O₃ adjustment (Sensitivity Analysis I), those using linear temperature controls (Sensitivity Analysis II), and models excluding data from the pandemic year 2020 (Sensitivity Analysis III). This consistency suggests that the observed regional patterns are not artifacts of co-pollutant confounding, pandemic-related anomalies, or model selection.

Discussion

In this study, our nationwide estimate (RR = 1.02 per 10 µg/m³) aligns closely with the effect sizes reported in major multi-city studies and large-scale cohorts [10, 26], supporting the overall validity of our exposure-response modeling. However, masking this national average, our analysis unveils a striking “Midwest Specificity” regarding the impact of PM_2.5 on respiratory mortality and provides empirical support for the “Universal Susceptibility” hypothesis within the context of regional pollution.

Mechanisms underlying high risk in the midwest

The exceptionally high relative risk observed in the Midwest (RR = 2.61) is likely attributable to the unique chemical composition of particulate matter in this region. Distinct from wildfire smoke in the West [27] or traffic-related emissions in the Northeast, PM_2.5 in the Midwest is characterized by secondary inorganic aerosols, particularly ammonium nitrate and sulfates. These components are formed through the reaction of agricultural ammonia emissions, derived from fertilizers and livestock, with sulfur dioxide emitted by coal-fired power plants [28, 29]. Crucially, our season-stratified analysis validates the persistence of this risk. The mortality risk in the Midwest remained exceptionally high in both cold and warm seasons. This seasonal consistency is pivotal: it effectively rules out transient seasonal confounders—such as cold stress in winter or heat waves in summer—as primary drivers of the observed association. Instead, it points to a chronic, year-round exposure to highly toxic PM_2.5 constituents (e.g., secondary inorganic aerosols) specific to this region. Toxicological evidence suggests that sulfate-nitrate rich particles possess higher oxidative potential and cytotoxicity compared to carbonaceous aerosols, thereby inducing more potent airway inflammation [30]. Furthermore, the Midwest has a higher baseline prevalence of CLRD and obesity, coupled with an aging population. This reduced physiological reserve renders the population more sensitive to fluctuations in air quality [31]. Additionally, it is important to note that the magnitude of the observed risk is higher than typically reported in daily time-series studies. This likely reflects the nature of monthly aggregated data, which captures cumulative or sub-acute exposure burdens over a 30-day window, potentially yielding larger effect sizes than models focusing solely on immediate, single-day lags.

The “air conditioning hypothesis” and protective effects in the south

The counterintuitive negative association observed in the South (RR < 1) is effectively explained by the “Air Conditioning Hypothesis” [32], a mechanism now empirically supported by our season-stratified findings. We observed a drastic divergence in risk estimates between seasons: the negative association was exclusively driven by the warm season (RR = 0.32), coinciding with the period of peak air conditioning usage. In contrast, during the cold season—when behavioral decoupling from the outdoor environment is minimized—the risk estimate moved toward the null (RR = 1.07).

Due to the hot and humid climate, the prevalence of central air conditioning in the South exceeds 90% [33]. Widespread AC usage creates “tight building envelopes” during warm months, effectively uncoupling outdoor pollutant concentrations from actual personal indoor exposure. Consequently, outdoor monitoring data significantly overestimate the inhaled dose during the warm season [34], leading to the false appearance of a “protective” effect. While competing risks from extreme heat may also contribute to mortality patterns during smog events [35], the disappearance of the negative association during the cold season confirms that the anomaly is primarily driven by behavior-mediated exposure misclassification rather than biological mechanisms.

Universal susceptibility vs. environmental justice

Contrary to micro-scale Environmental Justice studies that focus on neighborhood-level disparities [36], we did not observe significant effect modification by SVI. This indicates that in high-risk regions like the Midwest, air pollution manifests as widespread “Regional Haze” [37]. Unlike point-source pollution (e.g., adjacent to highways or industrial plants), this large-scale pollution pattern blankets vast geographic areas. It affects residents of affluent suburbs and vulnerable urban cores equally, thereby “flattening” the socioeconomic gradient of risk [38].

Strengths and limitations

Strengths of this study include its large-scale nationwide coverage, uniform methodological framework, and rigorous sensitivity analyses (including seasonal stratification and pandemic-year exclusion) that confirmed the robustness of regional heterogeneity. However, several limitations must be acknowledged. First, as an ecological study, findings are subject to the ecological fallacy and cannot be directly extrapolated to individual-level risks. Second, regarding exposure assessment, we estimated PM_2.5 concentrations at the geometric centroid of each county. While population-weighted centroids may better reflect personal exposure in large counties with uneven population distribution, PM_2.5 generally exhibits spatial homogeneity over monthly scales [17]. Given that our significant findings were concentrated in the Midwest and South where counties are generally smaller, we expect the discrepancy between geometric and population-weighted centers to be minimal [18]; any remaining non-differential error would likely bias results toward the null. Additionally, the spatial coverage of the EPA monitoring network presents a limitation, as 61.2% of counties (primarily sparsely populated areas) rely on IDW interpolation. However, our coverage statistics show that these unmonitored counties account for only 15.7% of the total study population (Supplementary Table S2). This ensures that our population-weighted risk estimates are primarily driven by high-accuracy direct observations from monitored regions. Third, this study did not explicitly quantify specific chemical constituents of PM_2.5, warranting future research to validate the compositional hypotheses. Fourth, due to privacy restrictions of the public database, our analysis was based on monthly mortality counts. We acknowledge that this temporal aggregation limits the detection of acute, daily lag effects and may introduce exposure misclassification for deaths occurring early in the month. However, unlike daily time-series models that capture transient triggers, monthly aggregation captures the sub-acute or cumulative exposure burden over a 30-day window. This methodological distinction likely explains why the magnitude of our estimated associations (e.g., in the Midwest) appears larger than those in studies focusing solely on single-day lags, as our results reflect the health impact of prolonged physiological stress.

Conclusions

This study confirms substantial regional heterogeneity in the impact of PM_2.5 on CLRD mortality across the United States, identifying the Midwest as a critical high-risk hotspot driven by specific chemical toxicity and cumulative exposure burdens. In contrast, the apparent protective effect in the South was identified as a seasonal artifact likely attributable to adaptive behaviors (e.g., air conditioning usage). Crucially, this elevated risk in the Midwest is not confined to socially vulnerable populations but transcends socioeconomic boundaries, reflecting a pattern of universal susceptibility. These findings underscore the imperative of incorporating regional specificity into air quality standards. For the Midwest specifically, mitigation strategies should prioritize the control of gaseous precursors from agricultural and industrial sources instead of focusing solely on primary particulate control to effectively protect public health.

Authors’ contributions

Dong Miao conceived and designed the study, analyzed the data, acquired funding, drafted the manuscript, and revised the manuscript. Yingping Tian, Xiaolei Ye and Gang He conceived and designed the study, supervised the study and critically revised the manuscript. Xinfeng Bai acquired funding, participated in data analysis, and revised the manuscript. Li’ao Wang and Meiying Wang performed data analysis, and revised the manuscript. Jun Gan, Guojing Bao, and Junfeng Zheng collected the data and revised the manuscript. Wei Chen and Yunsheng Hou participated in the study design and provided critical revision for important intellectual content. All authors read and approved the final manuscript.

Funding

This work was supported by the Hebei Provincial Medical Science Research Project (Grant No. 20240492) and the Hebei Provincial Government Clinical Medicine Outstanding Talent Training Program (Grant No. ZF2025315).

Data availability

The mortality data are available from the CDC WONDER database (https://wonder.cdc.gov). Environmental data are available from the US EPA Air Quality System (https://www.epa.gov/aqs). The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable. This study used publicly available, aggregated data from the CDC WONDER and US EPA databases, which does not involve human subjects requiring ethical approval.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.