Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Aug 6.
Published in final edited form as: Environ Res. 2025 Aug 6;285(Pt 3):122520. doi: 10.1016/j.envres.2025.122520

Chronic effects of wildfire smoke and criteria air pollutants on cardiovascular hospitalization rates in the contiguous US

Cheng Jin 1,2, Mahdieh Danesh Yazdi 3, Hanbing He 4, Edgar Castro 5, Joel D Schwartz 5, Robert O Wright 1, Yaguang Wei 1
PMCID: PMC12393821  NIHMSID: NIHMS2102806  PMID: 40774559

Abstract

Existing studies on the health effects of smoke fine particulate matters (PM2.5), a primary emission from wildfires, have often lacked comparison with other air pollutants, focused primarily on acute exposures, and not applied causal methods. In the present study, we obtained county-level, three-year average cardiovascular hospitalization rates for Medicare beneficiaries across the contiguous US between 2006–2016 from the Centers for Disease Control and Prevention. These data were linked with spatio-temporal estimates of smoke PM2.5, non-smoke PM2.5, nitrogen dioxide (NO2), ozone, and county-level confounders. We used a difference-in-differences method to evaluate causal effects of three-year moving average exposures (lag 0–2, 1–3, 2–4, or 3–5 year) to the four pollutants on hospitalization rates for total cardiovascular disease (CVD) and its two major subtypes: heart disease and stroke. We found that, for total CVD, the absolute change in hospitalization rate associated with smoke PM2.5 increased with longer lag periods: from −0.879 (95% confidence interval [CI]: −2.528, 0.771) at lag 0–2 to 7.538 (95% CI: 4.594, 10.481) at lag 3–5 per 1 μg/m3 increase in exposure per 1,000 people. The effect of non-smoke PM2.5 was smaller and diminished over time. NO2 and ozone had even smaller effects per 1 part per billion increases in exposure. Similar patterns were seen for heart disease. For stroke, all pollutants had minimal and mostly non-significant effects. More rural and lower-income counties experienced greater risks. These findings suggested the need to prioritize wildfire management in addition to traditional air quality control strategies.

Keywords: wildfire, smoke PM2.5, non-smoke PM2.5, cardiovascular diseases, causality

Introduction

Fine particulate matter (PM2.5) is a major ambient air pollutant that poses a significant threat to human health1. Although PM2.5 concentration in the US has decreased since the enactment of the Clean Air Act, progress has recently slowed down or even reversed2. A key driver of this trend is wildfires, which have substantially increased in frequency and intensity in recent years and emitted substantial amounts of smoke PM2.53. The trend is expected to persist as the climate continues to warm and dry, creating conditions that are more conducive to wildfires4.

Toxicological studies suggest that smoke PM2.5 was more toxic than non-smoke PM2.5 due to its smaller size and higher concentration of carbonaceous compounds, which have greater potential to cause oxidative stress and inflammation5,6. However, large-scale evidence directly comparing health effects of smoke and non-smoke PM2.5 remains limited. Addressing this gap is critical, as demonstrating a greater effect of smoke PM2.5 would suggest the need to prioritize wildfire management in addition to traditional air quality control strategies7. A major barrier to filling this gap has been the lack of spatio-temporal estimates of ambient smoke and non-smoke PM2.5 concentrations, which are commonly used as proxy exposure measures for large studies3. Recently, significant advances in satellite remote sensing technology, machine learning algorithms, and computational power have made it possible to generate high-resolution, full-coverage estimates of smoke PM2.58. Many models have since been developed to obtain proxy exposure measures with relatively high accuracy9,10.

Most existing studies evaluating the health effects of smoke PM2.5 have focused on acute exposures over a few days11–14. However, chronic exposure to total PM2.5 (including both smoke and non-smoke PM2.5) that lasts one year or more has been shown to pose much greater risk than acute exposure15. Despite this, long-term evidence for the health effects of smoke PM2.5 exposure remains limited16,17. Additionally, few studies have used causal modeling techniques, which provide greater assurance against confounding bias.

We linked existing spatio-temporal estimates of ambient smoke PM2.5, non-smoke PM2.5, and two other criteria pollutants (nitrogen dioxide [NO2] and ozone) with county-level cardiovascular hospitalization rates among Medicare beneficiaries (aged ≥65 years) across the contiguous US from 2006 to 2016. We focused on cardiovascular disease (CVD) because it is a leading cause of morbidity and mortality and is a well-established outcome of oxidative stress and inflammation caused by total PM2.5 mass18. NO2 and ozone have independently been linked to increased cardiovascular risk, and have been studied together with total PM2.5 in several previous studies as confounders for each other15,19. Including these pollutants may help isolate the effect estimates for smoke and non-smoke PM2.520. We evaluated the simultaneous effects of these pollutants over exposure periods of up to five years, using a difference-in-differences (DID) method that accounted for both measured and unmeasured confounders. Subgroup analyses were also performed to identify at-risk subpopulations.

Materials and Methods

Hospitalization data

We obtained county-level cardiovascular hospitalization rates per 1,000 Medicare Fee-for-Service beneficiaries aged 65 and older for every three consecutive years (i.e., 2006–2008, 2007–2009, and so forth through 2014–2016) across 3,221 counties in the contiguous US from 2006 to 2016. The data were sourced from the Centers for Medicare and Medicaid Services Medicare Provider Analysis and Review (MEDPAR) Part A file and released by the Centers for Disease Control and Prevention (CDC)’s Interactive Atlas of Heart Disease and Stroke21,22. The outcomes of interest included total CVD and its two major subtypes: heart disease and stroke. Hospitalizations prior to 2015 were identified using the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes: 390–434 and 436–448 for total CVD; 390–398, 402, 404, and 410–429 for heart disease; and 430–434 and 436–438 for strokes. Following 2015, the hospitalizations were identified using ICD-10-CM codes: I00–I78 for total CVD; I00–I09, I11, I13, and I20–I51 for heart disease; and I60–I69 for strokes. Major subtypes of heart disease included myocardial infarction, cardiac dysrhythmia, heart failure, and hypertension; and major subtypes of stroke included ischemic and hemorrhagic stroke.

Exposure assessment

From an open source, we obtained daily, ambient smoke PM2.5 levels at 10 km2 grid cells in the contiguous US from 2006 to 202023. The data were created by Childs et al. using gradient-boosted trees with an overall spatial cross-validated R2 of 0.679. The modeling process first identified “smoke days” based on either the presence of smoke plumes from National Oceanic and Atmospheric Administration Hazard Mapping System or the presence of 50 Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) points combined with over 75% missingness in Aerosol Optical Depth. Smoke PM2.5 were then estimated as the positive anomalies between monitored total PM2.5 and monthly median PM2.5 on non-smoke days. Finally, a gradient-boosted trees model was trained to predict smoke PM2.5 at locations without monitors. The model incorporated meteorological variables, fire-related features (distance to and size of nearest fire cluster), HYSPLIT points at different height quintiles, aerosol measurements, and cross-sectional features (elevation, land cover percentages). The grid cell-level smoke PM2.5 were aggregated to the county level by averaging estimates from grid cells, using the population of the intersection area between each grid cell and county as weights. The population was estimated by multiplying the grid cell’s population density by the size of the intersection area.

We estimated daily total PM2.5 at 1 km2 grid cells using a hybrid model that incorporated predictors including satellite-based measurements, land-use regression, and chemical transport models24,25. We used neural networks, random forests, and gradient boosting machines to generate three sets of PM2.5 estimates, and ensemble averaged them with a geographically weighted regression. The model showed strong predictive performance with cross-validated R2 on left-out monitors of 0.86. These total PM2.5 estimates at grid cells were aggregated to the county level by grouping all encompassed census blocks and calculating the average of the block-level estimates (assigned from the nearest grid cells), weighted by each block’s population. Non-smoke PM2.5 levels were then calculated by subtracting the smoke PM2.5 from the total PM2.5 at county-level.

Similar to total PM2.5, we predicted daily, 1 km2 grid cell level estimates for NO2 and ozone across the contiguous US using hybrid models25–28. Cross validated R2 were 0.79 for NO2 and 0.90 for ozone. The grid cell-level estimates were aggregated to the county level using population-weighted averaging, consistent with the total PM2.5.

For each county in each year, we evaluated the three-year moving average exposures to the pollutants during the current and previous two years (lag 0–2) to match the averaging period for hospitalization rates. Additionally, we evaluated 1-, 2-, and 3-year lagged exposures, i.e., lag 1–3, lag 2–4, and lag 3–5.

Other covariates

As potential confounders, we adjusted for meteorological variables, population demographics, neighborhood socioeconomic factors, healthcare accessibility, and vegetation, since these factors may be related to air pollutant exposure levels directly or indirectly through behavioral changes, and have also been associated with cardiovascular diseases29–32.

Meteorological variables, including summer and winter average temperatures and precipitation, were obtained from Daymet, which provided daily, minimum and maximum temperatures and total precipitation at 1 km2 grid cells33. For temperature, daily estimates were estimated using nearby station data, with coordinates projected by the Daymet Lambert Conformal Conic system and elevation from a gridded digital model. Nearby stations within a defined radius were selected, and weights were assigned using a truncated Gaussian kernel based on horizontal distance to the grid cell. Spatial temperature gradients were then estimated using weighted least squares on station coordinate differences, and final grid cell temperatures were computed via geographically weighted regression. For precipitation, spatial gradient parameters were estimated using smoothed observations, and the final estimates incorporated constraints on horizontal and vertical gradients, adjustments for localized storms, and limits on maximum values. The daily average temperature was calculated as the mean of daily maximum and minimum temperatures, which were then averaged for summer (June, July, and August) and winter (January, February, and December)34. Daily precipitation was averaged for summer and winter. These grid cell-level estimates were aggregated to the county level to match the spatial resolution of the hospitalization data.

County-level population demographic and socioeconomic factors were obtained from the US Census and the American Community Survey35–37. These factors included the annual percentage of non-Hispanic White and Black residents, percentage of the population under 65 years old, percentage without a high school diploma, percentage living below the poverty line, population density, and percentage of owner-occupied housing units. These data were available for the years 2000 and 2010, as well as annually from 2011 to 2016. Data for the years 2001 to 2009 were not available and were estimated using linear interpolation between the 2000 and 2010.

County-level measurements for healthcare accessibility were obtained from the Dartmouth Atlas of Health Care38. These included the annual percentage of people who had a blood lipids test, percentage of Medicare beneficiaries who had an eye examination, percentage of women who had a mammogram, percentage of people who had a hemoglobin A1c test, and percent of Medicare beneficiaries who had at least one ambulatory visit to primary care clinician. For each variable, missing values were imputed using multiple imputation based on available data for other healthcare accessibility measures, population demographics, and socioeconomic factors from the US Census and American Community Survey as described above39. To account for temporal and spatial structure, year and county were included in the imputation model. We generated 10 imputed datasets and took the average across them for epidemiologic analysis.

Lastly, we adjusted for the Normalized Difference Vegetation Index (NDVI) obtained from the Terra Moderate Resolution Imaging Spectroradiometer Version 6, as a measure of vegetation and rurality40. The data had a temporal resolution of 16 days and a spatial resolution of 250 meters, covering the contiguous US from 2000 to 2023. These were aggregated to the county level.

All covariates were averaged over a 3-year period (lag 0–2) to match the averaging period for hospitalization rates.

Statistical analysis

For each outcome (total CVD, heart disease, or stroke), we used a DID approach to estimate the absolute change in hospitalization rates associated with simultaneous exposures to the four pollutants (smoke PM2.5, non-smoke PM2.5, NO2, and ozone). DID is a quasi-experimental method that controls for both measured and unmeasured confounders and aims to identify “random shocks” in outcome that render the change in exposure independent of confounders41. While originally designed to contrast two locations with binary treatment or exposure, we previously extended it to accommodate multiple locations, time periods, and continuous exposures42. The key assumption of DID was the “parallel trends” assumption43. In this study, because air pollutant exposures were continuous, the “parallel trends” assumption meant that each county would have experienced the same change in hospitalization rates over time per unit increase in each exposure across different exposure levels. We relaxed this assumption by adjusting for county-specific dummy variables to account for unmeasured spatial confounders that vary between counties, including slowly changing or unmeasured characteristics unique to each county. We further adjusted for categorical calendar year to capture unmeasured temporal confounders that vary over time but not between counties. Finally, we adjusted for measured time-varying confounders that did vary between counties, including meteorological variables, population demographics and socioeconomic factors, healthcare accessibility, and NDVI.

Specifically, for each outcome, we assessed the effects of four pollutants over one lag period (lag 0–2, 1–3, 2–4, or 3–5), using a generalized nonlinear model with Gaussian link44. Assuming a linear relationship between each pollutant and hospitalization rate and that the “parallel trends” assumption held, the estimated coefficient represented the average change in hospitalization rate associated with a unit increase in that pollutant. Evaluating lag 0–2 exposure may lead to exposure misclassification, since beneficiaries might be hospitalized early in the period but were assigned the average exposure over the entire three years. As the lag period increased, the likelihood of exposure misclassification decreased. During lag 3–5, the exposure window occurred completely before any hospitalization, eliminating any exposure misclassification related to the exposure window. Given the inherent correlations among the three outcomes, we adjusted for multiple comparisons using Bonferroni correction for all effect estimates.

We also conducted subgroup analyses based on county-level urbanization and median household income. Urbanization was measured by the Rural-Urban Commuting Area (RUCA), which incorporated census tract-level commuting patterns45. We aggregated census tract-level RUCA to the county level, weighted by population, and stratified the all the county-year combinations at the 75th percentile of RUCA to perform separate analysis. Similarly, we stratified the county-year combinations at the 25th percentile of median household income and performed separate analysis. Two sample t tests were used to assess the statistical significance of subgroup differences.

We performed sensitivity analysis by including interaction terms between categorial calendar year and five regions across the country. This further relaxed the “parallel trends” assumption by allowing separate time trends for each region, which weakened the assumption of a common time trend across regions.

Results

Table 1 shows county-level hospitalization rates per 1,000 people for three outcomes and demographic characteristics between 2006–2016. The three-year average hospitalization rate for total CVD was 67.44 per 1,000. Heart disease was the major subtype, accounting for 50.39 per 1,000. The hospitalization rate for stroke was 11.80 per 1,000. The population was predominantly White (83.40%), with Black residents comprising 8.96%. The three-year average rates for healthcare accessibility measures were: hemoglobin A1c testing (84.27%), primary care visits among Medicare beneficiaries (80.78%), blood lipids testing (76.69%), eye examination among Medicare beneficiaries (65.80%), and mammogram among women (62.07%). Over half of residents (51.18%) did not have a high school diploma, 15.81% lived below poverty, and the average county-level median household income was $46,120.96.

Table 1.

Three-year average cardiovascular hospitalization rates and population demographic and socioeconomic factors across counties in the contiguous US.

Mean ± SD
Hospitalization rate per 1,000
 Total CVD 67.44 ± 22.19
 Heart disease 50.39 ± 17.83
 Stroke 11.80 ± 3.19
Covariates
 Percentage of Black, % 8.96 ± 14.5
 Percentage of White, % 83.40 ± 16.64
 Percentage of female, % 50.01 ± 2.27
 Percentage of people who had a blood lipids test, % 76.69 ± 7.03
 Percentage of Medicare beneficiaries who had eye examination, % 65.80 ± 6.79
 Percentage of women who had a mammogram, % 62.07 ± 7.68
 Percentage of people who had hemoglobin A1c test, % 84.27 ± 6.09
 Percentage of Medicare beneficiaries who had at least one ambulatory visit to primary care clinician, % 80.78 ± 7.32
 Percentage of the population who did not graduate from high school, % 51.18 ± 10.87
 Percentage of the population living below the poverty line, % 15.81 ± 6.36
 Median Household Income, $ 46,120.96 ± 12,536.44
 Median Property Value, $ 132,337.04 ± 82,513.92
 RUCA score 5.16 ± 3.06
Open in a new tab

Table 2 shows the distribution of three-year average levels of air pollutants and meteorological variables for all county-year combinations. The three-year average smoke PM2.5 was highly right-skewed, with the median concentration of 0.37 μg/m3 and a maximum of 7.78 μg/m3. Non-smoke PM2.5 had a more balanced distribution, with an average concentration of 8.38 μg/m3. The average concentrations of NO2 and ozone were 24.73 parts per billion (ppb; standard deviation [SD]: 13.14 ppb) and 74.65 ppb [SD: 39.67 ppb], respectively. Meteorological variables, including temperature and precipitation, varied significantly between seasons. The average summer and winter temperatures were 23.38°C and 1.23°C, respectively. The average summer and winter precipitation levels were 3.28 mm and 2.54 mm, respectively.

Table 2.

Distributions of three-year average levels of air pollutants, meteorological variables, and NDVI.

Variable Min 10th percentile 25th percentile Mean Median 75th percentile 90th percentile Max
Air Pollutant
 Smoke PM2.5, μg/m3 0.01 0.16 0.23 0.43 0.37 0.54 0.71 7.78
 Non-smoke PM2.5, μg/m3 1.73 5.17 6.97 8.38 8.57 9.93 11.14 17.08
 NO2, ppb 1.72 7.88 9.73 13.14 12.19 15.37 19.51 49.93
 Ozone, ppb 26.30 36.47 37.86 39.67 39.48 41.23 43.41 51.60
Meteorological Variable
 Summer temperature, °C 6.39 18.75 21.09 23.38 23.64 26.23 27.71 33.54
 Winter temperature, °C −23.92 −7.03 −3.44 1.23 1.07 6.03 9.74 21.73
 Summer precipitation, mm 0.00 1.41 2.49 3.28 3.34 4.11 4.83 11.8
 Winter precipitation, mm 0.12 0.62 1.17 2.54 2.53 3.51 4.38 14.27
 NDVI 0.110 0.311 0.404 0.510 0.526 0.633 0.678 0.803
Open in a new tab

Figure 1 shows the effects of three-year average air pollutant exposures on hospitalization rates for three cardiovascular outcomes. For total CVD, a 1-μg/m3 increase in smoke PM2.5 exposure at lag 0–2 was associated with a non-statistically significant decrease in hospitalization rate, with an estimated change of −0.879 (95% confidence interval [CI]: −2.528, 0.771) per 1,000 people. The effect became significantly positive at longer lag periods: 4.688 (95% CI: 2.701, 6.675) at lag 1–3, 7.514 (95% CI: 5.127, 9.901) at lag 2–4, and 7.538 (95% CI: 4.594, 10.481) at lag 3–5. A similar pattern was seen for heart disease, where the effect of smoke PM2.5 was non-significantly negative at lag 0–2 but became significantly positive at the longer lag periods. Compared to smoke PM2.5, non-smoke PM2.5 had positive effects on both total CVD and heart disease across all lag periods, but with smaller effect sizes. In addition, these effects decreased as the lag period extended: for total CVD, a 1-μg/m3 increase of non-smoke PM2.5 exposure at lag 0–2 was associated with a statistically significant increase of 1.602 (95% CI: 1.207, 1.998) in hospitalization rate per 1,000, and the effect decreased to 0.960 (95% CI: 0.330, 1.589) at lag 3–5; for heart disease, the effects decreased from 1.479 (95% CI: 1.145, 1.814) at lag 0–2 to 0.773 (95% CI: 0.240, 1.306) at lag 3–5. NO2 and ozone had much smaller effects per 1-ppd increase in exposure across all lag periods. For stroke, most effect estimates were small and not statistically significant. Numeric results are provided in Table S1 of the Supplementary Information. The effects per interquartile-range increase in each exposure are presented in Figure S1 of the Supplementary Information.

Figure 1.

Figure 1.

Open in a new tab

Percent increases (and 95% CIs) in hospitalization rates for total CVD, heart disease, and stroke per 1 μg/m3 increase in smoke PM2.5, 1 μg/m3 increase in non-smoke PM2.5, 1 ppb increase in NO2, and 1 ppb increase in ozone.

Figures 2–3 show the results of subgroup analyses. For total CVD and heart disease, the effects of smoke PM2.5 across all lag periods were mostly larger among beneficiaries living in more rural or lower-income counties. Similar pattern was seen for non-smoke PM2.5, with several subgroup differences being statistically significant. For NO2, the effects were significantly larger in more rural areas across all lag periods for total CVD and heart disease. Numeric results are provided in Table S2 and S3 of the Supplementary Information.

Figure 2.

Figure 2.

Open in a new tab

Percent increases (and 95% CIs) in hospitalization rates for total CVD, heart disease, and stroke per 1 μg/m3 increase in smoke PM2.5, 1 μg/m3 increase in non-smoke PM2.5, 1 ppb increase in NO2, and 1 ppb increase in ozone, stratified by RUCA. Higher RUCA indicates greater rurality. Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001.

Figure 3.

Figure 3.

Open in a new tab

Percent increases (and 95% CIs) in hospitalization rates for total CVD, heart disease, and stroke per 1 μg/m3 increase in smoke PM2.5, 1 μg/m3 increase in non-smoke PM2.5, 1 ppb increase in NO2, and 1 ppb increase in ozone, stratified median household income. Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001.

The effect estimates for all pollutants remained consistent after further relaxing the “parallel trends” assumption by allowing separate time trends for each region, although the CIs became wider due to reduced statistical power from including additional variables (Figure S2).

Discussion

This study investigated the chronic effects of smoke PM2.5, non-smoke PM2.5, NO2, and ozone on absolute changes in cardiovascular hospitalization rates among the national Medicare beneficiaries between 2006–2016. Specifically, we examined exposures up to five years prior to hospitalization and performed subgroup analyses to identify at-risk subpopulations, using a DID approach to adjust for both measured and unmeasured confounders. For total CVD and heart disease, we found adverse effects of chronic exposures to both smoke PM2.5 and non-smoke PM2.5 on hospitalization rates, with the effects of smoke PM2.5 being larger. Importantly, the effects of smoke PM2.5 persisted for five years, suggesting a long-lasting cardiovascular impact that may accumulate in the human body over time. The effects of non-smoke PM2.5 diminished with longer lag periods, and NO2 and ozone showed much smaller effects per unit increase in exposure. For stroke, the effects for all pollutants were small and mostly non-significant. Subgroup analyses suggested that beneficiaries living in more rural and lower-income counties experienced greater cardiovascular hospitalization risks associated with these pollutants.

Our study is among the few that directly compare the chronic effects of smoke and non-smoke PM2.517. Aguilera et al. reported up to ten times greater respiratory hospitalization risks from smoke PM2.5 compared to non-smoke PM2.5 in a regional study in Southern California12. Consistently, we found that at lag 3–5 years, the cardiovascular effects of smoke PM2.5 were over seven times those of non-smoke PM2.5. Our finding was supported by toxicological evidence, which suggested that the greater toxicity of smoke PM2.5 may be due to its smaller particle size, allowing it to more easily pass through the lungs and enter into circulation, as well as its higher concentration of carbonaceous particles, making it more likely to generate free radicals and cause oxidative stress and inflammation6,46. Collectively, these findings suggest that the current national standards for PM2.5, which are based on mass concentrations, may not be adequate. Given that smoke PM2.5 is more toxic and poses substantially greater risks than non-smoke PM2.5, more attention should be shifted toward strategies for wildfire control and smoke exposure mitigation, rather than relying solely on traditional emission controls7. For example, prescribed fires are effective in reducing wildfires but are currently targeted toward protecting property, rather than to minimize smoke exposure and its associated health impacts.

Most existing studies on the health effects of wildfire smoke have focused on acute exposure14. Only recently has a growing body of research begun to evaluate its chronic impacts. Matz et al. reported that both daily and annual exposures to smoke PM2.5 were associated with cardiorespiratory diseases in a Canadian population, with annual exposure showing much larger effect13. More recently, a national study by Ma et al. found that annual exposure to smoke PM2.5 was associated with increased cardiovascular mortality16. In our study, we extended the exposure window to up to five years. We found non-significant effects of smoke PM2.5 during lag 0–2, possibly due to exposure misclassification: beneficiaries hospitalized early in the three-year period were assigned the average exposure across the entire period. However, the effects became larger and statistically significant during lag 3–5 year, when there was no exposure misclassification since all hospitalizations occurred after the exposure window. These findings suggested that the impacts of smoke PM2.5 may persist for five years after exposure, much longer than previous studies have evaluated14.

In addition to smoke and non-smoke PM2.5, we found that ozone also had persistent effects on total CVD and heart disease, lasting for at least five years. This reinforced recent advances in studies of ozone’s chronic cardiovascular effects, which had increasingly expanded the exposure window for years47. Niu et al. reported that a 3-year average ozone exposure was associated with increased cardiovascular mortality48, and a more recent study by Zhang et al. reported an effect with 5-year ozone exposure49. Several plausible mechanisms have been proposed to explain how ozone may chronically contribute to CVD: chronic ozone exposure can trigger systematic inflammation and coagulation, all of which can increase CVD hospitalizations through cumulative effects over time50. Additionally, chronic ozone exposure may impair lipid metabolism and contribute to insulin resistance, eventually leading to diabetes, a major risk factor for CVD47. When comparing the chronic effects of ozone and PM2.5 on CVD, the relative effect sizes per unit increase were consistent with our previous study among national Medicare beneficiaries, the same population in the current analysis44. In that study, we found that the effect of total PM2.5 was approximately five times greater than that of ozone. In the current analysis, non-smoke PM2.5 showed about three times the effect of ozone. Since we separated total PM2.5 into smoke and non-smoke components, and found smoke PM2.5 to be much more harmful, the combined effect of total PM2.5 is likely be consistent with our previous finding. For NO2, the smaller effect sizes for total CVD and heart disease, compared to non-smoke PM2.5 and ozone, were also consistent with our previous study44. As a key marker of traffic-related air pollution, NO2 can react chemically in the air to form both ozone and secondary PM2.5, which may partly explain health effects that are originally attributed to NO2 exposure51.

Studies on chronic exposures to air pollutants on stroke hospitalization and mortality have reported mixed results52. Consistent with Oudin et al.53, Stafoggia et al.54, and Danesh Yazdi et al.44, we found little evidence on the chronic effects of air pollution on stroke. The underlying mechanisms are similar to those proposed for heart disease, including systemic inflammation, coagulation, and vasoconstriction. Over time, these processes may contribute to the development of common cardiovascular risk factors such as hypertension, cardiac arrhythmia, and accelerated atherosclerosis, which may not necessarily lead to stroke52. It is also possible that air pollution is more strongly associated with stroke onset rather than severe outcomes such as hospitalization and mortality, since more consistent findings have been reported for incident stroke in previous studies55.

We chose the DID approach because it provided stronger control for unmeasured spatial and temporal confounders by adjusting for county-specific dummy variables and categorical calendar year indicators. However, even if the “parallel trends” assumption held, that is, each county would have experienced the same change in hospitalization rates over time per unit increase in each exposure across different exposure levels, the cross-sectional nature of the study brought the possibility of reverse causation. Nevertheless, our results for lagged exposures at 3–5 years, during which hospitalizations occurred completely after the exposure, suggested that reverse causation was unlikely. Overall, the causal interpretation of our results should be made cautiously, and additional evidence is needed to establish the relationship.

Although the distribution of daily smoke PM2.5 levels is right skewed with spikes after wildfire onset, we used three-year exposure window to reflect the cumulative exposure burden. While the number of these spikes could be measured by counting “smoke days”, using this exposure metric has several caveats. First, setting a threshold for what qualified as a “smoke day” would be arbitrary as there is no clear biological rationale for this. In addition, treating 1 μg/m3 and 10 μg/m3 of smoke PM2.5 exposure as the same “smoke day” would lead to a loss of information. As one of our main objectives is to compare the effects of smoke and non-smoke PM2.5, using consistent exposure metrics (i.e., the three-year average) makes the estimates comparable.

Our study has several strengths. First, we directly compared the effects of smoke and non-smoke PM2.5, providing straightforward epidemiologic evidence on their relative effect sizes. Second, the use of DID approach provided some assurance that our findings are robust to unmeasured confounding. The consistency in the relative effect sizes of non-smoke PM2.5, NO2, and ozone further supported the reliability of our estimates. Third, the analysis was conducted on a national scale, covering the full spatial and temporal range, which enhanced the generalizability of our findings. Finally, the absolute effect estimates provided insights into the actual magnitude of impact, which was usually of greater public health importance than relative measures, e.g., risk ratio15.

Our study also has limitations. First, because wildfires occur only during limited periods each year, averaging smoke PM2.5 levels over a three-year exposure window reduced the variability of exposure, which may have lowered statistical power and masked the detection of effects. This may explain the wider CIs of smoke PM2.5’s effect estimates compared to other pollutants. Second, as an open sourced dataset, the hospitalization data were only available at the county level, which was relatively coarse and made the exposure estimates subject to measurement error. Due to the within-county variability, the primary error source was Berkson, as we assigned county-level average PM2.5 exposures to all individuals within a county. Another error source arose when an individual lived in a different county from where they were hospitalized. Assuming that whether and where individuals lived outside the hospitalization county was independent of exposure levels, this would introduce classical error. In our previous work, we have shown that Berkson error typically increased variability and widened confidence intervals without introducing bias, whereas classical error may attenuate effect estimates toward the null56.

Conclusion

In this study of the national Medicare beneficiaries, we evaluated the effects of chronic exposures to smoke PM2.5, non-smoke PM2.5, NO2, and ozone on absolute changes in cardiovascular hospitalization rates. For total CVD and heart disease, smoke PM2.5 showed the largest effect per unit increase in exposure among all the pollutants. Smoke PM2.5 also showed a long-lasting impact that may accumulate in human body over years. For stroke, the effects for all pollutants were small and mostly non-significant. Beneficiaries living in more rural and lower-income counties experienced greater impacts from those pollutants. These findings suggested that smoke PM2.5 was more harmful than non-smoke PM2.5 and other criteria pollutants, highlighting the need to prioritize wildfire management in addition to traditional air quality control strategies.

Supplementary Material

1

  • Smoke PM2.5 posed greater CVD risk than non-smoke PM2.5 NO2, and ozone.

  • Effect of smoke PM2.5 on CVD could persist for five years after exposure.

  • More rural and lower-income areas experienced greater risk.

Funding sources

This study was funded by the National Institute of Health grants P30ES023515, UL1TR004419, and R01ES036566.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Conflicts of interest

None declared.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Reference

  • 1.Thangavel P, Park D, Lee Y-C. Recent Insights into Particulate Matter (PM2.5)-Mediated Toxicity in Humans: An Overview. International Journal of Environmental Research and Public Health. [Google Scholar]
  • 2.Burke M, Childs ML, de la Cuesta B, et al. The contribution of wildfire to PM2.5 trends in the USA. Nature 2023;622(7984):761–766. DOI: 10.1038/s41586-023-06522-6. [DOI] [PubMed] [Google Scholar]
  • 3.Zhang D, Wang W, Xi Y, et al. Wildfire worsens population exposure to PM2.5 pollution in the Continental United States. Res Sq 2023. (In eng). DOI: 10.21203/rs.3.rs-3345091/v2. [DOI] [Google Scholar]
  • 4.Westerling AL. Increasing western US forest wildfire activity: sensitivity to changes in the timing of spring. Philosophical Transactions of the Royal Society B: Biological Sciences 2016;371(1696):20150178. DOI: 10.1098/rstb.2015.0178. [DOI] [Google Scholar]
  • 5.Møller P, R. JN, K FJ, et al. Role of oxidative damage in toxicity of particulates. Free Radical Research 2010;44(1):1–46. DOI: 10.3109/10715760903300691. [DOI] [PubMed] [Google Scholar]
  • 6.Enayati Ahangar F, Pakbin P, Hasheminassab S, et al. Long-term trends of PM2.5 and its carbon content in the South Coast Air Basin: A focus on the impact of wildfires. Atmospheric Environment 2021;255:118431. DOI: 10.1016/j.atmosenv.2021.118431. [DOI] [Google Scholar]
  • 7.Jaffe DA, O’Neill SM, Larkin NK, et al. Wildfire and prescribed burning impacts on air quality in the United States. Journal of the Air & Waste Management Association 2020;70(6):583–615. DOI: 10.1080/10962247.2020.1749731. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.O’Dell K, Ford B, Fischer EV, Pierce JR. Contribution of Wildland-Fire Smoke to US PM2.5 and Its Influence on Recent Trends. Environ Sci Technol 2019;53(4):1797–1804. DOI: 10.1021/acs.est.8b05430. [DOI] [PubMed] [Google Scholar]
  • 9.Childs ML, Li J, Wen J, et al. Daily Local-Level Estimates of Ambient Wildfire Smoke PM2.5 for the Contiguous US. Environ Sci Technol 2022;56(19):13607–13621. DOI: 10.1021/acs.est.2c02934. [DOI] [PubMed] [Google Scholar]
  • 10.Orr A, Adam Claire E, Graham J, et al. A State of the Science Review of Wildfire-Specific Fine Particulate Matter Data Sources, Methods, and Models. Environmental Health Perspectives;133(6):066001. DOI: 10.1289/EHP15672. [DOI] [Google Scholar]
  • 11.Chen G, Guo Y, Yue X, et al. Mortality risk attributable to wildfire-related PM2·5 pollution: a global time series study in 749 locations. The Lancet Planetary Health 2021;5(9):e579–e587. DOI: 10.1016/S2542-5196(21)00200-X. [DOI] [PubMed] [Google Scholar]
  • 12.Aguilera R, Corringham T, Gershunov A, Benmarhnia T. Wildfire smoke impacts respiratory health more than fine particles from other sources: observational evidence from Southern California. Nature Communications 2021;12(1):1493. DOI: 10.1038/s41467-021-21708-0. [DOI] [Google Scholar]
  • 13.Matz CJ, Egyed M, Xi G, et al. Health impact analysis of PM2.5 from wildfire smoke in Canada (2013–2015, 2017–2018). Science of The Total Environment 2020;725:138506. DOI: 10.1016/j.scitotenv.2020.138506. [DOI] [PubMed] [Google Scholar]
  • 14.Yu Y, Zou W, Jerrett M, Meng Y-Y. Acute health impact of wildfire-related and conventional PM2.5 in the United States: A narrative review. Environmental Advances 2023;12:100179. DOI: 10.1016/j.envadv.2022.100179. [DOI] [Google Scholar]
  • 15.Wei Y, Wang Y, Wu X, et al. Causal Effects of Air Pollution on Mortality in Massachusetts. Am J Epidemiol 2020. DOI: 10.1093/aje/kwaa098. [DOI] [Google Scholar]
  • 16.Ma Y, Zang E, Liu Y, et al. Long-term exposure to wildland fire smoke PM2.5 and mortality in the contiguous United States. Proceedings of the National Academy of Sciences 2024;121(40):e2403960121. DOI: 10.1073/pnas.2403960121. [DOI] [Google Scholar]
  • 17.Hao H, Xu K, Zhang D, et al. Long-Term Wildfire Smoke Exposure and Increased Risk of Heart Failure in Older Adults. Journal of the American College of Cardiology 2025;85(25):2439–2451. DOI: 10.1016/j.jacc.2025.04.058. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Yaguang W, Yijing F, Mahdieh Danesh Y, et al. Exposure-response associations between chronic exposure to fine particulate matter and risks of hospital admission for major cardiovascular diseases: population based cohort study. BMJ 2024;384:e076939. DOI: 10.1136/bmj-2023-076939. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Schwartz J, Wei Y, Dominici F, Yazdi MD. Effects of low-level air pollution exposures on hospital admission for myocardial infarction using multiple causal models. Environmental Research 2023;232:116203. DOI: 10.1016/j.envres.2023.116203. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Wei Y, Coull B, Koutrakis P, et al. Assessing additive effects of air pollutants on mortality rate in Massachusetts. Environmental Health 2021;20(1):19. DOI: 10.1186/s12940-021-00704-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Announcement: Interactive Atlas of Heart Disease and Stroke Available Online. JAMA 2012;308(9):858–858. [Google Scholar]
  • 22.Mitchell JB, Bubolz T, Paul JE, et al. Using Medicare Claims for Outcomes Research. Medical Care 1994;32(7) (https://journals.lww.com/lww-medicalcare/fulltext/1994/07001/using_medicare_claims_for_outcomes_research.4.aspx). [Google Scholar]
  • 23.Childs M, Li J, Wen J, et al. Daily local-level estimates of ambient wildfire smoke PM2.5 for the contiguous US. V1 ed: Harvard Dataverse; 2024. [Google Scholar]
  • 24.Di Q, Amini H, Shi L, et al. An Ensemble-based Model of PM2.5 Concentration Across the Contiguous United States with High Spatiotemporal Resolution. Environment International 2019;130:104909. ( 10.1016/j.envint.2019.104909). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Di Q, Wei Y, Shtein A, et al. Daily and Annual NO2 Concentrations for the Contiguous United States, 1-km Grids, Version 1.10 (2000-2016). Palisades, New York: NASA Socioeconomic Data and Applications Center (SEDAC); 2024. [Google Scholar]
  • 26.Di Q, Amini H, Shi L, et al. Assessing NO2 Concentration and Model Uncertainty with High Spatiotemporal Resolution across the Contiguous United States. Environmental Science & Technology 2020 2019;54(3):1372–1384. ( 10.1021/acs.est.9b03358). [DOI] [Google Scholar]
  • 27.Requia WJ, Di Q, Silvern R, et al. An Ensemble Learning Approach for Estimating High Spatiotemporal Resolution of Ground-level Ozone in the Contiguous U.S. Environ Sci Technol 2020;54(18):11037–11047. ( 10.1021/acs.est.0c01791). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Requia WJ, Wei Y, Shtein A, et al. Daily 8-Hour Maximum and Annual O3 Concentrations for the Contiguous United States, 1-km Grids, Version 1.10 (2000-2016). Palisades, New York: NASA Socioeconomic Data and Applications Center (SEDAC); 2024. [Google Scholar]
  • 29.Zhu H, Lin K, Liu W, et al. Mediating effect of the normalised difference vegetation index between PM2.5 and its components and cardiovascular disease: evidence from CHARLS. BMJ Open 2025;15(6):e100236. DOI: 10.1136/bmjopen-2025-100236. [DOI] [Google Scholar]
  • 30.Castro E, Healy J, Liu A, Wei Y, Kosheleva A, Schwartz J. Interactive effects between extreme temperatures and PM2.5 on cause-specific mortality in thirteen U.S. states. Environmental Research Letters 2025;20(1):014011. DOI: 10.1088/1748-9326/ad97d1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Jbaily A, Zhou X, Liu J, et al. Air pollution exposure disparities across US population and income groups. Nature 2022;601(7892):228–233. DOI: 10.1038/s41586-021-04190-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Patel L, Friedman E, Johannes SA, Lee SS, O’Brien HG, Schear SE. Air pollution as a social and structural determinant of health. The Journal of Climate Change and Health 2021;3:100035. DOI: 10.1016/j.joclim.2021.100035. [DOI] [Google Scholar]
  • 33.Thornton MM, Shrestha R, Wei Y, Thornton PE, Kao SC. Daymet: Daily Surface Weather Data on a 1-km Grid for North America, Version 4 R1. ORNL Distributed Active Archive Center; 2022. [Google Scholar]
  • 34.Wei Y, Wang Y, Lin C-K, et al. Associations between seasonal temperature and dementia-associated hospitalizations in New England. Environment International 2019;126:228–233. DOI: 10.1016/j.envint.2018.12.054. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Bureau UC. 2010 Census Summary File 1—United States. 2011. [Google Scholar]
  • 36.Bureau USC. US summary: 2000: US Census Bureau, 2002. [Google Scholar]
  • 37.Bureau UC. American Community Survey 1-Year Data (2011-2018). [Google Scholar]
  • 38.Cronenwett JL, Birkmeyer JD. The Dartmouth Atlas of Vascular Health Care. Cardiovasc Surg 2000;8(6):409–10. (https://www.ncbi.nlm.nih.gov/pubmed/11184739). [DOI] [PubMed] [Google Scholar]
  • 39.Li P, Stuart EA, Allison DB. Multiple Imputation: A Flexible Tool for Handling Missing Data. JAMA 2015;314(18):1966–1967. DOI: 10.1001/jama.2015.15281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Bari E, Nipa NJ, Roy B. Association of vegetation indices with atmospheric & biological factors using MODIS time series products. Environmental Challenges 2021;5:100376. DOI: 10.1016/j.envc.2021.100376. [DOI] [Google Scholar]
  • 41.Schwartz J, Wei Y, Yitshak-Sade Ma, Di Q, Dominici F, Zanobetti A. A national difference in differences analysis of the effect of PM2.5 on annual death rates. Environmental Research 2021;194:110649. DOI: 10.1016/j.envres.2020.110649. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Wang Y, Qiu X, Wei Y, Schwartz JD. Long-Term Exposure to Ambient PM2.5 and Hospitalizations for Myocardial Infarction Among US Residents: A Difference-in-Differences Analysis. Journal of the American Heart Association 2023;12(18):e029428. DOI: 10.1161/JAHA.123.029428. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Wang G, Hamad R, White JS. Advances in Difference-in-differences Methods for Policy Evaluation Research. Epidemiology 2024;35(5) (https://journals.lww.com/epidem/fulltext/2024/09000/advances_in_difference_in_differences_methods_for.6.aspx). [Google Scholar]
  • 44.Danesh Yazdi M, Wei Y, Di Q, et al. The effect of long-term exposure to air pollution and seasonal temperature on hospital admissions with cardiovascular and respiratory disease in the United States: A difference-in-differences analysis. Science of The Total Environment 2022;843:156855. DOI: 10.1016/j.scitotenv.2022.156855. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Hall SA, Kaufman JS, Ricketts TC. Defining Urban and Rural Areas in U.S. Epidemiologic Studies. Journal of Urban Health 2006;83(2):162–175. DOI: 10.1007/s11524-005-9016-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Wu W, Jin Y, Carlsten C. Inflammatory health effects of indoor and outdoor particulate matter. Journal of Allergy and Clinical Immunology 2018;141(3):833–844. DOI: 10.1016/j.jaci.2017.12.981. [DOI] [PubMed] [Google Scholar]
  • 47.Donzelli G, Suarez-Varela MM. Tropospheric Ozone: A Critical Review of the Literature on Emissions, Exposure, and Health Effects. Atmosphere. [Google Scholar]
  • 48.Niu Y, Zhou Y, Chen R, et al. Long-term exposure to ozone and cardiovascular mortality in China: a nationwide cohort study. The Lancet Planetary Health 2022;6(6):e496–e503. DOI: 10.1016/S2542-5196(22)00093-6. [DOI] [PubMed] [Google Scholar]
  • 49.Zhang Z, Wang C, Lin C, et al. Association of long-term exposure to ozone with cardiovascular mortality and its metabolic mediators: evidence from a nationwide, population-based, prospective cohort study. The Lancet Regional Health - Western Pacific 2024;52:101222. DOI: 10.1016/j.lanwpc.2024.101222. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Wang T, Wang J, Sun L, et al. Effect of Ozone Exposure on Cardiovascular and Cerebrovascular Disease Mortality in the Elderly. Toxics. [Google Scholar]
  • 51.Wang Y, Ding D, Kang N, et al. Effects of combined exposure to PM2.5, O3, and NO2 on health risks of different disease populations in the Beijing-Tianjin-Hebei region. Science of The Total Environment 2025;958:178103. DOI: 10.1016/j.scitotenv.2024.178103. [DOI] [PubMed] [Google Scholar]
  • 52.Kulick ER, Kaufman JD, Sack C. Ambient Air Pollution and Stroke: An Updated Review. Stroke 2023;54(3):882–893. DOI: 10.1161/STROKEAHA.122.035498. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Alexeeff SE, Deosaransingh K, Van Den Eeden S, Schwartz J, Liao NS, Sidney S. Association of Long-term Exposure to Particulate Air Pollution With Cardiovascular Events in California. JAMA Network Open 2023;6(2):e230561–e230561. DOI: 10.1001/jamanetworkopen.2023.0561. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Stafoggia M, Cesaroni G, Peters A, et al. Long-Term Exposure to Ambient Air Pollution and Incidence of Cerebrovascular Events: Results from 11 European Cohorts within the ESCAPE Project. Environmental Health Perspectives 2014;122(9):919–925. DOI: 10.1289/ehp.1307301. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Lee KK, Miller MR, Shah ASV. Air Pollution and Stroke. J Stroke 2018;20(1):2–11. DOI: 10.5853/jos.2017.02894. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Wei Y, Qiu X, Yazdi MD, et al. The Impact of Exposure Measurement Error on the Estimated Concentration-Response Relationship between Long-Term Exposure to 2.5 and Mortality. Environmental Health Perspectives 2022;130(7):077006. DOI: doi: 10.1289/EHP10389. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS2102806-supplement-1.docx (362.9KB, docx)

RESOURCES