Air pollution and cause-specific mortality in EU: a review integrating meta-analysis and meta-regression Authors: Ilaria Stanisci, Andrea Perri, Sandra Baldacci, Sara Maio, Giuseppe Sarno, Sofia Tagliaferro, Velia Malizia, Mirella Profita, Pietro Alfano, Filippo Sapienza, Giovanni Viegi, Stefania La Grutta Published: 2026 ABSTRACT AbstractAmbient air pollution remains a major public health concern in Europe despite recent improvements. This systematic review synthesizes evidence (2020–2025) on short- and long-term effects of PM10, PM2.5, NO₂ and O₃ on cause-specific mortality, integrating Italian BEEP and BIGEPI projects with European studies for the first time. Searches of PubMed, Web of Science, and Scopus (to 31 July 2025) identified 31 studies; meta-analyses were conducted when ≥ 5 studies were available (18 studies included for PM2.5). Risk of bias was assessed using ROBINS-E (v1, June 2023) by two independent reviewers. PM2.5 was significantly associated with mortality: all-cause RR ≈ 1.10 per 10 µg/m³ (I²=97.9%), cardiovascular RR = 1.026 (95%CI 1.014–1.039; I²=91.7%), and respiratory RR = 1.032 (95%CI 1.013–1.050; I²=88.3%), with stronger effects for long-term exposure. Long-term PM2.5 and NO₂ exposures were consistently linked to increased all-cause, cardiovascular, and respiratory mortality; PM10 showed weaker associations. Short-term increases in PM and NO₂ were associated with acute mortality rises, while O₃ evidence was heterogeneous. Limitations include substantial heterogeneity (I² 78–98%), evidence of small-study effects (Egger p < 0.05), and risk of bias rated “Some concerns” in most studies (16/18). Overall, air pollution continues to impose a measurable mortality burden in Europe, supporting stricter standards aligned with WHO guidelines. Funding: PNC PREV- A-2022-12376981.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-57501-6.Subject terms: Diseases, Environmental sciences, Risk factors Introduction Numerous studies have consistently reported adverse associations among long-term or short-term exposure to ambient air pollution and various health outcomes, in particular cause-specific mortality from respiratory and cardiovascular diseases1,2. The global death rate attributable to air pollution is 92/100.000 people, with the highest impact in low-middle income countries (i.e. 195/100.000 in South Asia and 137/100.00 in Africa), and the lowest in high-income countries (16.7/100.000); in Europe a value of 21.2/100.000 is estimated1. The most important outdoor air pollutants include gases (carbon monoxide (CO), sulfur dioxide (SO2), nitrogen dioxide (NO2), ozone (O3)) and particulate matter (PM) of various composition and size, namely thoracic particles (PM10, PM with aerodynamic diameter ≤ 10 μm), coarse particles (PM10-2.5, PM with aerodynamic diameter > 2.5 < 10 μm), fine particles (PM2.5, PM with aerodynamic diameter ≤ 2.5 μm) and ultrafine particles (PM0.1, PM with aerodynamic diameter ≤ 0.1 μm)3. In 2023, air pollution caused 7.9 million deaths: 4.9 million attributable to PM2.5, 2.8 million to household air pollution and 470.000 to O3; 86% of these deaths were due to non-communicable diseases1. Indeed, although the reduction in air pollution levels in recent decades in Europe, some European multicenter studies including millions of subjects consistently identified adverse health effects of chronic exposure to PM, NO2, and O3 on mortality4. Despite the large body of evidence on air pollution and mortality, including several large multicenter European studies, important gaps remain. In particular existing European evidence is often heterogeneous in terms of study design, exposure assessment, pollutant metrics, and population characteristics, which limits the comparability and synthesis of findings across countries. Moreover, most recent meta-analyses have been conducted at a global scale, where European-specific estimates are diluted, while Europe-focused quantitative syntheses remain limited. Furthermore, few studies have jointly evaluated short- and long-term effects of multiple pollutants using harmonized approaches, and the application of meta-regression techniques to systematically investigate sources of heterogeneity across European regions (e.g., differences in climate, population susceptibility, baseline pollution levels, and healthcare systems) is still scarce. These gaps are particularly relevant in light of recent regulatory developments, such as European Directive (EU) 2024/2881 on ambient air quality and cleaner air for Europe5, which introduced substantial changes and updates, with the aim of aligning European standards with the 2021 World Health Organization (WHO) guidelines6. In this evolving policy context, updated and region-specific estimates of the health effects of air pollution are needed to support evidence-based decision-making and to quantify the potential health benefits of stricter air quality standards. Within Europe, Italy represents one of the countries where comprehensive nationwide analysis investigating the association between health outcomes and ambient air pollution has been conducted within the “Big data in Environmental and occupational EPidemiology” (BEEP) and the “Use of BIG data for the evaluation of the acute and chronic health Effects of air Pollution in the Italian population” (BIGEPI) projects, funded by the Italian Workers’ Compensation Authority (INAIL). Such projects have investigated the health effects of air pollution and meteorological parameters on the whole Italian general population. This was achieved by integrating nationwide datasets encompassing land use, satellite information, modelled meteorological fields and atmospheric composition variables, mortality, hospital admissions, morbidity4,7–9. In this perspective, a comprehensive and up-to-date picture of health effects from air pollutants in Europe might provide the basis for a future assessment of the beneficial effects of implementing the new European air quality standards. The aims of this review were: to summarize the evidence on the long and short-term effect of air pollution (PM10, PM2.5, NO2 and O3) on cause-specific mortality obtained in Italy within the framework of BEEP and BIGEPI projects; to contextualize these findings within the most recent European evidence; to quantitatively assess the strength and consistency of associations through meta-analysis and meta-regression, with the aim of exploring heterogeneity across European settings and providing updated estimates relevant for current air quality policies. Methods Literature search A literature search was conducted for epidemiological studies that consider air pollutants and general population health outcomes similar to those of the BEEP and BIGEPI studies4,10–12. This review was conducted in accordance with the PRISMA guidelines. The PRISMA checklist is provided in the Supplementary Material (Table S1-S2). The review protocol was defined a priori and it was not registered in PROSPERO. A PECO (Population, Exposure, Comparator, Outcome) question was developed to search for such studies. Based on the findings of the BEEP and BIGEPI studies, keywords were identified to construct search strings for articles in the databases PubMed, Web of Science and Scopus. Details for each database have been reported in the Table S3. The following pollutants were considered as exposures: PM10, PM2.5, NO2 and O3. The following health outcomes were considered: cardiovascular, respiratory, cardiac, ischemic, cerebrovascular, mental, nervous, metabolic, diabetes mortality. The search was conducted for studies from European sources published in English from 2020 to 2025. Inclusion and exclusion criteria are reported in detail in Table S4. Geographical eligibility was operationalised as follows: studies were included if the study population was drawn from one or more European countries (single-country or pooled multi-country analyses); multi-continental studies were retained only when a separately identifiable European effect estimate was reported, and only the European-specific estimate was carried forward into the quantitative synthesis. Overlap between source populations (Table S5) was assessed before pooling: when two or more eligible papers reported the same outcome from overlapping populations (notably the ELAPSE consortium), the most recent and most comprehensive analysis was retained, while papers reporting non-overlapping outcomes from the same cohort were retained without redundancy. Only peer-reviewed articles with final bibliographic metadata available by the last-search date (31 July 2025) were eligible; preprints and ahead-of-print articles were excluded. Data extraction Two reviewers (IS and AP) independently performed data extraction using a pre‑piloted form, capturing author, year of publication, study design, sample size, population characteristics, exposures and outcome definitions. Duplicate extraction and a standardized template helped ensure consistency and reduce transcription errors13. Duplicate studies were eliminated. A selection based on title and abstract was performed by IS and FS Studies eligible for inclusion in the review were then read in full text. Extracted variables included the effect measure (risk ratio, hazard ratio or odds ratio) and its 95% confidence interval, the exposure increment and duration, the considered confounders, and the study‑level characteristics such as region and publication year. The flowchart describing the article selection process is reported in Fig. 1. Flowchart illustrating the study selection process. Risk of bias assessment Risk of bias in individual studies was assessed using the Risk Of Bias In Non-randomized Studies of Exposures (ROBINS-E) tool, version 1 (June 2023). Two reviewers (IS and FS) independently judged each of the seven domains (confounding, measurement of the exposure, selection of participants, post-exposure interventions, missing data, measurement of the outcome, and selection of the reported result) as Low, Some concerns, High or Very high. Disagreements were resolved by discussion with a third reviewer; the overall judgement for each study was defined as the most severe rating across domains. Study-level ratings are reported in Supplementary Table S6 and visualised in the traffic-light plot in Figure S1. Metanalysis and metaregression Of the 31 articles selected through literature review, those (n = 18) addressing the effect of PM2.5 exposure were analyzed in a metanalysis and metaregression. Indeed, to ensure statistically meaningful results, studies were first grouped by both health outcomes (e.g., all-cause, cardiovascular or respiratory mortality) and pollutant category (such as PM2.5). Each potential meta-analysis or meta-regression was carried out only when there were enough studies to provide a stable pooled estimate. Specifically, we tallied the number of eligible articles for every outcome–pollutant combination and proceeded with quantitative synthesis only when at least five independent studies were available; in our case, this condition was met only for PM2.5. Combinations with fewer than five studies were excluded from formal pooling to avoid imprecise estimates and over-interpretation of sparse data. To harmonize exposure metrics, values reported in ng·m⁻³ were converted to µg·m⁻³ by dividing by 1000. All effect measures were converted to risk ratios, and natural logarithms of the risk ratio (lnRR) were computed. When extracting outcome data, the direction and scale of each measure, follow‑up time and exposure duration were recorded to enable subgroup analyses. Publication year, region and other study‑level covariates were recorded a priori as potential effect modifiers. Log‑transformed risk ratios and their variances were pooled using a random‑effects model. This approach treats each study as estimating a different but related effect size and views the true effects as a distribution; the pooled estimate is the mean of that distribution. Between‑study variance (τ²) was estimated using the Paule–Mandel or restricted maximum likelihood estimators rather than the DerSimonian–Laird method because simulation work in preclinical meta‑analysis has shown that restricted maximum likelihood and Bayesian estimators produce more reliable estimates of heterogeneity and are preferred when heterogeneity is substantial14. Heterogeneity was quantified using Cochran’s Q, τ² and the I² statistic. The I² statistic can be biased in small meta‑analyses, either overestimating heterogeneity when the true fraction of heterogeneity is small or underestimating it when heterogeneity is large15. Therefore, τ² and 95% prediction intervals were reported to convey the range of true effects expected in future studies. We fitted random-effects meta-regression models to investigate whether variability in study effects could be partly explained by systematic differences in study characteristics, rather than treated as unexplained noise. In the available data, such a variable was publication year. We therefore modelled log-risk ratios as a function of study year, which acts as a proxy for temporal changes in exposure measurement, pollutant composition, regulatory environment and clinical practice, for explaining part of the observed heterogeneity. Random‑effects meta‑regression was fitted using weights equal to the inverse of the sum of the within‑study variance and τ² so that both within‑study variance and residual between‑study heterogeneity were taken into account16. Because meta‑regression associations are observational and can be prone to spurious findings if moderators are selected after inspecting the data, moderators were prespecified and analyses were limited to situations with at least ten studies16. Explanatory variables were evaluated using weighted least squares with the inverse‑variance weights. Pre-specified moderators were publication year, region, exposure-assessment method, co-pollutant adjustment, lifestyle-confounder adjustment and exposure window. Each moderator was eligible if ≥ 10 studies were available for the outcome and ≥ 3 studies were present per categorical level (Borenstein/Cochrane criterion). Univariable meta-regression models were fitted for every eligible moderator; for the all-cause meta-analysis a multivariable model jointly including region, exposure method and co-pollutant adjustment was also fitted, where degrees of freedom permitted. Sensitivity analyses were conducted to examine robustness of the findings. First, we performed a leave-one-study-out analysis by recalculating the pooled estimate after excluding each Article_ID in turn; the minimum and maximum pooled relative risks were recorded to identify influential studies. Second, we compared pooled Relative Risks-RR obtained with the Paule–Mandel estimator to those from the DerSimonian–Laird estimator. Finally, we checked whether the meta-regression conclusions changed when only long-term exposures were retained. Small-study effects and publication bias were investigated for each outcome through the Egger regression test. Results Long term effects The two Italian studies4,10 within the BEEP and BIGEPI projects provided the starting point for this review and offered consistent evidence of increased mortality associated with long-term exposure to air pollution. In a cohort of over 2.7 million adults, Nobile et al.4, reported small but statistically significant increases in non-accidental mortality for PM2.5 (Hazard ratio-HR 1.004, 95% Confidence Interval-CI 1.001–1.007), PM10 (HR 1.018, 95% CI 1.005–1.030) and NO2 (HR 1.010, 95% CI 1.002-1.018), as well as a significant association between PM10 and cardiovascular mortality (HR 1.029, 95% CI 1.013–1.045). Associations with respiratory mortality were generally weak or null. Similarly Gariazzo et al.10 found consistent associations of PM10 and NO₂ with cardiovascular (HR ≈ 1.04) and natural mortality (HR ≈ 1.02), supporting the role of traffic-related and particulate pollution in long-term health effects. Building on these initial findings, the extended literature search identified consistent evidence across Europe regarding the chronic effects of air pollution exposure. Results are summarised below by pollutant and detailed in Table S7. Across multiple European cohorts, PM2.5 emerged as a robust predictor of long-term mortality. In addition to the positive associations observed in Nobile et al.4 with non-accidental mortality, several studies reported increases in natural, cardiovascular and respiratory mortality. In line with reference studies17–19, found significant elevations in natural and cardiovascular mortality in Europe, while Klompmaker et al.17, Al Ahad et al.20, and Stafoggia et al.19 observed increased risks also for respiratory outcomes in the Netherlands (HR 1.058, 95% CI 1.009–1.111), in Scotland (HR 1.062, 95% CI 1.028–1.096), in European cohorts (HR 1.064, 95% CI 1.013–1.118). Italian, Swiss and UK data provided similar findings21–23. Both Italian baseline studies indicated associations between PM10 and long-term mortality, especially for cardiovascular4,10 and non-accidental outcomes4. Additional evidence from Sweden18, Scotland20, and Italy21,24 confirmed increased risks for cardiovascular and natural mortality. Al Ahad et al.20 found notable risk increases for respiratory mortality in Scotland (HR 1.025, 95% CI 1.005–1.045), while Zhang et al.25 documented significant associations with coronary artery disease in the UK. Long-term NO₂ exposure showed strong and consistent associations across studies. Nobile et al.4 and Gariazzo et al.10 found positive associations with cardiovascular and natural mortality. Larger European cohorts confirmed these findings19,22,26,27, typically around 4–6% higher risk per 10 µg/m³ NO2 (Pooled HRs 1.04–1.06). Other European studies also found significant associations between NO2 and respiratory mortality19,26. Overall, NO₂ was one of the most consistently harmful pollutants in the long-term perspective. The effects of long-term O₃ exposure showed more heterogeneity. Nobile et al.4 reported inverse or null associations for all mortality outcomes in Italy, a pattern partly echoed in Stafoggia et al.19 and Vienneau et al.22 in Europe, where warm-season O₃ showed protective associations. However, other studies documented harmful effects in Europe: Sun et al.28 found increased all-cause (pooled RR 1.014, 95% CI 1.009–1.019), cardiovascular (RR 1.019, 95% CI 1.004–1.035) and respiratory mortality (RR 1.025, 95% CI 1.010–1.040) linked to long-term O3 exposure, and Kasdagli et al.27 reported increased respiratory mortality in pooled analyses (RR 1.05, 95% CI 1.02–1.08). Overall, the evidence for O3 is less consistent, with effect estimates varying across regions and depending on seasonal exposure metrics. Short term effects For the short-term effects, the articles by Renzi et al.12 and Gariazzo et al.11, within the framework of BEEP and BIGEPI projects, were the starting point for our literature research. Renzi et al.12 analysed short-term exposure to PM10 reporting a significant increase in non-accidental mortality in the Brindisi industrial area (Italy) (Increased Risk-IR% 6.7%, 95% CI 0.9–12.7). Gariazzo et al.11 evaluated throughout Italy the short-term exposure to PM2.5, observing consistent increases in cardiovascular (IR% 2.32, 95% CI 1.21–3.43), respiratory (IR% 4.55, 95% CI 1.78–7.40) and natural mortality (IR% 2.08, 95% CI 1.44–2.72). Respiratory mortality was also associated with PM10 (IR% 3.54, 95% CI 2.13–4.97) and NO2 (IR% 6.68, 95% CI 1.04–12.62). These findings are in line with other European studies consistently showing that short-term elevations in air pollutant concentrations are followed by acute increases in mortality. Results are summarised below by pollutant and detailed in Table S8. As regards PM2.5, consistent evidence was observed in Central and Eastern Europe. In Poland, Czernych et al.29 documented increased risks of myocardial infarction mortality associated with PM2.5, supporting the role of PM2.5 in triggering acute cardiovascular events. Belgian data from Demoury et al.30 showed modest but statistically significant increases in all-cause mortality, while those related to cause-specific outcomes were generally weaker. Evidence from a multi-country analysis by Yu et al.31, highlighted considerable heterogeneity across Europe, with positive associations observed in countries such as Finland and the United Kingdom, and null or inverse associations reported in Norway, Spain and Sweden. Short-term exposure to PM10 was consistently associated with acute mortality, particularly for non-accidental, cardiovascular and all-cause outcomes. In line with Italian studies11,12 other studies supported the association between PM10 and cardiovascular mortality in France32 in Poland29(for stroke and myocardial infarction mortality, and in Greece for cardiorespiratory and all-cause mortality33. In Belgium,30 significant increases were found for all-cause mortality. Short-term exposure to NO₂ showed consistent associations with acute cardiovascular and cerebrovascular mortality across several European settings. In Switzerland, Saucy et al.34 found increased odds of cardiovascular mortality associated with NO2 exposure (Odds Ratio - OR 1.05, 95% CI 1.01–1.1) while in Poland, Czernych et al.29 documented significant short-term increases in stroke and myocardial infarction mortality associated with NO2, (RR 1.04, 95% CI 1.01–1.06 and RR 1.05, 95% CI 1.02–1.08, respectively). Belgian evidence30 further supported these findings, showing positive associations between NO2 and all-cause mortality. In Italy, Gariazzo et al.11 reported heterogeneous results, with positive associations for respiratory mortality and inverse or null estimates for cardiovascular and natural mortality. Nevertheless, overall, NO2 emerged as one of the pollutants most consistently associated with acute cardiovascular outcomes in the short-term perspective. Finally, the short-term effects of O3 exposure appeared more heterogeneous across studies and regions. In Poland, Czernych et al.29 reported mostly null associations between O₃ and stroke mortality, alongside inverse associations for myocardial infarction mortality. Southern European studies reported mixed results. In Greece33 estimated notable attributable fractions of cardiorespiratory and all-cause mortality were linked to O3; in Belgium,30 reported small and generally non-significant short-term effects of O3 on mortality were found. Overall, evidence for short-term O3 effects was less consistent compared with PM and NO2. Metanalysis and metaregression A total of eighteen studies were included in all-cause mortality meta-analysis. The pooled random-effects estimate indicated that a 10 µg/m³ increase in PM2.5 was associated with approximately a 10% increase in all-cause or natural mortality. Heterogeneity was substantial (Q = 805.8, df = 17, I² = 7.9%, p < 0.001), reflecting variation in study design, geographic setting and exposure assessment. Meta-regression showed that effect estimates tended to rise with publication year (coefficient = 0.0276), and long-term exposures were significantly stronger than short-term exposures (coefficient = 0.1620, p = 0.036). τ² from the Paule–Mandel estimator was 0.0193. These findings support an overall positive association between PM2.5 and all-cause mortality and suggest that longer follow-up periods yield larger risk estimates (Fig. 2). Forest plot of per-study RRs of all-cause mortality per 10 µg/m³ PM2.5 (k = 14). Box sizes proportional to the inverse-variance weight; horizontal whiskers span the 95% CI; the diamond represents the random-effects pooled estimate. Long-term cohort studies are marked (L); short-term time-series/case-crossover (S). The cardiovascular mortality meta-analysis included fourteen cohort and time-series studies. The pooled RR was 1.026 per 10 µg/m³ increase (95% CI 1.014–1.039). Heterogeneity remained considerable (I² = 91.7%). Meta-regression found that neither publication year (β = 0.0097, p = 0.712) nor region (Europe vs. other; β = 0.0584, p = 0.263) explained heterogeneity. This suggests that the relationship between PM2.5 and cardiovascular mortality has been relatively stable over recent years and across regions, despite differing populations and adjustment strategies. Each study contributed a single effect estimate for the RR of cardiovascular mortality associated with a 10 µg/m³ increase in PM2.5. The reported effect sizes varied widely: the most conservative estimate corresponded to a RR of approximately 1.006, indicating a 0.6% increase in risk, whereas the most pronounced estimate was about 1.073, implying a 7.3% increase. Most of the remaining studies clustered around modest increases, with RR in the range of 1.017 to 1.030. For respiratory mortality, thirteen studies were synthesized, encompassing both long-term and short-term designs. The pooled random-effects estimate was 1.032 (95% CI 1.013–1.050) per 10 µg/m³ increase in PM2.5. Heterogeneity was high (I² = 88.3%). Study-level effect estimates ranged from null to substantially elevated risk, with one multimorbidity cohort reporting RR > 1.7 and another cohort reporting risk ratios below unity. Meta-regression identified a declining temporal trend: the intercept corresponded to a RR of 1.04 at the mean publication year, and the slope suggested a 3.61% change per year. This negative trend may reflect declining ambient PM2.5 concentrations, improved exposure assessment and more comprehensive confounder control in recent studies, although publication bias toward smaller effects in recent years cannot be ruled out. In our synthesis, each study contributed exactly one RR estimate for respiratory mortality per 10 µg/m³ increase in PM2.5. The pooled log-risk ratio (~ 0.032) translates to a RR of about 1.032 (i.e., a 3.2% increase in respiratory mortality per 10 µg/m³) with a 95% CI from 1.013 to 1.050. Individual studies, however, varied considerably: the most conservative estimate was close to 1.00, while the most pronounced estimate approached 1.15, implying roughly a 15% increase in mortality. Most of the remaining studies reported modest elevations in risk, clustering around relative risks of 1.02–1.07. Across outcomes, long-term cohort studies generally reported stronger associations than short-term time-series analyses (Fig. S2). The pooled effect for all-cause mortality was larger than those for cardiovascular and respiratory mortality, indicating that PM2.5 exposure is most strongly related to overall mortality risk. The high heterogeneity underscores the importance of methodological differences across studies and reinforces the utility of meta-regression to explore potential moderators. The red lines in Fig. 3 depict fitted meta-regression slopes based on published coefficients: positive slopes for all-cause and cardiovascular mortality (+ 2.8% and + 0.97% per year, respectively) and a negative slope for respiratory mortality (–3.6% per year). Meta-regression bubble plots showing RR per study against publication year for each mortality outcome. Taken together, these results provide robust evidence that ambient PM2.5 is associated with increased mortality from multiple causes, with the magnitude of association varying by outcome and over time (Figs. 2 and 3). Sensitivity analyses showed that the pooled all-cause RR varied from 1.03 to 1.08 across leave-one-study-out analyses; removing the Tang et al.23 study reduced the pooled RR to 1.03, whereas excluding Vienneau et al.22 study increased it to 1.08. These two studies therefore exerted substantial influence, while removal of other studies altered the pooled estimate by less than 0.5%. The cardiovascular and respiratory mortality results exhibited similar patterns of influence, though the ranges were narrower (approximately 1.02–1.04). Overall, the sensitivity analyses demonstrate that the association between PM2.5 and mortality remains positive across a variety of analytic choices, but the magnitude of the pooled effect is sensitive to high-influence studies and to the choice of heterogeneity estimator. The Egger regression indicated small-study effects in all three meta-analyses (p = 0.024 for all-cause, < 0.001 for cardiovascular and 0.003 for respiratory mortality), warranting a cautious interpretation (Table S9). Pre-specified moderators only partially explained between-study heterogeneity. In the all-cause meta-analysis, exposure-assessment method (R²=22%, p = 0.028) and exposure window (R²=19%, p = 0.018) explained the largest single shares of variance. In the cardiovascular meta-analysis, region was the dominant moderator (R²=26%, p = 0.014), with larger pooled RRs in Southern than in Northern Europe. The multivariable model for the all-cause outcome jointly including region, exposure method and co-pollutant adjustment explained 39% of between-study variance. Leave-one-out analyses produced pooled RR ranges of 1.031–1.082 for all-cause, 1.011–1.019 for cardiovascular and 1.019–1.034 for respiratory mortality (Figure S3). Case-deletion influence diagnostics flagged Brunekreef17, Stafoggia19 and Liu35 as influential for the all-cause estimate (Figure S4). Pooled estimates were largely insensitive to the choice of between-study variance estimator (REML vs. Paule-Mandel vs. DerSimonian-Laird, agreement within ± 0.007). Restricting the synthesis to long-term exposures yielded systematically larger pooled RRs (all-cause 1.078, cardiovascular 1.019, respiratory 1.034), consistent with the a priori hypothesis of cumulative chronic-exposure effects, and exclusion of the single overall High-risk study did not change the qualitative conclusions. Considering between-study heterogeneity (Figure S5), small-study effects, and the precision and consistency of pooled estimates, the strength of the evidence was judged moderately strong for all-cause and cardiovascular mortality and limited for respiratory mortality. Discussion This review synthesizes evidence from recent European studies on the effects of major ambient air pollutants on the short- and long-term mortality, confirming a substantial and consistent public health impact of air pollution. Across study designs and populations, PM and NO2 consistently emerged as the pollutants most strongly associated with mortality outcomes. Long-term exposure to PM2.5 was consistently associated with increased all-cause, cardiovascular and respiratory mortality across European cohorts4,17–19,21–23,26. PM10 showed similar associations with long-term mortality risks, particularly for cardiovascular and natural causes4,10,18,21,24,25. Evidence from both national and regional studies suggests that thoracic PM contributes meaningfully to long-term health effects, likely through mechanisms overlapping with those of PM2.5. Short-term exposure to PM2.5 and PM10 was associated with acute increases in non-accidental, cardiovascular and all-cause mortality in European countries11,12,29,30,32,33. Our metanalyses for PM2.5 showed a significant association with mortality across Europe: all-cause mortality RR ≈ 1.10 per 10 µg/m³, cardiovascular mortality RR = 1.026 and respiratory mortality RR = 1.032. Stronger effects were observed for long-term exposures. NO2 showed strong and relatively consistent associations with mortality in both long-term and short-term studies performed in Europe4,10,17,19,22,26,29,30,34. Inconsistent results about the health effect of O3 have been reported: suggestive associations with mortality27,28,33,36, but also null or inverse associations4,19,22,29,30. Sometimes, the uncertainty in interpreting results is caused by the interaction between O3 and PM2.5 levels37. In the summer season, when levels of O3 rise, a positive association with other pollutants has been often observed. The impact of O3 seen in summer could stem from the higher concentrations measured during warmer months, as well as from the longer time people spend outdoors, which leads to larger exposure. Conversely, in the winter months, when the formation of O3 through photochemical reactions is limited, a negative relationship between O3 and primary pollutants released from cars and heating systems is seen. Therefore, developing multiple pollutant models taking into account the interfering effects of other air pollutants is widely advised37,38. Recent evidence has indicated that oxidative stress triggered by airborne contaminants is crucial in understanding the effects of air pollution. The initial interaction of inhaled environmental pollutants occurs at the liquid layer that envelops the respiratory epithelium, and the reactions that follow exposure are driven by oxidative processes within this air-lung interface. These processes may lead to oxidative stress and subsequently boost the production of inflammatory substances by human airway epithelial cells. Oxidative stress refers to a condition where the equilibrium between oxidants and antioxidants is upset. This disruption can happen when the creation of oxidizing molecules, such as free radicals, surpasses the existing antioxidant defenses. The effects of oxidative stress can manifest as either systemic or local inflammation, thus impacting the cardiopulmonary system39. The three most common pollutants responsible for oxidative stress are NO2, PM10, and O3. Most human genetic studies relating to air pollutants have focused on ozone exposure. Ozone acts as a potent oxidizer and interacts with the fluid lining of the bronchial epithelium, leading to free radical formation. It reduces levels of protective antioxidants and boosts the generation of inflammatory mediators39. NO2 may exert direct toxic effects through airway inflammation and oxidative stress, although it also serves as a marker of traffic-related pollution mixtures, including ultrafine particles and other co-emitted pollutants40. The potential for particulates to cause oxidative harm is influenced by their size and surface characteristics. Generally, smaller particulate matter tends to be more toxic, leading to higher oxidative stress and inflammation. Indeed, nanoparticles (ultrafine particles with diameter < 100 nm) are more toxic and can generate reactive oxygen species to a greater extent, exacerbating pre-existing respiratory and cardiovascular disease, also through a dose–response effect39. Consistent with this, fine particles can translocate into the systemic circulation, promoting chronic inflammation, oxidative stress, endothelial dysfunction, and progression of atherosclerosis, thereby contributing to increased cardiovascular and respiratory mortality41,42. Short-term exposures are also biologically plausible as triggers of acute events. These effects are likely mediated by rapid systemic responses, including autonomic imbalance, vascular dysfunction, pro-thrombotic changes, and exacerbation of pre-existing conditions41,43. The consistency of these findings across diverse geographic contexts further supports a likely causal relationship between air pollution exposure and adverse health outcomes. Several caveats temper interpretation. The Egger regression suggested small-study effects in all three meta-analyses (p < 0.05); 16 of 18 meta-analysed studies carried ROBINS-E ratings of “Some concerns”, driven mainly by residual confounding by individual lifestyle factors and exposure-modelling resolution. Residual cohort overlap within the ELAPSE cluster could not be entirely excluded, but pre-specified leave-one-out analyses on the overlapping papers did not move pooled estimates by more than 5%. Among the limitations, we also acknowledge that the review protocol was not prospectively registered (e.g., in PROSPERO). However, the study was conducted following a pre-specified protocol, with all methodological decisions defined prior to the quantitative synthesis and transparently reported in the manuscript. The substantial heterogeneity observed in these meta-analyses arises from real differences across studies, rather than random error. In all-cause analyses, heterogeneity arose from variation in study design (cohort vs. time-series), geographic setting, exposure assessment and population characteristics44. The respiratory mortality synthesis noted that high I² values reflected differences in study design, population characteristics, exposure measurement and adjustment for confounders. Regional differences appeared particularly important: a recent WHO update reported that higher RR were found for lung cancer mortality in Europe and lower in the Americas and Western Pacific regions. This may reflect differences in pollutant composition and sources; for example, larger penetration of diesel engines in European fleets compared with the United States. Heterogeneity also varied with ambient pollutant levels; meta-regression suggested that higher PM concentrations were associated with larger effect estimates. Differences in exposure duration (long- vs. short-term), exposure windows and lag structures can further contribute to heterogeneity. High levels of heterogeneity do not necessarily undermine evidence of harm; Orellano et al.44 argue that a persistently positive association across contexts is an indication of external validity. Potential sources of bias deserve careful consideration. Although funnel plots in the WHO review suggested possible publication bias, Egger’s test may be inappropriate in the presence of high heterogeneity and comparing estimates from large and small studies may provide a better indication. Residual confounding remains a salient concern: confounding factors such as body mass index and smoking are not consistently available in administrative cohorts, leading to moderate risk-of-bias ratings in the confounding domain. Exposure misclassification may bias results toward the null because most studies rely on ambient monitoring or modelling rather than personal exposure. Cross-pollutant confounding may persist even when two-pollutant models are used; effect estimates are often attenuated when adjusting for additional pollutants. Selection bias is possible in cohorts that exclude individuals with missing data or specific health conditions, potentially limiting generalizability. Several strengths bolster confidence in these findings. The analyses synthesize a large, recently updated body of evidence spanning multiple European countries, incorporating both long-term cohort studies and short-term time-series analyses. Using random-effects models allows for between-study variability, and meta-regressions explore potential effect modifiers, including publication year and region. The systematic standardization of exposure increments enables direct comparison across studies and applying a threshold of at least five studies per outcome–pollutant combination helps to avoid sparse-data bias. Importantly, the positive associations persist across diverse contexts, supporting external validity. Sensitivity analyses excluding high-bias studies did not materially change the pooled estimates, indicating robustness. Conclusions Our findings demonstrated that, despite the reduction in air pollution levels in recent decades in Europe45, this exposure remains an avoidable health risk that represents a serious burden to society in terms of general, cardiovascular and respiratory mortality. EU air quality standards are currently laxer than the WHO AQG published in 2021, and studies have shown significant health effects even below the most stringent current standard. Therefore, our findings add new scientific evidence supporting the need to further reducing population exposure in order to achieve global health benefits. Significant efforts are needed to prevent the onset and the exacerbation of chronic diseases, not only by reconsidering current air quality legislation and regulations. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Author contributions IS and AP: Data acquisition, Analysis, Investigation, Methodology, writing original draft & editing. SB, SM and GS: Supervision, Methodology, Writing review & editing. ST : Methodology, Writing review & editing. VM and MP : Supervision, Project administration, Funding, Methodology, Writing review & editing. PA : Methodology, Writing review & editing. FS : Data acquisition & Investigation. GV and SLG : Supervision, Methodology, Writing review & editing. Funding Project implemented with the technical and financial support of the Ministry of Health – PNC PREV- A-2022-12376981. Data availability The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Ilaria Stanisci and Andrea Perri. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.