Effect of Air Pollution on Heart Failure: Systematic Review and Meta-Analysis

Heart failure (HF) is a complex clinical syndrome often representing the final stage of multiple cardiovascular diseases and remains highly prevalent worldwide despite declining age-standardized incidence. In 2017, an estimated 63.4 million people were living with HF, with substantial increases in years lived with disability. Air pollution contributes to millions of premature deaths globally, mainly via fine and ultrafine particulate matter (PM2.5 and UFP). Observational and experimental studies have linked air pollution to coronary artery disease, stroke, and myocardial infarction. Prior meta-analyses (e.g., Shah et al., 2013) reported short-term associations between air pollution and HF outcomes but were dominated by high-income countries with relatively low pollution levels. Newer studies from low- and middle-income countries (LMICs) and on long-term exposures have produced inconsistent findings, potentially due to differences in populations, designs, pollutant types and concentrations, and confounding control. This review and meta-analysis aimed to comprehensively assess associations of both short- and long-term exposure to particulate and gaseous air pollutants with HF hospitalization, incidence, and mortality in adults, and to inform prevention strategies.

Previous evidence indicated short-term increases in HF hospitalizations or deaths following exposure to particulate and gaseous pollutants, with prior meta-analyses largely reflecting data from high-income settings and lower pollutant concentrations. Over the past decade, a growing body of studies from LMICs and on long-term exposure has emerged. However, results have been inconsistent: for example, a cohort in Toronto reported positive associations between long-term PM2.5 and congestive HF, whereas a large South Korean cohort reported negative associations. Differences in design, demographics, pollutant concentrations, exposure assessment methods, and outcome definitions contribute to heterogeneity across studies, underscoring the need for an updated, comprehensive synthesis across settings and exposure windows.

• Design: Systematic review and meta-analysis following 2020 PRISMA guidance and the Navigation Guide for risk-of-bias assessment.
• Search strategy: Three databases searched through 31 August 2022 for epidemiologic studies assessing associations between air pollutants (PM2.5, PM10, NO2, SO2, CO, O3) and HF outcomes (hospitalization, incidence, mortality) in adults.
• Eligibility and grouping: 100 studies included, grouped by exposure window: short-term (≤30 days; n=81) and long-term (≥1 year; n=19). No studies with intervals between 30 days and 1 year were eligible. Short-term studies included time-series and case-crossover designs (plus two cohort analyses using survival models). All long-term studies were cohort designs.
• Data extraction and standardization: Extracted adjusted relative risks (RRs) and 95% CIs. Standardized increments: 10 µg/m3 for PM2.5 and PM10; 10 ppb for NO2, SO2, O3; 1 ppm for CO. Assumed linear exposure–response consistent with generalized linear modeling in source studies. When multiple outcomes were reported from the same population, hospitalization with a larger number of events was used for overall risk; hospitalization and mortality were also analyzed separately in stratified analyses when reported. For repeated or overlapping populations, prioritized time-series analyses (short-term), longer follow-up (long-term), broader pollutant coverage, or updated reports. Stratified estimates (e.g., by location, age, temperature) were incorporated to reflect overall risk. For short-term studies with multiple lags, estimates were pooled separately by lag; the shortest lag in each study contributed to the overall short-term estimate. For long-term studies with varied exposure windows, the most significant estimate per study was used for overall pooling.
• Risk of bias: Assessed per Navigation Guide across recruitment, blinding, confounding, exposure assessment, incomplete outcome data, selective reporting, conflicts of interest, and other biases. Dual independent assessment with adjudication; conservative judgments used when disagreements persisted.
• Statistical analysis: Random-effects meta-analyses conducted when ≥2 studies assessed the same pollutant–outcome relation. Heterogeneity quantified using I2; I2 ≥75% considered considerable. Publication bias evaluated using funnel plots and Egger’s test (when ≥10 studies), with p<0.05 indicating significant bias; Trim-and-Fill used to adjust for potential bias. Stratified analyses (short-term) by geography, age, outcome, design, covered area, exposure assessment method, and exposure window length; (long-term) by covered area, exposure window length, exposure stage, and geography. Meta-regression (for pollutants with ≥10 studies) explored heterogeneity sources, including publication year, sample size, population characteristics, region, exposure assessment, exposure window (long-term), outcome definition (short-term), and design (short-term). Sensitivity analyses excluded studies with extreme effect sizes, special periods (e.g., wildfire/storm), small sample sizes (<10,000), special populations (CVD/HF patients), and high risk of bias.
• Software: Stata 16.0. Two-sided p<0.05 considered significant.

• Corpus: 100 studies across 20 countries (81 short-term; 19 long-term) covering wide pollutant concentration ranges (e.g., PM2.5: 2.9–102.1 µg/m3; PM10: 13.0–131.5 µg/m3; NO2: 6.57–77.03 ppb; SO2: 0.92–32.01 ppb; CO: 0.002–5.60 ppm; O3: 1.88–95.66 ppb).
• Short-term exposure (≤30 days):
  • PM2.5: RR per 10 µg/m3 = 1.018 (95% CI: 1.011, 1.025), indicating a 1.8% increased HF risk.
  • PM10: RR per 10 µg/m3 = 1.016 (95% CI: 1.011, 1.020), indicating a 1.6% increased HF risk.
  • NO2, SO2, CO: Significant positive associations with HF (overall RRs ~1.010–1.037); O3 overall not significant [RR = 1.010 (95% CI: 0.998, 1.021)].
  • Lag structure: Stronger associations for multi-day lag 0–1 compared with single-day lag 0; effects generally attenuated over 3 days but remained detectable up to 7 days. O3 significantly associated with HF at lag 0–6 only [RR = 1.025 (95% CI: 1.008, 1.042)].
  • Subgroups: Stronger adverse associations in LMICs than in HICs; generally larger effects in Asia vs. Europe/North America; Oceania studies (PM2.5 only) showed higher combined risk (RR = 1.081, 95% CI: 1.019, 1.147). Monitoring-station-based exposure assessments tended to yield larger effects than modeled exposures. Time-series and case-crossover designs showed similar results; single vs. multicity coverage yielded similar associations.
• Long-term exposure (≥1 year):
  • PM2.5: RR per 10 µg/m3 = 1.748 (95% CI: 1.112, 2.747) for HF incidence/mortality (pooled most significant estimates). In publication-bias table, non-adjusted pooled RR = 1.708 (1.034, 2.821); after Trim-and-Fill, 2.219 (1.294, 3.807).
  • PM10: RR per 10 µg/m3 = 1.212 (95% CI: 1.010, 1.454).
  • NO2: RR per 10 ppb = 1.204 (95% CI: 1.069, 1.356).
  • O3, CO, SO2: Long-term associations generally not statistically significant overall.
  • Subgroups: Similar estimates across exposure window lengths (1 year vs. >1 year), exposure stage (prior to outcome vs. prior to baseline), covered area (single vs. multicity), and geographic locations for PM2.5, NO2, and O3.
• Heterogeneity and bias:
  • Considerable heterogeneity (I2 often >90%) across pollutants and exposure terms. For short-term exposure, meta-regression did not identify substantial sources of heterogeneity. For long-term PM2.5, sample size, exposure assessment, exposure window, population selection, and continent explained some heterogeneity; for long-term NO2, publication year and sample size contributed.
  • Publication bias: Minimal for short-term PM2.5; apparent for some other short-term pollutants, but adjusted results changed little and were often slightly larger. No significant publication bias detected for long-term exposures at test level, and adjustments did not materially alter results.
• Overall: Both short- and long-term exposures to PM2.5, PM10, and NO2 were significantly associated with increased HF risk. Short-term SO2 and CO were significant; O3 was generally not significant in either exposure term.

The meta-analysis addresses the central question by demonstrating consistent adverse associations between ambient air pollution and HF across diverse settings and exposure windows. The findings extend prior work by incorporating substantial evidence from LMICs and long-term exposure studies, revealing stronger associations in LMICs—likely reflecting higher pollutant concentrations and potentially differing pollutant sources and population susceptibilities. Short-term effects were more pronounced for multiday lag windows (lag 0–1), supporting biologically plausible delayed responses. Long-term associations for PM2.5, PM10, and NO2 suggest cumulative impacts on HF incidence and mortality. These results align with mechanistic evidence implicating oxidative stress, systemic inflammation, autonomic imbalance, HPA-axis activation, endothelial dysfunction, atherosclerosis progression, and myocardial remodeling as pathways linking air pollution to HF development and exacerbation. The synthesis underscores the public health significance of air pollution control measures in reducing HF burden globally, particularly in regions with high pollution levels.

This systematic review and meta-analysis provides comprehensive evidence that exposure to PM2.5, PM10, NO2, SO2, and CO is associated with increased HF risk for short-term exposures, and that long-term exposure to PM2.5, PM10, and NO2 is significantly associated with HF incidence and mortality. O3 associations were generally not significant. The findings highlight the need for sustained environmental and public health policies aimed at reducing ambient air pollution to mitigate HF morbidity and mortality. Future research should jointly evaluate short- and long-term exposures, delineate dose–response relationships, assess multipollutant and component-specific effects, and better characterize susceptible subpopulations and HF phenotypes across diverse settings, including LMICs.

• High between-study heterogeneity across pollutants and exposure terms, partly unexplained by meta-regression (particularly for short-term analyses).
• Evidence of publication bias for several short-term pollutants, although Trim-and-Fill adjustments minimally affected pooled estimates.
• Analyses focused on single-pollutant models; potential synergistic or antagonistic multipollutant effects were not evaluated.
• Sparse long-term evidence for gaseous pollutants (especially CO and SO2) and limited data from LMICs; long-term LMIC evidence was largely from a single Chinese cohort with short follow-up for some outcomes.
• Variability in exposure assessment methods, outcome definitions, and lag/exposure windows may limit comparability and generalizability.