Non-Linear evidence from generalised additive models across national environmental health

Highlights

• Gas emissions (with a p-value < 0.05) were consistently associated with economic growth (80%) but were also linked to higher overall mortality.
• Solid fuel emissions showed the strongest negative health effects, particularly on infant mortality (a 45% increase in developing countries) and cardiovascular diseases (a 57% increase in prevalence).
• Liquid fuel emissions had more mixed effects, contributing to GDP per capita at 63% but with less consistent health impacts.
• Cross-country classification revealed that income level alone does not explain performance; institutional quality and healthcare systems are critical determinants.
• Results confirm that economic growth alone is insufficient for improving health and environmental outcomes, which points to the importance of proactive, context-specific policies.

1 Introduction

It is commonly accepted that greenhouse gas (GHG) emissions, especially those from energy use, are a primary cause of climate change, with significant ramifications for both economic stability and public health. Although everyone is concerned about the increase in global emissions, the repercussions are not equally felt in all nations. The principal causes of climate change are these emissions, primarily carbon dioxide (CO₂), methane (CH₄), and nitrous oxide (N₂O). Through the greenhouse effect, a natural phenomenon exacerbated by human activity, especially the burning of fossil fuels, these gases trap heat in the atmosphere. Sea levels have risen, ice has melted, temperatures have increased, and extreme weather events have become more frequent as a result (1). Energy consumption patterns differ significantly across world regions, contributing to persistent global inequalities in both emissions and climate vulnerability (2, 3). Carbon intensity, the amount of CO₂ emitted per unit of energy consumed, varies widely between countries, depending on the energy mix. Earlier research used a panel dataset to investigate the effects of various impact factors, such as economic growth, energy intensity, energy consumption, urbanization, passenger vehicles, and transport infrastructure investments on carbon emissions in one of Africa's leading economies, with results showing that GDP and passenger vehicles contributed significantly to CO₂-eq emissions (3). The carbon intensity of fuels varies greatly; coal is the most carbon-intensive fuel and accounts for the largest portion of global CO₂ emissions, despite not being the most extensively utilised fuel worldwide. Although natural gas emits significantly less CO₂ per unit of energy than coal, both oil and natural gas are significant contributors to emissions. Nevertheless, methane, a powerful GHG, is still released during its extraction and transportation (4). Nuclear and renewable energy sources, on the other hand, emit very little direct pollution. Developed countries tend to have higher per capita energy consumption but often benefit from cleaner energy technologies, while developing countries may rely more heavily on carbon-intensive fuels due to economic constraints (5). Different energy-related emissions also have divergent effects on key indicators. For example, solid and liquid emissions are linked to negative health outcomes, whereas gaseous emissions have been associated with positive effects on economic growth (GDP), highlighting the complex trade-offs between energy, the environment, and development (6).

The Environmental Kuznets Curve (EKC) hypothesis posits that environmental degradation initially worsens with economic growth but begins to decline after a certain income threshold, as societies invest in cleaner technologies, strengthen environmental regulations, and shift toward more sustainable practices. This draws an inverted U-shaped relationship between environmental degradation and economic growth (7, 8). For instance, urbanisation contributes to rising emissions through increased energy use and industrial concentration, although demographic and social development within urban areas may help mitigate emissions through efficiency gains and better infrastructure (9). However, the improvement in environmental quality is not an automatic process. The most likely explanation is that greater prosperity leads citizens to demand stronger environmental protection, which results in stricter standards and better enforcement of environmental laws in richer countries (7). The concept of “environmental dumping” may contribute to accentuating the trend. According to this “pollution haven” hypothesis, as countries grow wealthier, they may stop producing pollution-intensive goods and instead import them from less developed nations with weaker environmental regulations, effectively relocating polluting industries (7). Moreover, developed countries may reduce their domestic pollution by outsourcing polluting industries, but if their demand for pollution-intensive goods persists, this merely shifts pollution abroad (10). The observed relationships are not inevitable and depend on the technological, political, and economic conditions of the time. Current low-income countries can learn from the past and adopt environmental protection measures at earlier stages of their development, thanks to increased awareness of the dangers and the availability of cleaner technologies (7). In addition, the “pollution haven” hypothesis is nuanced: in a finite world, today's low-income countries will not always be able to outsource their polluting industries to even poorer nations (10).

The EKC appears more applicable to local and short-term pollutants, where abatement costs are low and effects are immediately noticeable (e.g., sulphur dioxide, particulates). In contrast, for pollutants with long-term effects and high abatement costs, such as CO₂ emissions and greenhouse gases, some studies do not observe an inverted U-shape but rather a continuously positive relationship with income (10, 11). Nevertheless, other studies highlight specific cases where an inverted U-pattern does emerge, such as in Spain, driven by changes in the energy mix (12), or in the United Kingdom, where stricter legislation played a decisive role (13). Carbon emissions have historically been linked to economic growth and poverty reduction due to the central role of fossil fuels in powering industrialization and development (14). Developed countries have historically emitted more CO₂ due to early industrialization and prolonged reliance on fossil fuels. GDP growth is often made possible by increasing electricity consumption, which in many countries remains heavily reliant on fossil fuels (15). While fossil fuel consumption can facilitate poverty reduction, especially in developing nations by providing affordable energy and employment in energy-intensive industries, it also accelerates CO₂ emissions and environmental degradation (16). These findings encourage greater domestic and foreign investment to boost economic growth, but such growth can also increase a country's exposure to risk and hinder long-term sustainability (17). For example, in countries like Cameroon, this dual effect presents a dilemma, as its dependence on fossil fuels risks becoming a barrier to sustainable development if emissions are not brought under control (18).

Beyond the economy, CO₂ emissions and related pollutants like sulphur dioxide (SO₂) and ozone (O₃) directly impact agriculture by damaging crops and reducing yields, threatening food security and productivity (19). Moreover, air pollution stemming from carbon emissions has direct consequences for public health, increasing hospital admissions for chronic respiratory diseases (20), inflating healthcare costs, and reducing labour productivity through sick leave, all of which indirectly reduce GDP (21). Poverty and income inequality, on the other hand, have been identified as drivers of environmental degradation (22). Other variables, such as population size, labour force, per capita income, and reliance on non-renewable energy sources also correlate positively with CO₂ emissions (23). Crucially, the cost of inaction on climate change is projected to outweigh the investment required for transitioning to renewable energy and adopting robust climate policies (19). Climate change could undermine growth and stability due to its economic toll from extreme weather, infrastructure damage, loss of biodiversity, and increased healthcare needs. Carbon emissions also have a profound and multifaceted impact on human health, both directly and indirectly. As a major driver of environmental degradation and climate change, CO₂ is closely associated with increased mortality rates (6, 24). Numerous studies have demonstrated a strong positive correlation between carbon emissions and mortality, a trend observed across multiple regions, including sub-Saharan Africa (24, 25), Asia and the Pacific (26), and India (15). Environmental pollution resulting from these emissions contributes to climate change and has a wide range of harmful effects on human health. Documented consequences include perinatal complications, infant mortality, respiratory and cardiovascular diseases, allergies, cancers, increased oxidative stress, endothelial dysfunction, and various mental health disorders (27). Children are especially vulnerable, with exposure linked to low birth weight, developmental delays, and an increased incidence of respiratory infections (21). Among the most immediate and direct health effects are those affecting the respiratory system. Inhalation of air pollutants such as particulate matter, ozone, and nitrogen oxides can cause or aggravate both acute and chronic respiratory illnesses, including asthma, bronchitis, and lung cancer (2). In cities like São Paulo, where air quality regulations are either weak or poorly enforced, a clear positive relationship has been identified between the concentration of air pollutants and hospital admissions for respiratory conditions (20). In many developing countries, cooking and heating with solid fuels like wood, charcoal, or coal is a major cause of indoor air pollution. These behaviours are closely linked to higher rates of respiratory diseases, especially in women and children who spend more time indoors and are therefore more likely to get sick (21). Beyond respiratory effects, long-term exposure to air pollution often intensified by heatwaves linked to climate change also significantly affects cardiovascular health. Research indicates that polluted air elevates the risk of heart disease, hypertension, stroke, and cardiovascular mortality; oxidative stress and systemic inflammation, which lead to endothelial dysfunction and compromise vascular health and circulation, primarily underpin these effects (21).

Overall, a substantial body of research shows that carbon-intensive economic expansion is associated with environmental deterioration and negative health effects (6, 21). However, the links between emissions, economic growth, and mortality remain complex and vary considerably across countries. In some cases, rising income levels coincide with declining pollution, while in others, economies remain dependent on high-carbon development models. Furthermore, the health impacts of emissions especially in relation to economic growth are often analysed in isolation rather than within a unified analytical framework. This segmented approach constrains a comprehensive understanding of how environmental, economic, and health processes interact over time. Empirical studies frequently employ linear modelling techniques or concentrate on pollutants, regions, or income categories. Cross-national analyses often neglect nonlinear dynamics and do not consistently differentiate between developed and developing economies. Additionally, not many studies categorise countries based on their overall patterns in emissions, health, and economic performance, which makes it hard to find common features or create useful policy categories. Consequently, it remains unclear whether economic growth alone can simultaneously foster environmental improvement and better health outcomes. To bridge these gaps, this study compiles a broad and integrated dataset and applies Generalized Additive Models (GAMs) to account for potential nonlinear relationships among emissions, economic growth, and mortality. It begins with an in-depth case study of the United Kingdom and subsequently expands the analysis to a global comparison of developed and developing countries. By categorising nations based on their shared trends in environmental, health, and economic indicators, the research offers a more cohesive evaluation of development pathways. The results show that while growth can help improve conditions in certain situations, significant differences between countries, especially in how they use energy, emphasize the need for specific policies and changes beyond just focusing on growth.

2 Materials and methods

This research investigates the relationship between carbon emissions and health indicators using statistical methods. GAMs were used to examine how different sources of carbon emissions relate to health and economic factors in 40 countries with varying income levels over time. All analyses were performed in R using the mgcv package.

2.1 Environmental kuznets curve

As stated earlier in the Introduction section of this research, the EKC theoretically has an inverted U shape indicating that it initially increases, reaches a peak (turning point), and begins to decline as other forms of energy are explored that are less harmful to the environment; this is depicted in Figure 1.

The Environmental Kuznets Curve theory is demonstrated by the overall data in Figure 2: in high-income countries, emissions are stabilising or even declining, largely due to energy efficiency gains, clean energy deployments, and stronger environmental regulations. In contrast, emissions in low- and middle-income countries are still rising because economic growth is prioritised over environmental sustainability, given limited resources and the fact that these countries have not yet reached inflection points. The EKC has been tested with data from the transportation sector, and the theory still holds, revealing that attention should be given to alternative systems of transportation rather than conventional methods to limit the effects of these emissions (3, 8). While developing countries currently have lower per capita emissions, their total emissions are rising rapidly as industrialization progresses. Although Asia has now become the region with the highest overall emissions, surpassing North America and Europe, per capita emissions remain highest in developed nations, as shown in Figure 3.

2.2 Dataset construction

The analysis was conducted on a sample of 40 countries, classified as either developed or developing according to the World Bank. Figure 4 provides an overview of the selected countries along with their corresponding World Bank income classifications. As illustrated, the countries are spread across the globe, selected to provide both economic and geographic representativeness. Clear disparities in development are apparent across continents. However, African countries are less represented than those from other regions, primarily due to limited data availability or the absence of certain fossil fuel usage, most commonly natural gas or, in some cases, coal. All data were obtained from publicly available sources, and no primary data collection was performed. The data were cleaned and organized in an Excel file, with each sheet corresponding to a specific country. Table 1 summarizes the sources and characteristics of all indicators used in the analysis.

Table 1. Overview of health, environmental, and economic indicators used in the analysis.

2.3 Generalized additive models

This work primarily relies on the use of Generalized Additive Models (GAMs) as the core statistical methodology. Unlike traditional linear or polynomial models, GAMs are a flexible extension of Generalized Linear Models (GLM) that allow for the modelling of non-linear relationships between explanatory variables and the dependent variable by estimating smooth functions for each predictor, thereby capturing non-linear effects without imposing a rigid functional form (28). This flexibility is particularly relevant in the context of this study, where the relationships between carbon emissions (from gaseous, liquid, and solid sources) and various indicators, such as death rates or Gross Domestic Product (GDP), may be complex and not strictly linear (29).

The basic structure of a GAM is represented as a function (28):where, The term “Additive” means that each Xj affects Y through its own smooth function, and these effects are added together. This approach makes it possible to identify general trends without imposing a predefined functional form, consistent with the EKC hypothesis. The additive structure can detect different effects across country groups or emission types by analysing the effect of each variable separately.

g(E[Y])=β0+f1(X1)+f2(X2)+…+fp(Xp)

• Y is the response variable (like GDP or death rate).
• (E[Y]) is the expected value of the response.
• g(.) a link function (like identity, log, or logit, depending on the type of Y).
• β0 is the intercept.
• X1,X2,X3,…,Xp are predictor variables
• fj(Xj) are smooth, non-parametric functions of the predictors Xj, estimated from the data (here emissions of CO₂ from gas, liquid, solid fuels).

Figure 5 illustrates how Generalized Additive Models represent the effect of each explanatory variable on the dependent variable. Three smooth curves represent the gaseous, liquid, and solid CO₂ emissions in this case. Each curve shows how one type of emission affects the outcome (in this case, the death rate) while keeping the other variables the same. Table 2 shows the GAM results for the Death Rate indicator, the R² of approximately 0.94 implies that 94% of the variability in DR is explained by the predictors and the deviance explained indicates that 95.6% of the total deviance is accounted for by the model suggesting its adequacy. The GAM structure is:

g(Deathrate)=β0+sgas(Gasemissions)+sliquid(Liquidemissions)+ssolid(Solidemissions)

Table 2. GAM results for the DR indicator for the United Kingdom.

On the horizontal axis, the values of the explanatory variable are displayed, and the vertical axis shows the smooth function s(x,k), which corresponds to the partial effect of that variable on the outcome. The curve is centred around zero: values above zero indicate that the variable contributes positively to the response (compared to the average), while values below zero indicate a negative contribution. The smooth shape of the curve is non-parametric, meaning it is directly fit to the data rather than imposed by a specific functional form. The dashed lines around the curve represent the confidence interval, illustrating the uncertainty of the estimate. The notation s(x,k) specifies both the smoothed function and the number of basic functions k used to construct the spline. This parameter controls the flexibility of the curve but is not itself a coefficient to interpret. In other words, these plots do not display the raw value of the dependent variable, but rather how each predictor influences it once the effects of all other variables in the model are considered. Interpreting these graphs allows us to visualise whether the effect is positive, negative, or non-linear, and at which ranges of emissions these effects are strongest. In the GAM, β0 is constant for each indicator and does not affect the analysis of variable-specific effects. Therefore, it will not be discussed in the results. The modelling was performed using the gam() function from the mgcv package in R. Smooth terms were used to capture the non-linear patterns, and model performance was assessed using goodness-of-fit indicators such as explained deviance and adjusted R². The gam() function uses penalized regression splines (like thin plate splines or cubic splines) to model non-linear relationships between predictors and the response. Smoothness is not fixed in advance but is estimated during model fitting using criteria like: The models were estimated separately for each health and economic indicator, using per capita carbon emissions as predictors. For each smooth function in the GAM, the associated p-value was first examined. When the p-value was very high, indicating a lack of statistical significance, the corresponding emissions variable was considered to have no meaningful effect on the outcome. For smooth terms with p-values below 0.1, the graphical representation of the function was then inspected to identify the overall trend, whether increasing, decreasing, or approximately constant. In addition, the R² of each model was evaluated to ensure that the overall fit was satisfactory before interpreting individual effects. This combined approach of statistical testing, visual inspection, and model validation ensured that both the strength and the direction of each variable's effect were accurately assessed.

• GCV (Generalized Cross-Validation),
• UBRE (Un-Biased Risk Estimator),
• AIC, REML (Restricted Maximum Likelihood),
• NCV (Naive Cross-Validation) (30).

2.4 Log transformation

In this study, the response variables (indicators) used in the Generalized Additive Models (GAMs) were log-transformed prior to the analysis. This was necessary because many of the health and economic indicators, such as mortality rates, health expenditures, and GDP per capita, exhibit highly skewed distributions, with a few extreme values that can disproportionately influence the model. Using a logarithmic transformation helps to even out the data, making it more balanced and easier to compare countries that vary significantly in size, income, or pollution levels. Additionally, the countries considered in this research have different currency units; therefore, transforming by a logarithmic function provides an unbiased framework for the analysis. By rescaling the indicators in this way, the models provide a more robust assessment of the relationships between CO₂ emissions and the selected outcomes across both developed and developing countries.

3 Results

To illustrate the methodology and its outcomes, this section first examines the case of the United Kingdom before extending the analysis to the broader, global perspective.

3.1 United Kingdom example

3.1.1 Emission trends

Since 1960, the UK's emissions by fuel type have undergone a major transformation: coal (solid emissions), which was the dominant source in the 1960s and 1970s, has sharply declined, reaching near zero after the closure of the last coal plants in the 2020s (31). Gas emissions increased significantly from the 1990s onward, becoming the main contributor as coal was phased out. Oil or liquid emissions have remained relatively stable over time, with a slight increase in their share due to the persistence of transport-related demand. Overall, emissions in the United Kingdom are decreasing across all fuel types as depicted by Figure 6. We will examine how this trend in emissions affects the various indicators.

3.1.2 GDP per capita results

If we look closely at the results for a specific indicator, here GDP per capita, we obtain the following data.

The model fit is very strong, as indicated by R² = 0.978 and deviance explained = 0.982. The R² shows that approximately 97.8% of the variability in GDP per capita is explained by the model, meaning that the selected smooth terms capture nearly all the observed variation. Similarly, the deviance explained indicates that 98.2% of the total deviance (a generalization of variance for GAMs) is accounted for by the model, confirming an excellent overall fit. Together, these metrics suggest that the GAM accurately represents the relationship between emissions and GDP per capita for the United Kingdom.

The GAM results presented in Table 3 and Figure 7 show distinct effects of different types of emissions on GDP per capita.

• - edf (estimated degrees of freedom) indicates the flexibility of the smooth: higher values mean a more complex, non-linear relationship. For example, s(Gas) has an edf of 5.39, showing a strongly nonlinear effect, while s(Liquid) is close to 1.22, indicating a nearly linear effect.
• - F is the F-statistic for the smooth term, measuring how strongly the variable contributes to explaining variation in the response. Higher F values correspond to more significant effects, Gas has the highest F-Statistic value (9.29672522), followed by Solid (7.18856246) and liquid has the least with 3.42467603, the p-value also agrees with this trend.
• - P-value indicates the statistical significance of the smooth: very low p-values such as s(Gas) at 1.35 × 10⁻⁶ show a highly significant effect, while values near 0.05, such as s(Liquid) with p-value 0.054557829 are marginally significant. These coefficients form the basis of our inference and recommendations.
• - Significance uses a conventional coding: *** means highly significant, and. means marginally significant.
• - Trend summarises the direction of the effect over the range of the variable. Here, s(Gas) has a positive effect on the response variable, while s(Liquid) and s(Solid) have negative effects.

Table 3. GAM results for the GDP per capita indicator for the United Kingdom.

Overall, these results indicate that while gas emissions are associated with an increase (positive trend) in GDP per capita, solid emissions significantly reduce it, and liquid emissions may have a slight negative impact, both showing a negative trend at varying significance levels, as seen in Table 3.

3.1.3 Other indicator trends

The GAM results for the United Kingdom reveal a clear post-EKC pattern (8, 29). Concerning health expenditure, liquid and solid emissions reduce per capita spending, suggesting that pollution undermines both economic output and the capacity to invest in health. Regarding health outcomes, liquid fuels stand out as the most harmful; they consistently increase cardiovascular mortality and prevalence, contribute to higher infant mortality, and are linked to elevated overall death rates. Solid fuels also worsen mortality and prevalence, although with more mixed effects, while gas shows weaker or marginal associations, sometimes even appearing protective in the case of infant mortality. Respiratory outcomes are more nuanced, with liquid fuels increasing prevalence but reducing mortality, whereas gas primarily raises respiratory deaths.

3.2 Overall results

The results are gathered in Table 4. The categories are divided into nine groups based on income levels and include a global classification. The results vary depending on the income group and the type of emissions: gaseous, liquid, or solid. Each type of fuel can increase, decrease, or have no effect on the concerned indicator. The following sections will detail the most important results.

Table 4. Results per indicator per income group and globally.

3.2.1 Global results

The results indicate a strong positive association between gas consumption and GDP per capita. Overall, 80% of observations show an increasing effect, with a stronger association in high-income countries (90%) compared to middle-income countries (70%). Liquid fuels also demonstrate a positive relationship (63% total increase effect). On the other hand, solid fuels have mixed effects. In middle-income countries, there is a 60% increase in association, but in high-income countries, there is a 40% decrease in effect. This shows that advanced economies are less dependent on solid fuels. These findings indicate a close correlation between the adoption of cleaner fuels, especially gas, and economic development. For death rates, gas consumption is associated with reduced overall death rates, with 48% of total observations indicating a decreasing effect. The effect is stronger in middle-income countries (55% decrease) than in high-income countries (40% decrease). Liquid fuels show a similar trend (50% total decrease), while solid fuels demonstrate a 50% decrease effect overall. This pattern suggests that modernising energy sources may contribute to improvements in overall mortality, particularly in transitioning economies. Taken together, these patterns highlight a trade-off between short-term economic benefits from gas and liquid fuels and significant long-term health costs, particularly from liquid and solid emissions.

3.2.2 Comparison between developed and developing countries

When comparing high- and middle-income groups, notable contrasts emerge. In high-income countries, gas is almost systematically associated with higher GDP per capita (90%) and contributes to reducing infant mortality (95%), but it also increases deaths from chronic respiratory diseases (70%) and cardiovascular prevalence (65%). In these countries, liquid fuels primarily boost GDP by 60%, but they also contribute to higher death rates (30%–35%) and a higher prevalence of chronic diseases. In middle-income countries, gas still promotes growth (70%), but its protective effects on infant mortality are weaker (80%), and it is more often linked to higher death rates (55%) compared to high-income countries. Moreover, liquid emissions in middle-income countries appear even more harmful for mortality (65% increase in the general death rate and 55% in infant mortality). Solid fuels, finally, show widespread negative health effects across both groups, but their economic role is more pronounced in middle-income countries (60% increase in GDP compared to only 30% in high-income countries). This suggests that while high-income countries bear significant health costs from emissions, middle-income countries face a sharper trade-off where economic reliance on fossil fuels comes at the expense of more severe public health impacts.

Table 5 summarises the main differences between high-income and middle-income countries, showing the percentage of countries within each group affected by the indicator.

Table 5. Discriminating criteria between high-income and middle-income countries, by type of fuel and the proportion of countries concerned.

3.3 Classification of countries according to shared GAM profile

As an example, the GAM profiles for the UK's GDP and death rate are shown in Figure 5 and Figure 7, respectively. In a similar manner, we acquired the GAM profiles for every other nation we chose. The countries with comparable tendencies are grouped together in the GAM results. As a result, nations can be categorised into five groupings, as Table 6 illustrates:

• - Group 1: Developed countries meeting developed criteria
• - Group 2: Developing countries meeting developing criteria
• - Group 3: Developed countries meeting developing criteria
• - Group 4: Developing countries meeting developed criteria
• - Group 5: Countries exhibiting no specific pattern

Table 6. Classification based on GAM results.

If a country meets at least half of the criteria assigned to its group, it is not reclassified. If it meets half of the criteria of the other group, it is placed in Groups 3 or 4, depending on its original group. If it does not meet half of the criteria of either group, it is placed in the last group. The classification shows that the official level of development does not always correspond to actual environmental and economic performance. Some developed countries display weak performance, while some developing countries exceed expectations.

Group 1, composed mainly of Organisation for Economic Co-operation and Development (OECD) economies, broadly follows the expected “developed” pattern. In these countries, growth and emissions are usually not linked, which means that economic growth is less dependent on fossil fuel use. Nevertheless, the health impacts of certain pollutants remain visible, highlighting that development alone does not fully eliminate environmental and health externalities.

Group 2, made up of developing countries that meet “developing” criteria, reflects the expected pattern where fossil fuels still provide economic benefits but at the expense of health. In these contexts, resource-intensive sectors often drive growth, with short-term economic gains frequently offset by deteriorating health outcomes.

Group 3 includes developed countries that instead align with “developing” criteria, showing that economic advancement does not guarantee superior environmental outcomes. Countries in this group tend to link growth and pollution, which means that as incomes rise, so do emissions and health problems.

Group 4 illustrates the opposite phenomenon: developing countries that behave more like high-income economies in terms of emission impacts. These countries may benefit from cleaner energy transitions, stronger environmental regulation, or specific features of their economies that limit the negative side effects of growth.

Group 5 unites countries that lack a consistent trend, highlighting the limitations of a universal framework. This leftover category shows that local factors, such as dependence on certain fuels, the strength of the health system, geographic weaknesses, or political decisions, can lead to paths that differ from global trends.

4 Discussion

4.1 United Kingdom case

The contradictions of energy transitions are highlighted by the UK results. Gas's significant contribution to economic restructuring following the demise of coal and its availability as a reasonably flexible and efficient energy source are reflected in its high positive impact on GDP per capita. However, there are drawbacks to this economic advantage: gas still raises respiratory mortality because the pollutants released concurrently with CO₂ aggravate pre-existing lung conditions and promote respiratory tract inflammation (21). This indicates that the move away from coal corresponded to a change in the burden of sickness rather than a “clean” solution. Solid fuels support the policy direction of the coal phase-out followed in the UK since the 1980s by confirming their reputation as the most detrimental, both economically and health wise (33). But probably most worrisome are the long-lasting and serious health effects of liquid fuels, which are mostly caused by the transportation and industrial sectors. Even if they have a weaker direct correlation with GDP, they are consistently linked to greater mortality rates from cardiovascular illness because accompanying pollutants (PM₂, NO₅, and CO) cause oxidative stress, inflammation, and circulatory diseases that worsen heart ailments. Because these pollutants harm foetal development, damage the lungs and immune systems of newborns, and increase the risk of serious respiratory and cardiovascular disorders, they are also associated with infant mortality (21). This suggests that their externalities might outweigh their financial gains, raising concerns about the suitability of current mitigation and regulation strategies.

Looking at long-term emission trends, the UK's post-EKC pattern seems to be mostly caused by decreases in the heating and power sectors since the 1990s, while emissions from other sectors have remained relatively constant (33). This implies that improvements are the consequence of targeted decarbonisation measures centred on electricity generation and domestic heating rather than an overall economic shift. This creates an imbalance: while cutting back on coal-based power generation has had clear benefits for the economy and people's health, ongoing liquid emissions from industry and transportation are still hindering progress. These disparate sectoral dynamics highlight the limitations of viewing the UK's trajectory as a simple EKC case: growth has only been accompanied by lower emissions in areas where there has been significant regulatory intervention, and “market forces” by themselves have not been enough to cut emissions in other sectors. When combined, these results highlight how the UK's trajectory does reflect a post-EKC pattern, although an uneven and partial one: while the worst effects have been lessened by the phase-out of coal, health outcomes are still being compromised by reliance on gas and liquid fuels.

4.2 Global trends

The global patterns underscore the complexity and heterogeneity of emission impacts across countries and indicators. Gas emissions appear to offer economic dividends, with growth benefits observed in most cases, as they increase GDP and health expenditure per capita. This is in line with the fact that gas is often seen as a cleaner and more efficient transitional fuel than coal and oil, especially when it comes to generating electricity. However, the mixed health results, especially the rise in heart disease caused by harmful pollutants from burning gas (which cause inflammation, oxidative stress, and problems with blood circulation), show that the advantages of gas are not always positive. Furthermore, the rising global share of gas use may induce a rebound effect, whereby increased availability and consumption lead to higher overall energy use, potentially amplifying its environmental and health impacts, which remain underrecognized in policy discussions (5). Liquid fuels emerge as the most insidious: while their economic impact is less consistent, their health costs are substantial, spanning increased mortality and chronic disease prevalence. This impact reflects the pervasive use of oil in transport and industry, sectors that are harder to decarbonise and where exposure to pollutants is widespread and sustained. Solid fuels, meanwhile, remain the most clearly detrimental to public health, reinforcing well-established evidence of their toxic effects on respiratory and cardiovascular systems. The occasional reduction in health expenditure associated with solid fuels likely reflects constrained public budgets or underreporting rather than actual improvements in population health. These findings reveal a persistent trade-off: long-term health burdens often accompany economic gains, particularly in areas where liquid and solid fuels are prevalent. This challenges the optimistic narrative of a universal Environmental Kuznets Curve, suggesting instead that without targeted health and environmental interventions, economic growth alone is insufficient to ensure improved well-being. The results also call attention to the need for more integrated approaches that simultaneously address energy, economic, and health objectives, especially in countries where the transition away from fossil fuels remains partial or uneven.

4.3 Comparison between developed and developing countries

The comparison between high- and middle-income countries highlights the differentiated ways in which fossil fuel dependence interacts with economic development and health outcomes. In high-income countries, gas emerges as a double-edged sword: it systematically supports higher GDP per capita and helps reduce infant mortality, but it also generates substantial health burdens through a higher prevalence of cardiovascular and respiratory diseases. This situation reflects both technological advantages, such as cleaner combustion technologies and better health systems that mitigate infant mortality, and structural vulnerabilities, because once a certain income level is reached, economic growth is no longer sufficient to offset the negative effects of emissions on public health. In middle-income countries, the trade-offs are sharper. Gas remains growth-enhancing. Protective health effects are weaker and associated with higher mortality, underlining the lack of compensatory health infrastructures. This scenario contrasts with developed countries where increases in disease prevalence result in lower mortality thanks to strong healthcare systems. Moreover, in many of these countries, a significant portion of the environmental impact stems from the offshoring of waste processing and polluting industries by developed nations, effectively transferring the ecological burden while allowing higher-income countries to maintain cleaner local environments. However, since the emissions in question are associated with energy consumption, this issue has no direct impact on our study. The harmful effects of liquid fuels are even more noticeable: they help industries grow, but they also lead to higher rates of death in infants and the general population, indicating that their health risks are especially serious in places with weaker regulations, lower fuel quality standards, and less effective healthcare. Solid fuels illustrate the clearest developmental divide: while they still contribute significantly to GDP growth in middle-income countries, their negative health consequences remain widespread, showing that reliance on coal and biomass continues to underpin economic activity at the expense of population health (3). By contrast, in high-income countries, solid fuels have largely lost their economic role, and their health costs remain the main legacy of their historical use. Overall, these contrasts suggest that the Environmental Kuznets Curve dynamic is highly context-dependent. High-income countries show that growth does not eliminate health costs from fossil fuels but rather transforms them into disease burdens. Middle-income countries, meanwhile, remain locked in a more acute growth-health trade-off, where fossil fuels, particularly liquids and solids, still drive economic gains but at the expense of elevated mortality and constrained health expenditure. This indicates that global inequalities in health and environmental transitions are not simply a matter of income level, but also of energy mix, regulatory capacity, and the extent to which growth is decoupled from fossil fuel dependence.

4.4 Classification highlights

The new classification presented in this study highlights the divergence between formal development status and actual environmental economic performance (See Figure 8). The presence of Group 3 shows that economic advancement does not guarantee superior environmental outcomes. For example, countries in this group tend to exhibit weaker decoupling between growth and pollution, pointing to structural energy dependencies or policy shortcomings. Greece continues to experience relatively high pollution levels due to its lingering reliance on coal and the effects of economic crises that have delayed energy transitions. The UAE exhibits extremely high per capita emissions, driven by fossil fuel dependence, intensive energy consumption for air conditioning and desalination, and rapid urbanisation (34). Similarly, the United States maintains high per capita emissions and only gradual reductions in certain sectors, reflecting large-scale consumption, transport dependency, and variable environmental regulation (35). Group 4 suggests that targeted energy transitions, regulatory frameworks, or rapid technological uptake can allow developing economies to leapfrog certain stages of the environmental Kuznets trajectory. Kazakhstan and Brazil have undergone significant industrialisation and economic diversification, with large service sectors reducing their share of highly polluting industries. Export-oriented industries often adopt higher environmental standards to meet global demand. They have also established formal environmental regulations and agencies, such as Brazil's IBAMA (36) and Kazakhstan's environmental code reforms (37), which, while imperfectly enforced, create frameworks similar to those in high-income countries. Group 5 emphasises the persistence of heterogeneity. Rather than being anomalies, these cases highlight the importance of contextual analysis when evaluating the interplay between growth, emissions, and health. Taken together, this typology underlines that official income classification is a poor proxy for environmental and health performance. The existence of “mismatched” countries in Groups 3 and 4 shows that trajectories are not predetermined by income but are mediated by policy, energy mix, and institutional choices. Instead of supporting a simple EKC story, these findings suggest that we need to understand transitions in a more varied and country-specific way, where both rich and poor countries can either fall behind or perform better than what we expect.

4.5 Limitations

Due to a lack of data, this study only contains a small number of nations from some continents, especially Africa. More nations from various continents could be added to the sample. Other methods, such as dynamic, spatial, or hybrid machine learning–econometric models, could be deployed in the statistical methodology. A more comprehensive perspective and stronger conclusions would result from this (8, 29). In addition to offering a better foundation for the creation and application of policies, the findings would deepen knowledge of the patterns and their implications (3).

5 Conclusions

This research set out to investigate the complex relationships between economic growth, environmental pressures, and health outcomes using Generalized Additive Models (GAMs). By analysing different types of emissions (gaseous, liquid, and solid fuels) and their impacts on economic and health indicators across countries, it was possible to shed light on how development status shapes these dynamics. The results show that although the Environmental Kuznets Curve (EKC) theory helps explain some aspects, the situation is much more complicated, with important differences between high- and middle-income countries. A classification of countries into distinct groups based on their GAM profile further underlines the heterogeneity of outcomes. Some developed countries displayed patterns more typical of developing ones, failing to convert growth into environmental or health improvements, while certain developing countries showed surprisingly strong performance in line with more advanced economies. This discrepancy demonstrates that “official” development categories, such as the World Bank's, do not always reflect real performance in environmental and health terms. Instead, institutional quality, regulatory frameworks, and healthcare systems emerge as crucial factors in explaining cross-country differences. Comparing high- and middle-income countries reveals additional insights. Once developed economies reach a certain growth threshold, the negative health impacts of emissions tend to outweigh the economic benefits. Infant mortality closely correlates with solid emissions, while gas emissions often increase overall mortality. Nevertheless, paradoxically, in these same countries, strong healthcare systems help mitigate deaths from chronic respiratory and cardiovascular diseases, even when disease prevalence is high. In middle-income countries, the picture is different: emissions, particularly from gas and solid fuels, still support growth, and some health indicators, like infant mortality, improve despite higher pollution levels. However, these benefits come at the cost of rising mortality from chronic diseases, pointing to a delayed but looming health burden. The study demonstrated that the interplay between emissions, growth, and health is complex and shaped by both economic status and institutional capacity. While global trends point toward eventual improvements in health and the environment as countries grow richer, relying solely on growth is insufficient and even risky. Economic growth alone does not guarantee environmental improvement or better health outcomes. Proactive policies remain indispensable. Therefore, future research should not only refine the modelling of emissions and health but also incorporate institutional and behavioural dimensions, which are key to understanding why some countries succeed in decoupling growth from environmental harm while others do not. By combining cross-country emission data, health indicators, and economic variables, this study contributes to the wider field in several ways. It provides empirical evidence supporting and challenging the EKC hypothesis, highlighting cases where it does not hold. It also emphasises the need to integrate health outcomes into environmental-economic analyses, showing that policy decisions focusing only on GDP growth may overlook significant public health risks (3, 8, 34). Furthermore, the study provides policymakers with insights into tailoring environmental regulations and energy strategies to national contexts, rather than adopting a uniform approach (38). Given the diverse results across countries, national and international collaborations are encouraged so that countries with a fair level of control over emissions can easily partner with those that have yet to address this challenge, thereby maintaining environmental balance across regions and ultimately making the world safer and healthier for its inhabitants. Most studies focus on specific countries or sectors, such as transportation and industries (3, 29), but this study considers several countries across various factors, providing a broader perspective on the dynamics of these emissions and avoiding limitations in our approach.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. AI was used in the english proof reading process.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Rakesh Kumar, Auburn University, United States

Reviewed by: Oluwole Joseph Oladunni, Durban University of Technology, South Africa

Shrabani Mukherjee, Shiv Nadar University Chennai, India