Long-Term Wildfire Smoke Exposure and Increased Risk of Heart Failure in Older Adults

Abstract

BACKGROUND

Fine particulate matter, defined as particles <2.5 μm in diameter (PM_2.5), is the most important environmental risk factor for global mortality. Wildfires have been increasing globally, posing a global health challenge. Fire smoke–specific PM_2.5 is believed to have more cardiovascular toxicity compared with other PM_2.5 sources; however, the impact of fire smoke PM_2.5 and heart failure (HF) remains undefined. We sought to investigate the association between long-term exposure to fire smoke PM_2.5 and HF and compare it with nonfire PM_2.5 in a large national cohort.

OBJECTIVES

This study sought to evaluate the association between long-term exposure to wildfire smoke PM_2.5 and the risk of incident HF among older adults across the contiguous United States.

METHODS

This retrospective cohort study analyzed data from Medicare fee-for-service beneficiaries between 2007 and 2018. We linked individuals to high-resolution exposure estimates of fire smoke and nonfire smoke PM_2.5 at a 1 × 1-km spatial resolution that were aggregated to ZIP (Zone Improvement Plan) codes by averaging the values of all 1-km grid cells whose centroids fell within each ZIP code boundary and followed them for incident HF. The relationship between PM_2.5 and incident HF were examined using Cox proportional hazard models, adjusting for individual demographic characteristics and area-level socioeconomic risk factors.

RESULTS

The cohort included approximately 22 million fee-for-service enrollees, with a follow-up totaling 115 million person-years. The mean smoke PM_2.5 exposure, calculated as the average of all past 2-year moving average exposures across person-years, was 0.51 μg/m³. The HR associated with each 1-μg/m³ increase in the past 2-year average smoke PM_2.5 was 1.014 (95% CI: 1.007–1.020), which was notably higher than the corresponding HR for nonsmoke PM_2.5 (1.005; 95% CI: 1.003–1.006). This association corresponds to an estimated 20,238 (95% CI: 10,727–29,612) additional heart failure cases annually among U.S. older adults. Additionally, the number of days exposed to smoke PM_2.5 exceeding 1 μg/m³ and 2.5 μg/m³ over the past 2 years were both significantly associated with an elevated risk of HF. We additionally found that the association between smoke PM_2.5 and HF was stronger in women, Medicaid-eligible individuals, and those living in lower income areas, indicating higher susceptibility.

CONCLUSIONS

In this national cohort, long-term exposure to fire smoke–related PM_2.5 was linked to a higher risk of HF compared with nonsmoke PM_2.5, with greater susceptibility in women and socially vulnerable populations. These findings underscore the need for targeted interventions and policies to reduce wildfire smoke exposure and its cardiovascular impacts.

CENTRAL ILLUSTRATION

Long-Term Smoke Fine Particulate Matter Exposure and Heart Failure Risk in U.S. Older Adults

PM_2.5 = fine particulate matter with aerodynamic diameter ≤2.5 mm; SES = socioeconomic status.

Heart failure (HF) is a major cause of morbidity and mortality globally. It is estimated that >60 million individuals live with HF globally. In the United States alone, HF affects approximately 6 million adults with an estimated direct cost ranging between 39 and 60 billion U.S. dollars.¹ HF represents an escalating public health burden,² particularly with an increasingly aging population.³ Although age-standardized incidence has declined since 2000, the overall prevalence of HF continues to rise, imposing substantial economic and health care costs while reducing quality of life, particularly among older adults.⁴ Preventive strategies have focused on conventional risk factors, such as hypertension, diabetes, and renal disease, yet a significant portion of the HF burden remains linked to social and environmental determinants.⁵ The American Heart Association has implicated air pollution, especially fine particulate matter (PM_2.5; particle diameter <2.5 μm), as a cause in the development of cardiovascular morbidity and mortality.⁶

Although emissions control policies have effectively reduced ambient PM_2.5 levels across the United States in recent decades,⁷ the rising frequency, intensity, and duration of wildfires pose an escalating public health concern.⁸ Wildfires release significant amounts of smoke, resulting in rapid increases in PM_2.5 levels that can persist for prolonged periods and travel vast distances because of intense heat and prevailing winds, ultimately degrading air quality over a long spatiotemporal range.⁹ In 2021, estimates indicated that wildfires have contributed up to 25% of PM_2.5 levels across the United States,¹⁰ with even higher contributions in some Western regions. These recurring pollution events raise serious concerns about the cumulative health impacts of wildfire smoke exposure, as repeated pollution spikes may contribute to chronic health issues over time.¹¹ Importantly, it is predicted that wildfires are more likely to impact marginalized communities, exacerbating preexisting environmental racism and further contributing to differences in health care.¹¹

Globally, studies have established that PM_2.5 exposure is linked to an elevated risk of cardiovascular outcomes, including HF.¹² However, although the short-term respiratory impacts of wildfire smoke are well documented, evidence for cardiovascular outcomes is less consistent.¹³ Recognition is growing that long-term exposure to fire-specific PM_2.5 may pose distinct and underappreciated health risks, particularly for older adults, yet long-term studies remain limited.¹⁴ Despite this growing awareness, large-scale cohort studies examining the effects of long-term exposure on HF risk are still lacking. To fill these knowledge gaps, we conducted the first national, population-based cohort study to quantify the association between long-term fire smoke PM_2.5 exposure and HF risk among U.S. older adults (≥65 years) from 2007 to 2018. Leveraging nationwide Medicare claims data, we applied high-resolution, smoke-specific PM_2.5 estimates generated by machine learning algorithms to improve the accuracy of exposure assessment, addressing the limitations of previous studies that relied on proxies such as total PM_2.5 concentrations or binary smoky day classifications.¹⁵

METHODS

STUDY POPULATION.

We constructed our study cohort using the national Medicare denominator file and the Chronic Conditions Warehouse (CCW) data set from the Centers for Medicare & Medicaid Services. The CCW claims data include predefined indicators for chronic conditions among fee-for-service Medicare beneficiaries, providing the year of initial diagnosis and corresponding diagnosis codes. The denominator file contains annual enrollment information for each Medicare beneficiary, including demographics, death dates, ZIP code, and Medicaid enrollment status, serving as a proxy indicator for socioeconomic status (SES).

The cohort included all Medicare beneficiaries continuously enrolled in the Fee-For-Service program with both Parts A (hospital insurance) and B (medical insurance) in the contiguous United States from 2007 to 2018. We tracked changes in residential ZIP code for each beneficiary annually to capture exposure variations over time.

Our primary outcome was the incidence of HF. To exclude any individuals with preexisting HF, we started the follow-up period 3 years after each patient’s initial enrollment. This washout period increased the likelihood that only new cases of HF were included, enhancing the accuracy of our incidence estimates. Because we restricted enrollment to beneficiaries aged ≥65 years at baseline, the 3-year washout period resulted in a study population that was aged ≥68 years at the start of follow-up. Participants were then monitored until they developed HF, died, were lost to follow-up, or reached the end of the study period. Importantly, we excluded the washout period from the follow-up time to eliminate immortal time bias,¹⁶ ensuring that actual follow-up began as early as 2010. Data were processed within a secure computing infrastructure compliant with the Health Insurance Portability and Accountability Act at the Rollins School of Public Health, Emory University.

OUTCOME CLASSIFICATION.

HF cases were identified using an algorithm within the CCW, incorporating Medicare claims data from various settings, including inpatient, outpatient, and physician visits, using International Classification of Diseases (ICD)-9th and -10th Revision diagnosis codes. The year of initial HF diagnosis was determined after beneficiaries were enrolled in Medicare Parts A and B. Detailed ICD codes and the algorithms used for classifying HF cases are provided in Supplemental Table 1.

EXPOSURE ASSESSMENT.

Daily fire smoke and nonsmoke PM_2.5 concentrations for the contiguous United States from 2007 to 2018 were estimated at a 1 × 1-km spatial resolution using a comprehensive 2-stage machine learning modeling framework.¹⁷ This framework combined Community Multiscale Air Quality model simulations with satellite observations, meteorological data, and ground-based measurements, including low-cost sensor networks like PurpleAir. Two separate random forest models were developed: one to estimate total PM_2.5 concentrations in smoke-impacted regions and another to estimate nonsmoke (background) PM_2.5 concentrations using data from regions and days classified as “no-smoke” based on Hazard Mapping System smoke plume polygons and a Community Multiscale Air Quality–derived smoke ratio below 0.03. The nonsmoke model was then applied to smoke-impacted regions to predict background PM_2.5 levels, whereas the total PM_2.5 model estimated overall concentrations. Smoke-specific PM_2.5 was calculated as the difference between predicted total and nonsmoke PM_2.5. Cross-validation results showed high agreement between predicted and observed PM_2.5 values at both monthly and annual levels (Pearson correlation R² = 0.9), demonstrating the reliability of the exposure estimates.¹⁷ The wildfire smoke PM_2.5 model also demonstrated strong performance in smoke-impacted grids, with cross-validation R² values of 0.75 (overall), 0.59 (spatial), and 0.67 (temporal) and corresponding Root Mean Square Errors of 4.59 μg/m³, 5.88 μg/m³, and 5.18 μg/m³, respectively, supporting its applicability for nationwide wildfire exposure estimation.

To link exposure to the Medicare beneficiaries, we calculated annual fire smoke and nonsmoke PM_2.5 concentrations at the ZIP Code Tabulation Area level by averaging daily 1-km² estimates whose centroids fell within each ZIP Code Tabulation Area, using spatial boundaries defined by the U.S. Census Bureau. These annual averages were then assigned to each beneficiary based on their ZIP code of residence and calendar year of follow-up, as reported in Medicare enrollment data. Because Medicare updates beneficiary residential addresses annually, we were able to capture residential mobility throughout the study period, reducing potential exposure misclassification.

Long-term exposure was defined as the 2-year moving average of smoke PM_2.5 concentrations during the 2 calendar years before the year of HF diagnosis. For example, for a beneficiary diagnosed with HF in 2016, exposure was calculated using data from 2014 and 2015. To additionally characterize wildfire smoke intensity and frequency, as suggested in prior literature,¹¹ we calculated the annual number of days with fire smoke PM_2.5 exceeding 1 μg/m³ and 2.5 μg/m³. We then defined long-term exposure to smoke-event frequency as the past 2-year annual average number of days above each threshold during the same prediagnosis window. Additionally, our choice of thresholds was supported by the empirical distribution of daily wildfire smoke PM_2.5 in our data set—approximately 1 μg/m³ corresponds to the 75^th percentile and 2.5 μg/m³ to the 95th percentile of daily concentrations. These cutpoints therefore capture a broad range of exposure conditions, from more frequent, moderate smoke days to less common, high-intensity episodes.

COVARIATES.

Individual-level data, including age, sex, race, Medicaid status, and date of death, were obtained from the Medicare denominator file. Comorbidities such as myocardial infarction, atrial fibrillation, diabetes, hypertension, and chronic obstructive pulmonary disease were also obtained from the CCW database and considered potential effect modifiers in the association between wildfire smoke exposure and HF risk. Additional area-level covariates included geographic region (Northeast, Southeast, Midwest, Southwest, and West), ZIP code–level meteorological data (relative humidity and temperature) from the North American Land Data Assimilation System, health care capacity (eg, inpatient bed count) from the American Hospital Association Survey, and behavioral risk factors (eg, smoking rates) from the Behavioral Risk Factor Surveillance System. ZIP code–level socioeconomic factors, such as median household income, percentage Black, percentage of auto transportation, percentage renting, and population, were obtained from the U.S. Census and the American Community Survey. Personal-level alcohol use disorders, drug use disorders, and tobacco use disorders were also obtained from the CCW other chronic or potentially disabling conditions segment.

STATISTICAL ANALYSIS.

We used stratified Cox proportional hazards models with generalized estimating equations (GEEs) to estimate HRs for the risk of incident HF associated with long-term exposure to wildfire smoke PM_2.5. The underlying time scale was calendar year, consistent with the annual updates of Medicare enrollment and diagnostic information. Because wildfire smoke PM_2.5 was modeled as a time-varying covariate (updated annually by ZIP code), the proportional hazards assumption does not apply in the conventional sense and was therefore not formally tested. Subjects were censored at either death or loss of follow-up or end of the study period.

Long-term exposure was evaluated using 3 metrics: the moving averages of smoke PM_2.5 concentrations and the annual number of days with smoke PM_2.5 exceeding 1 μg/m³ and 2.5 μg/m³ over the past 2 years, respectively. Given the low correlation between smoke and nonsmoke PM_2.5 (r = −0.26), both were included simultaneously in the model to estimate their independent associations with heart failure risk.

Temporal and spatial variations were accounted for by including categorical indicators for calendar year and geographic region. Meteorological variables were averaged over the same 2-year exposure window as wildfire smoke PM_2.5 to ensure temporal alignment with the exposure. We included temperature and relative humidity because they are related to both pollution levels and cardiovascular disease risk.^18–20

All covariates were included as linear terms unless noted otherwise. GEEs with robust SEs were used to adjust for residual autocorrelation within ZIP codes. Models were stratified by age at study entry (in 1-year increments), sex, race, and Medicaid eligibility to allow for flexible baseline hazard functions, and HRs with 95% CIs were calculated for each 1-μg/m³ increase in smoke or nonsmoke PM_2.5 concentrations. We did not adjust for preexisting medical comorbidities in the main models, because these conditions are not considered confounders in the context of our research question. Rather, they may lie on the causal pathway between exposure and outcome (as mediators) or introduce bias if adjusted for inappropriately (as potential colliders). In this study, we were specifically interested in exploring how the presence of these comorbidities may modify the association between wildfire smoke PM_2.5 and incident HF in older adults. Therefore, comorbidities were treated as potential effect modifiers rather than adjustment variables. The directed acyclic graph for this study is displayed in Supplemental Figure 1.

We further quantified the HF burden attributable to smoke PM_2.5 exposure using 2 indicators, number of attributable cases and population attributable fraction, both of which reflect the population-level impact of exposure.²¹ Additionally, penalized spline models were used to test for potential nonlinear exposure–response (E-R) relationships between smoke and nonsmoke PM_2.5 and HF. Effect modification by sex, SES-related variables (Medicaid dual enrollment, ZIP code–level median household income), and preexisting comorbidities was assessed by fitting separate models for each subgroup. These effect modifiers were selected based on prior evidence of their role in shaping susceptibility to cardiovascular effects of PM_2.5, as documented in previous studies.²² In addition, we repeated the analysis by state and further aggregated the estimates across different geographic regions using a random-effects meta-analysis to consider potential heterogeneity among states caused by differing wildfire composition, climatic conditions, and population characteristics. Finally, we applied an interaction term (subgroup × PM_2.5) to calculate the P value for interaction.

We conducted several sensitivity studies to assess the robustness of our findings. First, to assess the potential impact of outcome misclassification, we conducted 1 complementary sensitivity analysis. We fit GEE linear regression models to estimate additive effects on HF incidence rates. This modeling approach is less susceptible to bias from non-differential misclassification of outcomes, because random error is absorbed into the residual variance of the outcome, providing more robust effect estimates.²³ In addition, to further reduce the risk of including prevalent HF cases at baseline, we implemented an extended “clean period” of 5 years—excluding individuals who received a diagnosis of HF within the first 5 years of follow-up. We excluded the washout period from the follow-up time to eliminate immortal time bias, ensuring that actual follow-up began as early as 2012. Compared with our primary analysis, which used a 3-year clean period, this longer window increases the likelihood of capturing true incident cases, albeit at the cost of a reduced sample size. Second, we conducted a nonmover analysis by only including enrollees who did not relocate during the follow-up period to reduce potential exposure errors related to address change. Third, we included personal-level variables such as alcohol use disorders, drug use disorders, and tobacco use disorders in the model. These variables were not included in the primary analysis because they were derived from the CCW “Other Chronic or Potentially Disabling Conditions” segment, which uses a different identification window that is not directly compatible with the standard annual CCW algorithm used for other covariates. Moreover, these diagnoses tend to capture more severe cases and may not reflect general behavioral patterns in the population, potentially introducing misclassification bias. Therefore, we restricted their inclusion to sensitivity analyses to assess robustness of our findings. Fourth, to further align with the conceptual exposure framework proposed by Casey et al,¹¹ we added an alternative metric using the threshold of >0 μg/m³ for daily wildfire smoke PM_2.5 to capture any detectable smoke exposure. This sensitivity analysis allowed us to assess the consistency of our findings across a broader range of exposure definitions that reflect varying levels of smoke frequency and duration. Finally, we assigned exposure as the moving averages of smoke PM_2.5 concentrations over the past 3 years to evaluate the effect of longer exposure window on heart failure.

The Rollins High Performance Computing Cluster at Emory University was used for all computations. All statistical analyses were implemented in R version 4.0.2. This study was approved by the Emory Institutional Review Board office.

RESULTS

STUDY POPULATION CHARACTERISTICS.

Table 1 presents the demographic characteristics and exposure summaries of the HF study cohort. The cohort comprised approximately 21.9 million Medicare beneficiaries, with a median follow-up period of 5 years, contributing a total of 114.8 million person-years. The mean age at entry was 74.4 years, with a higher proportion of women (57.6%). Most participants were White (88.2%), followed by Black (6.4%), and individuals of other races (5.4%). Approximately 10.2% of the cohort was eligible for Medicaid, indicating lower SES. During the study period, we observed 4.9 million HF events. Supplemental Table 2 provides additional demographic information at the ZIP code level. Supplemental Figure 2A depicts the incidence of HF per 100,000 Medicare enrollees across the contiguous United States, highlighting clusters of higher rates in the South and Midwest. Supplemental Figure 2B presents the temporal trend of HF incidence among Medicare enrollees from 2010 to 2018, demonstrating a gradual decline over time.

TABLE 1.

Demographic Characteristics and Exposure Summary of the Heart Failure Study Cohort After a 3-Year Washout Period

Number of events	4,854,497
Population size	21,936,621
Total person-years	114,792,473
Median follow-up years	5
Mean age at entry, y (IQR)	74.42 (10.00)
Sex
Male	9,296,082 (42.38)
Female	12,640,539 (57.62)
Race
White	19,352,389 (88.22)
Black	1,398,412 (6.37)
Othera	1,185,820 (5.41)
Medicaid eligibility
Eligible	2,239,698 (10.21)
Ineligible	19,696,923 (89.79)
Air pollution concentrations
Smoke PM2.5
Mean moving average of past 2 y, μg/m3 (IQR)	0.51 (0.47)
Mean annual average days with smoke PM2.5 >1 μg/m3 of past 2 y (IQR)	83.00 (66.00)
Annual average days with smoke PM2.5 >2.5 μg/m3 of past 2 y (IQR)	20.00 (18.00)
Nonsmoke PM2.5
Mean moving average of past 2 y, μg/m3 (IQR)	8.52 (3.39)

Values are n (%) unless otherwise indicated.

^aOther indicates Asian, Hispanic, Native American, Pacific Islander, and multiracial individuals.

PM_2.5 = fine particulate matter with aerodynamic diameter ≤2.5 μm.

AIR POLLUTION LEVELS.

The past 2-year moving average smoke PM_2.5 concentration was 0.51 μg/m³ (Q1-Q3: 0.47 μg/m³), with an average of 83 days (Q1-Q3: 66 days) per year with wildfire smoke PM_2.5 levels exceeding 1 μg/m³ and 20 days (Q1-Q3: 18 days) exceeding 2.5 μg/m³. The past 2-year moving average nonsmoke PM_2.5 concentration was 8.52 μg/m³ (Q1-Q3: 3.39 μg/m³) (Table 1). Additional details on the exposure distributions for each metric are provided in Supplemental Table 3. Figure 1 shows the geographic distribution of smoke and nonsmoke PM_2.5 across the contiguous United States, averaged from 2010 to 2018. Smoke PM_2.5 concentrations were highest in the Southeast (mean: 0.80 μg/m³) and West (mean: 0.74 μg/m³), whereas nonsmoke PM_2.5 concentrations peaked in the Midwest (mean: 9.11 μg/m³) and West (mean: 8.13 μg/m³). Supplemental Figure 3 illustrates the temporal trends of smoke and nonsmoke PM_2.5 during the study period, with smoke PM_2.5 peaking in 2017 and showing significant fluctuations throughout, whereas nonsmoke PM_2.5 exhibited a relatively steady decline over time. The correlation coefficient between smoke PM_2.5 and nonsmoke PM_2.5 was −0.26, indicating a low negative correlation.

FIGURE 1. Smoke and Nonsmoke Fine Particulate Matter Spatial Distribution in the United States.

(A) Mean annual smoke PM_2.5 concentrations (μg/m³) across the contiguous United States during the study period. (B) Mean annual nonsmoke PM_2.5 concentrations (μg/m³) across the contiguous United States during the study period of 2007 to 2018. PM_2.5 = fine particulate matter with aerodynamic diameter ≤2.5 μm.

HEALTH EFFECTS ESTIMATES.

Figure 2 and Supplemental Table 4 present the associations between long-term exposure to smoke PM_2.5, nonsmoke PM_2.5, and HF. Both smoke and nonsmoke PM_2.5 were significantly associated with an increased risk of HF. The HR per 1-μg/m³ increase in smoke PM_2.5 was 1.014 (95% CI: 1.007–1.020), which was notably higher than the corresponding HR for nonsmoke PM_2.5 (1.005; 95% CI: 1.003–1.006) (Central Illustration). Additionally, the annual number of days exposed to smoke PM_2.5 exceeding 1 μg/m³ and 2.5 μg/m³ over the past 2 years were both significantly associated with an elevated risk of HF. For broader interpretability, we also reported effect estimates per 10-μg/m³ and per Q1-Q3 increase in smoke and nonsmoke PM_2.5 in Supplemental Table 4. Under the assumption of a small long-term increase of 1 μg/m³ in smoke PM_2.5 exposure and a baseline HF incidence of 26.5 per 1,000 Medicare beneficiaries, the population attributable risk was 1.34% (95% CI: 0.71–1.96), corresponding to approximately 20,238 (95% CI: 10,727–29,612) additional HF cases annually among U.S. adults aged ≥65 years.

FIGURE 2. HRs of Smoke and Nonsmoke Fine Particulate Matter Exposure on Incident Heart Failure.

(A) HRs of heart failure associated with per 1-μg/m³ increase in mean concentration of smoke and nonsmoke PM_2.5 for moving average of past 2-year exposure window. (B) HRs of heart failure associated with per 10-day increase in annual smoke PM_2.5 (SP) days for moving average of past 2-year exposure window. The estimated HRs were obtained from Cox proportional hazard models, and error bars stand for the 95% CIs. Note that A and B reflect different exposure metrics and are not directly comparable in terms of effect size. The corresponding HR values, including estimates per 10-μg/m³ and per IQR increase, are provided in Supplemental Table 4. Abbreviation as in Figure 1.

Figure 3 illustrates the estimated E-R relationships between smoke and nonsmoke PM_2.5 exposure and HF incidence. In both cases, the E-R curves appear relatively flat at lower concentrations and become approximately linear beyond ~0.3 μg/m³ for smoke PM_2.5 (Central Illustration) and ~6 μg/m³ for nonsmoke PM_2.5, indicating increasing risk at higher exposure levels.

FIGURE 3. Exposure-Response Curves.

The exposure–response curves for moving average of past 2-year of smoke (A) and nonsmoke (B) PM_2.5 and heart failure. The exposure–response curves are shown for exposure windows concentration range between 1st and 99th percentiles of the pollutants (ie, with 2% poorly constrained extreme values excluded). Abbreviation as in Figure 1

STRATIFICATION ANALYSIS.

We observed heterogeneity in the associations between smoke PM_2.5 and HF risk across different demographic and community subpopulations. Specifically, the associations were stronger among women, Medicaid-eligible participants (indicative of lower SES), and those living in lower median household income ZIP codes. Additionally, regional variation was evident: The strongest associations were observed in the Southeast and Southwest, whereas weaker or null associations were found in the Midwest, Northeast, and West. Detailed HR estimates can be seen in Table 2.

TABLE 2.

Subgroup Estimates of HRs and 95% CIs by Individual Demographics and Community Contextual Modifiers per 1-μg/m³ Increase in Smoke PM_2.5

Modifier	Group	Smoke PM2.5
		HR (95% CI)a	P Valueb
Sex	Male	1.010 (1.003–1.017)	0.0002
	Female	1.017 (1.010–1.024)
Medicaid eligibility	No	1.012 (1.005–1.018)	<0.0001
	Yes	1.027 (1.017–1.038)
ZIP code-level median household incomec	Quartile 1	1.042 (1.026–1.057)	<0.0001
	Quartile 2	1.017 (1.000–1.035)
	Quartile 3	0.993 (0.983–1.002)
	Quartile 4	0.993 (0.981–1.004)
Number of any of the 5 comorbidities (myocardial infarction, atrial fibrillation, chronic obstructive pulmonary disease, diabetes, and hyperlipidemia)	0	1.022 (1.012–1.033)	<0.0001
	1	1.015 (1.007–1.023)
	2	1.018 (1.011–1.025)
	≥3	1.018 (1.011–1.025)
U.S. geographic region	Midwest	0.980 (0.935–1.027)	<0.0001
	Northeast	0.860 (0.711–1.040)
	Southeast	1.144 (1.120–1.167)
	Southwest	1.055 (1.022–1.090)
	West	1.004 (0.967–1.047)

^aHR per 1-unit increases in annual mean exposure (moving average of past 2-year exposure) adjusting for selected covariates in different subgroups.

^bInteraction term P value.

^cThe ranges for quartiles 1, 2, 3, and 4 are ≤$43,159.37, $43,159.37-$53,791.26, $53,791.27-$70,968.25, >$70,968.25.

PM_2.5 = fine particulate matter with aerodynamic diameter ≤2.5 mm.

SENSITIVITY ANALYSIS.

The associations between long-term exposure to smoke PM_2.5 and HF incidence remained robust across multiple sensitivity analyses. First, we evaluated the potential impact of outcome misclassification through the approach of fitting a linear regression model for HF incidence rates. In the analysis, the result was consistent with or slightly stronger than those from the primary analysis, suggesting that any misclassification of the outcome likely biased our estimates toward the null (Supplemental Table 5). Moreover, a more stringent “cleaner period” that excluded individuals diagnosed with HF during their first 5 years of follow-up produced results consistent with the primary analysis (Supplemental Table 6). Second, restricting the analysis to participants who did not change their residential address throughout the follow-up period yielded findings in agreement with the main analysis, indicating significant positive associations between smoke PM_2.5 exposure and HF risk (Supplemental Table 7). Third, incorporating additional personal-level covariates into the model did not significantly alter the results (Supplemental Table 8). Fourth, using an alternative threshold of >0 μg/m³ to define days with any detectable wildfire smoke PM_2.5 also yielded consistent results with those based on the >1- and >2.5-μg/m³ cutpoints, reinforcing the robustness of our findings (Supplemental Table 9). Finally, extending the exposure window for smoke PM_2.5 produced results consistent with those of the main analysis, further supporting the robustness of the observed associations (Supplemental Table 10).

DISCUSSION

In this large, nationwide population-based cohort of Medicare beneficiaries, we found that long-term exposure to smoke PM_2.5 was associated with an increased risk of HF incidence. Notably, the association between smoke PM_2.5 and HF risk was stronger than that observed for PM_2.5 from other sources, suggesting potentially greater relative toxicity per unit of exposure. Furthermore, we observed stronger associations between smoke PM_2.5 on HF risk with longer exposure durations—a pattern not evident for nonsmoke PM_2.5—suggesting that repeated or prolonged exposure to smoke-related pollution may contribute to greater long-term HF burden. Importantly, our results also reveal that women and individuals with lower SES are particularly susceptible to smoke PM_2.5 exposure, exhibiting a higher risk of developing HF.

Our finding aligns with previously published systematic reviews^15,24 linking smoke exposure to cardiovascular outcomes, including HF. Although many studies have investigated smoke-related health impacts, most have focused on short-term outcomes such as emergency department visits and hospitalizations.^25,26 In contrast, our study highlights the chronic effects of prolonged exposure, and, to date, few investigations have specifically assessed the relationship between long-term fire-specific PM_2.5 exposure and HF incidence. Given the limited body of prior research in this area, direct comparisons to established benchmarks are challenging. Nonetheless, although the individual HR may appear to be modest, the potential population-level impact is considerable because of the widespread exposure to wildfire smoke.²⁷ A long-term increase of just 1 μg/m³ in smoke PM_2.5 exposure is estimated to result in approximately 20,238 additional HF cases annually among U.S. adults aged ≥65 years. These findings underscore that despite small relative risks at the individual level, the cumulative burden of wildfire smoke exposure is substantial and warrants public health attention. We also observed that the impact of smoke PM_2.5 on HF was stronger than that of nonsmoke PM_2.5 per 1-μg/m³ change in exposure. This disparity may be attributed to the distinct characteristics of smoke PM_2.5 compared with other sources.^28,29 Specifically, the unique chemical composition of smoke, including a higher proportion of polar organic compounds³⁰ prone to oxidative reactions, contributes to increased inflammation,³¹ oxidative stress,³² and cardiovascular toxicity. In addition, previous studies have highlighted that smoke PM_2.5 contains elevated levels of toxic metals, such as zinc, arsenic, and cadmium, which are emitted during the incomplete combustion of vegetation.³³ These metals further increase the toxicity of smoke PM_2.5 by generating additional free radicals and exacerbating oxidative stress.³⁴ Smoke is also characterized by a higher proportion of ultrafine particles (diameters ≤0.1 μm),³⁵ which can penetrate deep into the lungs and translocate to the blood-stream, compounding its adverse cardiovascular effects.³⁶ These characteristics may contribute to greater cardiovascular harm from wildfire smoke PM_2.5 compared with PM_2.5 from urban or other sources. Moreover, numerous studies have established associations between cardiovascular disease and natural disasters, including hurricanes, typhoons, and earthquakes.³⁷ In terms of wildfire, the physical and mental stress resulting from the fear of injury and death, property damage or loss, evacuation, potential unemployment, and limited access to health care after such disasters may contribute to an increased incidence of cardiovascular disease.

Our findings on nonsmoke PM_2.5 and HF incidence are generally consistent in direction with previously published meta-analyses,⁴ which report positive associations between long-term ambient PM_2.5 exposure and HF risk. However, interpretation within a U.S. context remains limited. A recent meta-analysis⁴ identified 19 studies on this topic, but only 5 were conducted in the United States, and only Yazdi et al³⁸ specifically assessed HF incidence; the others focused on mortality or hospital readmissions. This limited U.S.-specific evidence, along with differences in study design and exposure assessment, makes direct comparisons difficult. Although the magnitude of our effect estimates for nonsmoke PM_2.5 are smaller than that reported by Yazdi et al³⁸, several contextual factors may contribute. Their study focused on the southeastern United States, where smoke—mainly from prescribed burns—constitutes a larger share of PM_2.5 emissions compared with the nationwide profile in our cohort. They also did not differentiate smoke from nonsmoke PM_2.5, whereas we assessed them separately. Differences in exposure patterns, with prescribed burns occurring more consistently and wildfires more episodically, may also play a role. Additionally, the earlier study period (2000–2012) likely reflected higher ambient PM_2.5 levels; for instance, Jin et al³⁹ reported a mean exposure of 10.2 μg/m³, compared with 8.5 μg/m³ in our cohort. Our findings also suggest that the association between nonsmoke PM_2.5 and HF becomes more apparent above ~6 μg/m³ (Figure 3B), which may partially explain the smaller observed effect. Finally, variation in outcome definitions, such as reliance on hospital admission codes vs incident diagnoses, could influence the comparability of results.

The assessment of the dose–response relationship is crucial for informing public health policies. In our study, the E-R curve for wildfire smoke PM_2.5 and incident HF risk showed a flatter shape with wider CIs at lower concentrations (around 0.5 μg/m³ and below), followed by a more linear increase in risk at moderate to higher concentrations. Few toxicological or epidemiological studies have specifically examined the E-R relationship between smoke PM_2.5 and HF, and there is no clear consensus from existing studies on ambient PM_2.5 and HF. For example, Bai et al⁴⁰ observed a linear E-R curve for the incidence of HF and long-term exposure to ambient PM_2.5 in Canada, but a Chinese study observed no association.⁴¹ These differences underscore the need for future studies to verify our findings.

When exposed to fire smoke, women were observed to be at a higher risk of HF in this study as compared with men. Few studies have specifically examined whether sex modifies the association between smoke exposure and HF; however, prior evidence suggests that women are more sensitive to PM_2.5 exposure overall. For example, Bai et al⁴⁰ followed 5.1 million Canadian adults and observed a stronger association between long-term PM_2.5 exposure and incident HF in women compared with men. Similarly, a meta-analysis by Heo et al,⁴² which pooled data from 7 studies, concluded that women experienced significantly higher rates of HF hospitalization after short-term PM_2.5 exposure. Several biological and hormonal factors may contribute to this increased vulnerability. After menopause, the decline in estrogen, which has cardioprotective properties, may increase susceptibility to the harmful cardiovascular effects of air pollutants.⁴³ Additionally, physiological differences, such as smaller airway diameters and a heightened inflammatory response,⁴⁴ may also play a role in increasing women’s sensitivity to PM_2.5.

Our study also indicates that low SES was another risk factor that may render certain subpopulations more susceptible to smoke PM_2.5-associated HF incidence, and this result aligned with a previous study.²⁶ This increased susceptibility can be attributed to several factors. Lower income and poverty limit the ability to meet basic needs and access protective resources like health care and safe housing.⁴⁵ Chronic exposure to poverty-related stress increases allostatic load, weakening the body’s resilience to additional stressors such as fire smoke.²⁶ Populations with low SES also face higher prevalence of undertreated medical conditions⁴⁶ and have reduced access to health-promoting resources,⁴⁷ exacerbating the adverse health effects of smoke. They may also have limited access to high-efficiency particulate air filters, high-quality masks, and timely public health communications that promote protective behaviors during wildfire events, further increasing their vulnerability.⁴⁸ Furthermore, prolonged social stress can alter endocrine function and contribute to epigenetic changes, compounding their susceptibility to health risks.⁴⁹ Our findings add to the growing evidence of increased sensitivity to air pollution among populations with low SES, emphasizing the importance of focusing on environmental justice to address both their elevated exposure levels and vulnerability.

Our findings that Medicare beneficiaries with multiple comorbidities experienced a higher risk of HF after wildfire exposure are consistent with previous literature.³⁹ We also observed a sensitivity to smoke PM_2.5 among individuals without any of the 5 comorbidities. Potential explanations are that healthier individuals may have more active lifestyles, resulting in greater outdoor exposure to air pollution,⁵⁰ including smoke, compared with those with comorbidities, who may have more restricted activities,⁵¹ and that individuals with comorbidities may be more likely to use medications such as beta-blockers or statins, which can blunt cardiovascular responses to environmental stressors, potentially offering some degree of protection.⁵²

STUDY STRENGTHS AND LIMITATIONS.

Our study has several important strengths. To our knowledge, this is the first large-scale, nationwide cohort study to assess the association between long-term smoke PM_2.5 exposure and HF incidence in the contiguous United States. We included approximately 22 million Medicare enrollees, contributing millions of incident cases, which offered significant statistical power and robustness to our analysis. Unlike most previous studies that primarily focused on proxies such as smoke days or total ambient PM_2.5 concentrations¹⁵ during fire seasons, our study leveraged high-resolution spatial-temporal smoke PM_2.5 exposure data, which allowed more precise estimation of individual exposure levels based on residential ZIP code. This approach enabled us to include participants from both urban and rural areas, where air pollution monitoring data might be lacking, ensuring a comprehensive analysis of the effects of smoke across diverse settings. Furthermore, we applied a 3-year clean period before data collection to improve the likelihood of capturing new HF diagnoses rather than prevalent cases. This design strengthens the inference regarding incidence rather than prevalence and enhances the validity of our findings. Additionally, the study used the CCW database, which integrates data from diverse health care settings, including nursing facilities, inpatient and outpatient care, insurance claims, and home health services. Given that approximately 50% of new HF cases are identified during outpatient visits,⁵³ the inclusion of these settings enhances the accuracy and representativeness of our cohort. Finally, our stratified analysis explored potential individual- and community-level modifiers such as sex, SES, and comorbidity information, allowing for a deeper understanding of the variations in risk among different population subgroups.

Several limitations should be acknowledged. First, despite the strong performance of our exposure model, predicted smoke PM_2.5 concentrations may still contain measurement errors. This residual exposure measurement error may have arisen because we assigned exposure based on residential ZIP code rather than specific home addresses. Nevertheless, any resulting errors are likely independent of the disease outcome status, meaning that the resulting bias would likely attenuate the observed associations toward the null, potentially under-estimating the true effect of smoke exposure. Second, there is a potential for outcome misclassification, as our study relied on Medicare claims data, an administrative health care database that identifies health conditions based on primary ICD diagnosis codes. To mitigate this concern, we implemented a 3-year washout period to exclude individuals with prior HF diagnoses, both of which enhance the likelihood of capturing incident cases. In addition, prior validation studies and meta-analyses^54,55 have reported moderate sensitivity (63%–76%) and high specificity (>90%) for Medicare claims-based HF definitions, supporting the general reliability of this approach. To further address potential misclassification, we applied 1 complementary method and found that the results remained consistent or slightly stronger, suggesting bias toward the null. Additionally, our sensitivity analysis using a more stringent 5-year clean period yielded similar findings, further supporting the robustness of our outcome definition. Third, unmeasured confounding is possible because of the lack of control for individual-level HF risk factors, such as lifestyle habits or family history. However, we used air pollution as the exposure metric, which is generally more closely associated with area-level factors than with individual-level behaviors. By adjusting for these area-level covariates, we aimed to reduce the potential impact of confounding.⁵⁶ Additionally, sensitivity analyses indicated that our results remained robust after incorporating individual-level covariates, supporting the validity of our findings. Moreover, our exposure assessment was conducted at the annual level because of the structure of the Medicare CCW data set, which only updates diagnostic and follow-up information annually. As a result, we were unable to assign more precise, date-specific exposure windows, which may have introduced some degree of temporal exposure misclassification. To minimize this bias—and in line with the study’s focus on long-term exposure—we used a 2-year moving average to capture chronic exposure patterns while ensuring that exposure preceded the onset of HF. Fourth, residual spatial confounding is also a potential concern in studies involving geographically patterned exposures such as wildfire smoke PM_2.5. Although we adjusted for a range of area-level covariates and individual-level characteristics, unmeasured spatially structured factors may remain. To mitigate this, we used several design features, including time-varying exposure and covariate data, adjustment for geographic region and calendar year fixed effects, and clustering at the ZIP code level using a GEE framework. These features enabled us to capture within-ZIP code variation over time, reducing bias from stable unmeasured geographic confounders. Fifth, although our findings suggest a stronger association per unit increase in wildfire smoke PM_2.5 compared with nonsmoke PM_2.5, we acknowledge that these 2 pollutant types differ in their distribution patterns and chemical composition. Therefore, direct comparisons should be interpreted with caution, and additional toxicological and epidemiological research is warranted to better characterize the relative toxicity of different PM_2.5 sources. Finally, our analysis only included Medicare fee-for-service enrollees, meaning that individuals enrolled in Medicare Advantage plans (Medicare-HMO) were not represented. As such, our findings may not be fully generalizable to the entire Medicare population.

CONCLUSIONS

Our study provides significant evidence that long-term exposure to smoke PM_2.5 is associated with an increased risk of HF, and smoke PM_2.5 exhibits greater toxicity compared with PM_2.5 from other sources. Additionally, our findings indicate that women and individuals of lower SES are more vulnerable, experiencing disproportionately higher risks from smoke exposure. This study highlights the critical need for targeted interventions to mitigate the health impacts of smoke, particularly among the most vulnerable populations, and underscores the importance of considering both social determinants of health and environmental justice in these efforts. These findings emphasize the need for targeted public health strategies to protect vulnerable groups from smoke. Improved air quality regulation, enhanced community interventions, and expanded health care resources are crucial for mitigating these health impacts, especially in high-risk populations.

Supplementary Material

APPENDIX For supplemental tables and figures, please see the online version of this paper.

ABBREVIATIONS AND ACRONYMS

CCW

Chronic Conditions Warehouse

E-R

exposure–response

GEE

generalized estimating equation

HF

heart failure

ICD

International Classification of Diseases

PM_2.5

fine particulate matter with aerodynamic diameter ≤2.5 μm

SES

socioeconomic status