This study provides, to our knowledge, the first municipality-level assessment of life expectancy differences attributable to long-term air pollution exposure in Slovakia, integrating exposure response functions with empirical mortality and socioeconomic data. Across multiple approaches, air pollution emerges as one of the strongest determinants of spatial differences in life expectancy, with losses approaching 1.5–2 years in the most polluted municipalities, comparable in magnitude to educational disparities and exceeding effects associated with healthcare accessibility. The health burden is unevenly distributed across the exposure gradient. Life expectancy losses rise steeply with pollution levels and are concentrated in municipalities with high use of residential solid-fuel combustion and poor dispersion10,22. Differences between modelling approaches are largest at lower exposures but converge in highly polluted areas, where all methods consistently indicate substantial life expectancy deficits, underscoring the robustness of these findings in the most affected communities. By combining regression-based estimates with WHO and SCHIF frameworks, the analysis highlights the importance of accounting for historical exposure trajectories and lagged mortality responses in regions undergoing rapid emission transitions. The results indicate that present day air quality alone does not fully capture long-term health consequences in Central and Eastern Europe, where major pollution reductions occurred only after the 1990s.
The statistical regression model developed in this study incorporates a broader set of determinants than air pollution alone and therefore captures a more complex structure of life expectancy (LE) variation across Slovak municipalities. Whereas WHO and SCHIF rely solely on exposure contrasts and assume equilibrium mortality responses, the regression integrates empirical health outcomes from 2008 to 2024 together with socioeconomic, demographic, and environmental covariates implicitly embedded in the long-term mortality data. This allows the model to reflect both historical exposure patterns and the lagged recovery of population health following the large reductions in pollution that occurred after the 1990s in Slovakia46,47. This finding indicates that, even some municipalities currently located in the lowest pollution quantiles often exhibit higher ΔLE in the regression model than predicted by SCHIF or WHO, because their present-day mortality still carries the imprint of past, substantially higher exposure levels19. In this sense, the regression model captures a time integrated, lag sensitive response rather than the steady state contrasts assumed in standard exposure response functions. These differences are most visible in the lower half of the exposure distribution (Q1-Q10), where SCHIF predicts very small impacts and WHO produces nearly constant relative changes, while the regression values remain appreciably higher. This suggests that part of the observed LE gradient in cleaner areas is influenced not only by current ambient air quality but also by residual effects of historical pollution, as well as by local determinants not represented in exposure only frameworks, as discussed in ref. 40. Indoor air quality, heating technologies, housing insulation, ventilation patterns, the prevalence of solid fuel use in colder regions, socioeconomic disparities, healthcare accessibility, and demographic structure all modulate underlying health risks16. These factors are indirectly absorbed into the regression coefficients but cannot be disentangled explicitly within WHO or SCHIF approaches.
When the four approaches were synthesized into a combined estimate (inverse-variance weighted fusion), the resulting curve closely followed the regression and SCHIF trajectories from approximately Q15 upward, indicating strong agreement in the high exposure domain where the mortality burden is primarily driven by chronic inhalation of combustion derived particles20. The Bayesian hierarchical fusion, although more uncertain at low exposures, converged toward the same pattern in higher quantiles. This reinforces the conclusion that despite methodological differences, all approaches identify a consistent and robust increase in ΔLE at the upper end of the pollution distribution (approximately from 22 μg/m3). In contrast, the divergence at low exposures highlights the limitations of steady state models and the importance of incorporating temporal dynamics and contextual factors when interpreting health impacts in historically transitioning regions48,49. A methodological caveat is that the four input methods are not statistically independent, they rely on overlapping exposure fields and partially shared epidemiological evidence34. The inverse-variance weighted fusion therefore likely underestimates the true combined uncertainty, because between method correlations are not explicitly modelled.
Overall, the regression model complements WHO and SCHIF by providing a data driven perspective that integrates real world heterogeneity, historical exposure trajectories, and lagged mortality responses. At the same time, its behavior in low exposure areas illustrates that air pollution is only one component of a broader constellation of environmental and social determinants shaping regional differences in life expectancy. The regression model should not be interpreted as providing individual-level causal estimates. Higher pollution levels, particularly in areas dominated by solid-fuel combustion, are likely to co-occur with broader socioeconomic disadvantage that may not be fully captured by the included covariates, and residual confounding cannot be excluded. However, the exposure-response relationship is consistently monotonic across the exposure distribution and remains robust after adjustment for key structural determinants, including educational attainment and the spatial distribution of marginalized Roma communities. These variables capture major axes of social inequality in Slovakia and explain a substantial share of variation in life expectancy. Within this adjusted framework, the persistence and coherence of the air pollution gradient support its interpretation as a meaningful contributor to population health differences, while acknowledging the limitations inherent to ecological analysis. The remaining covariates included in the regression model showed associations with life expectancy that were consistent in direction and magnitude with previously reported evidence. Unmeasured or imperfectly measured confounders, such as dietary and physical activity patterns, smoking prevalence, alcohol consumption, occupational exposures, and detailed healthcare quality may still bias the estimated associations, particularly in the lower exposure quantiles where the air pollution signal is weaker19. Collinearity between socioeconomic indicators and pollution metrics further complicate causal attribution and suggests that our regression results should be viewed as integrated “signatures” of place based disadvantage rather than isolated pollution effects50. The IVW synthesis provides a pragmatic way to combine estimates derived from different frameworks. While these approaches share common exposure fields and life tables, they rely on different exposure response functions and therefore represent alternative interpretations of the health impacts of PM2.5 exposure. By incorporating the uncertainty ranges of the individual estimates, IVW offers a transparent method for deriving a consolidated estimate suitable for policy-oriented applications. This is particularly important in applications where a single estimate is required for spatially resolved analyses, such as economic assessments of air pollution impacts at the municipal level.
The WHO HRAPIE function assumes a log-linear exposure response relationship, resulting in a constant relative risk per unit of concentration across the entire range of pollution levels. While epidemiologically conservative, this formulation produces artefacts in settings with substantial spatial heterogeneity4. In our analysis, the WHO model generated comparatively high ΔLE values even in the lowest exposure quantiles, implying unrealistically large benefits in municipalities that are already relatively clean. Because the log-linear function scales proportionally, backward shifts in historical exposure preserve relative differences between areas. Therefore, the Q20-Q1 difference remains nearly unchanged regardless of the baseline level in the WHO model. This behavior contrasts with empirical evidence showing diminishing marginal risks at low concentrations and tends to flatten the true spatial gradient of life expectancy differences40,51. Consequently, the WHO approach is more suitable for national level burden of disease assessments than for fine scale or historical reconstructions such as those performed here19. The SCHIF approach, in contrast, applies a shape constrained nonlinear function that increases slowly at low exposures and steeply in the upper range. This yields very small ΔLE estimates in the cleanest areas but a rapid rise in health impact among municipalities with higher pollution, patterns that closely match our regression-based results, particularly above approximately the 15th exposure quantile. SCHIF therefore captures the nonlinear amplification of chronic mortality risk in polluted environments more realistically than the WHO function, which is confirmed also in many studies19,33,52,53. Its main limitation lies in the sensitivity to the assumed inflection region; small shifts in the curve’s low exposure domain can alter national ΔLE estimates, particularly in countries like Slovakia, where a sizeable share of the population resides near the lower middle portion of the curve. SCHIF may also understate long-term effects of persistent low-level exposure, which continues to generate cardiovascular and inflammatory burden even below the apparent threshold region34,52.
Together, the two methodologies thus define a useful uncertainty envelope. WHO provides a relatively high estimate in low exposure settings but with a compressed Q20-Q1 gradient of about half a year, whereas SCHIF yields smaller ΔLE at the clean end but a much steeper rise toward high exposures that aligns better with the regression results. The convergence of SCHIF and the regression curves in upper quantiles, combined with the comparatively flat WHO profile, indicates that nonlinear functions are more appropriate for capturing the steep life expectancy deficit in the most polluted municipalities40. A further limitation common to both WHO and SCHIF is that their exposure response functions are derived predominantly from Western European and North American cohorts, where the particle mixture is more strongly dominated by traffic and secondary aerosols than by residential solid-fuel combustion51. Extrapolating these relationships to Slovak conditions, therefore, assumes that equal PM2.5 increments have comparable toxicity across different source mixtures, an assumption that may not fully hold in regions with a high contribution of coal and biomass burning. Moreover, both methods operate on area level average concentrations and do not account for within municipality exposure gradients or indoor-outdoor differences, which are likely to be substantial in settings with heterogeneous heating technologies and building characteristics20,21.
Across different settings, reductions in PM2.5 have repeatedly been linked with measurable gains in life expectancy, typically on the order of 0.04–0.08 years per 1 µg/m³ decrease in long-term exposure, as shown for example in US cohorts and global burden assessments4,5,54. In our Slovak municipal level analysis, both the WHO and SCHIF applications fall into this plausible range, a mean ΔLE of about 1.36 years (WHO, 1990 baseline) and 0.70 years (SCHIF, 1990 baseline) for exposure reductions roughly in the order of 10 to 20 µg/m³ implies effect sizes very close to 0.04–0.08 years per 1 µg/m³. When the regression model is transformed into an approximate linear slope, it yields an LE gain of about 0.08 years per 1 µg/m³. That is at the upper end of the range reported by published studies, which is consistent with the historically high pollution levels experienced in many Slovak municipalities19. At the same time, our comparison of PM2.5 and B(a)P based regressions confirms that bulk particle mass is a robust but could be somewhat conservative proxy for health impacts. Both PM2.5 and B(a)P models predict similar losses of roughly 1.8 to 2.0 years in the highest exposure quantiles, with overlapping confidence intervals, yet the B(a)P curve is slightly steeper and more nonlinear, especially in mid exposure quantiles, in high quantile both B(a)P and PM2.5 results are very similar, indicating a common source of pollution55. This pattern aligns with evidence that PAH enriched particles retain independent associations with mortality and cardiopulmonary outcomes even after adjustment for PM2.5, suggesting that using PM2.5 alone may underestimate the true benefit of interventions targeting solid fuels combustion sources in some areas4. In our context, the strong correlation between B(a)P and PM2.5 implies that most of the risk is captured by PM2.5 fields, but the B(a)P results indicate that part of the remaining uncertainty, and likely a fraction of the residual LE gap is linked to this more toxic PAH fraction.
From a measurement perspective, B(a)P poses additional challenges. Monitoring networks are sparser, time series are shorter, and analytical detection limits are higher and more variable than for PM2.57. As a result, our B(a)P based regression is fitted on a more restricted and potentially noisier subset of the exposure space, which may contribute to the wider confidence intervals and the more nonlinear pattern at high quantiles. Furthermore, neither the PM2.5 nor B(a)P based models explicitly adjust for pollutants such as NO2 or O3, nor for particle composition beyond a single PAH tracer56. In practice, any LE effect attributed to PM2.5 or B(a)P likely reflects a broader mixture of combustion-related pollutants, and the partitioning of risk between these components remains uncertain. Finally, the magnitude of life expectancy differences attributable to air pollution in our study, ~0.5–2 years across municipalities is comparable to major behavioral risk factors such as unhealthy diet, physical inactivity, smoking or harmful alcohol use, and clearly exceeds the impact of any other single environmental exposure57. Targeted air quality policies, particularly those addressing residential combustion, should therefore be a central pillar of public health strategy, alongside measures targeting lifestyle, behavioral and socioeconomic risk factors. Air pollution represents a complex mixture of pollutants, and the focus on PM2.5 and B(a)P does not capture the full spectrum of potential exposures. However, in the Slovak context, these pollutants are closely linked to residential combustion and represent the dominant exposure pathways. Other pollutants, such as NOx, may play an important role at the local scale, particularly in traffic influenced by urban environments. However, given the resolution and structure of national scale air quality modelling, these localized effects cannot be consistently captured at the municipality-level across the entire country.
Our reconstruction of historical exposure relied on a linear backward extrapolation of monitored concentrations to 1990. It is notable that even though the overall trend decreases, some stations report long-term increases of PM concentrations in summer, which can be attributed to cyclical climate change and resulting dry periods creating dry and dusty periods28,58. The assumption of linear extrapolation is methodologically common25,59,60, but historically imperfect, as the emission trajectory of Slovakia (in supplementary material, chapter S4), and of most Central and Eastern European countries was strongly nonlinear around the political and economic transition after fall of socialism in 198961. During the early 1990s, several major industrial sources rapidly declined or shut down, pollutant abatement technologies were progressively introduced, and residential heating shifted from coal to natural gas in large parts of the country11. These structural changes produced a rapid decrease in primary particulate and PAH emissions that far exceeded the gradual decline observed after the year 200062. It is therefore plausible that ambient concentrations of PM2.5 and B(a)P declined more steeply between 1990 and 2000 than assumed in the linear extrapolation based solely on recent monitoring data.
The reconstruction of historical exposure trajectories represents an important source of uncertainty. In the absence of spatially resolved monitoring data prior to 2010, the study relies on a linear extrapolation of observed trends. Although the exact year-to-year evolution of concentrations may have differed locally, several lines of evidence support the plausibility of this approximation. First, independent exposure reconstructions suggest that PM2.5 concentrations in Slovakia during the early 1990s were typically below 30 µg/m³ (28,29), consistent with the baseline reconstruction used in this study. Second, sensitivity analyses indicate that alternative nonlinear scenarios frequently produce unrealistically high concentrations exceeding 35 µg/m³ on average. Finally, monitoring data show no statistically significant regional differences in long-term decline rates, suggesting that applying a single national trend represents a reasonable simplification of historical dynamics. The historical exposure reconstruction should be interpreted as an approximation of long-term exposure trends rather than an exact reconstruction of annual concentrations in individual past years, including year-specific meteorological effects.
The dynamics of historical air pollution change could also be spatially heterogeneous. Industrial restructuring and heating modernization progressed more rapidly in urban centers than in rural, solid fuel dependent regions61,63. As a result, historical exposure patterns likely differed slightly from the present ones64. Cities that today fall into the lower pollution quantiles may have experienced some of the highest exposures around 1990, whereas rural municipalities, now among the most polluted may have improved more slowly50,65. This temporal inversion could help explain why, in our regression model, lower pollution quantiles (dominated by today’s urban or near urban areas) sometimes show greater ΔLE than expected from SCHIF. Their long-term cumulative exposure could be underestimated by the linear back projection. This spatial contrast is reflected in models in current patterns, with higher PM2.5 and B(a)P concentrations persisting in south-eastern regions dominated by residential solid fuel heating and weaker atmospheric dispersion, while lower levels prevail in southern lowlands with more effective ventilation. The close covariation of PM2.5 and B(a)P supports a shared origin in residential solid-fuel combustion, consistent with European receptor studies attributing the majority of B(a)P emissions to this source66.
Collectively, these considerations highlight that health impact assessments based solely on present day exposure distributions omit an essential component of long-term cumulative risk19. Incorporating the nonlinear emission transitions of the post communist period is therefore crucial for accurate life expectancy attribution in Central and Eastern Europe. Linear extrapolation likely underestimates both true historical concentrations and the corresponding ΔLE gradient, particularly in the nonlinear region of the SCHIF curve, where modest changes in concentration produce disproportionately large mortality effects4. In addition, our exposure reconstruction treats population as spatially static, without explicitly modelling migration between high and low pollution municipalities over the life course. This may blur true individual exposure histories and contribute to uncertainty in the attribution of post 1990 LE gains to air-quality improvements. COVID 19 lockdowns affected short term pollution levels but had only a limited influence on annual mean PM2.5.
Several notable limitations need to be addressed. Residential solid-fuel combustion is among the most uncertain emission sectors in Central and Eastern Europe, as national inventories are based on incomplete information on household heating technologies, fuel quality, and user behavior. Local practices, including waste burning, are rarely documented and may contribute to remaining spatial inconsistencies, particularly in rural areas. Reconstruction of historical exposure is further limited by the scarcity of monitoring data before the late 1990s. As a result, the linear back extrapolation of PM2.5 and B(a)P concentrations could be biased by several µg/m³, especially during the period of rapid economic and political change after 1989, however historical studies provide evidence that air pollution used in this study is well aligned. While sensitivity analyses indicate that our main conclusions are robust, more detailed historical emission inventories and chemically consistent analyses would substantially improve future assessments. Both the WHO HRAPIE and SCHIF exposure response functions are based largely on evidence from Western Europe and North America. Applying these relationships to Slovak conditions assumes broadly comparable particle toxicity, baseline health risks, and population susceptibility that are problematic. Differences in pollution mixtures, demographic structure, and healthcare access may influence the true shape and slope of the exposure response relationship. The regression analysis relies on administrative mortality data, which introduces additional uncertainty in small municipalities with limited populations. Individual municipal estimates should therefore be interpreted with caution, with greater confidence placed on aggregated patterns and gradients. Exposure misclassification is unavoidable. Because PM2.5 concentrations were used as one of the predictor variables in the B(a)P interpolation framework, part of the observed PM2.5 and B(a)P spatial correlation may reflect shared model structure in addition to genuine emission patterns. This approach may reduce sensitivity to municipalities with atypical pollutant composition, where the B(a)P fraction is relatively high despite only moderate PM2.5 concentrations, or vice versa. However, independent monitoring data still show strong co-variation between measured annual mean PM2.5 and B(a)P concentrations, supporting that much of the observed spatial overlap reflects real emission patterns dominated by residential combustion. Also, air quality models smooth fine scale spatial variability, and available monitoring data cannot capture microenvironmental or indoor-outdoor exposure differences. Indoor air quality may also contribute to total exposure, particularly in households using solid fuels for heating. While cooking-related emissions from open combustion sources are relatively limited due to widespread gasification, indoor exposure may still vary with socioeconomic conditions and heating practices. Due to the lack of high-resolution data on indoor air quality and household energy use, this aspect could not be explicitly quantified and represents a limitation of the study. In addition, the use of single pollutant models limits attribution within complex combustion-related mixtures, while concurrent changes in healthcare, demographics, and behavior cannot be fully separated from air quality effects. Nevertheless, the consistency of results across mechanistic, statistical, and combined modelling approaches, particularly at higher exposure levels, supports the robustness of the main findings.
This study estimates municipality-level life-expectancy loss associated with long-term PM2.5 and B(a)P exposure in Slovakia, integrating air quality models, long-term monitoring, life table analysis, multivariable regression, and explicit fusion of multiple health impact frameworks. Across all approaches, life expectancy loss increases monotonically with exposure. Depending on the method, the mean national ΔLE associated with current pollution ranges from 0.72 years (SCHIF, 1990 baseline) to 1.20 years (WHO), while regression estimates imply ~1.17 years nationally and an average slope of 0.08 years per 1 µg/m³, consistent with international evidence. At higher exposure levels (~Q15–Q20), except WHO, all methods converge, indicating a robust life expectancy deficit of about 1.8–2.2 years in the most polluted municipalities. Combining WHO, SCHIF, and regression based estimates through inverse-variance weighted fusion yields a smooth, coherent dose-response curve with relatively narrow uncertainty, representing a stable summary of health burden at high exposures. In cleaner municipalities, results are more divergent, WHO predicts comparatively large losses, SCHIF very small ones, and the regression model suggests modest but non negligible deficits, likely reflecting historical exposure and contextual factors. PM2.5 and B(a)P based regression results converge in the highest exposure quantiles, indicating a common pollution source, while modest differences at intermediate exposures suggest that PAH-related toxicity may account for part of the residual life-expectancy gradient. Overall, ambient air pollution explains an important, but partial, share of regional life expectancy differences in Slovakia. Effective health-gain strategies should therefore prioritize reductions in residential PM2.5 and PAH emissions while being integrated with broader social and healthcare policies. The fused exposure response curve provides a practical tool for translating heterogeneous evidence into policy relevant estimates of life expectancy gains across the exposure gradient. The results presented in this study should be interpreted as associational rather than causal, reflecting spatial relationships between long-term exposure and life expectancy across municipalities. Future research should aim to strengthen causal inference by integrating higher-resolution exposure data, individual-level health information, and quasi-experimental study designs.