The study arrives at a moment when China faces converging pressures. Healthcare costs are rising steadily, air-pollution exposure remains a persistent public health burden despite major clean-air campaigns, urbanization continues to reshape where and how people live, and digital transformation is sweeping through hospitals, insurers, and public administration at remarkable speed. Previous research has examined how pollution, energy production, and technological change influence health spending, but most of those studies relied on mean-based regression techniques that implicitly assume the effect of, say, an AI investment or a change in particulate-matter concentrations is identical across the entire distribution of spending outcomes. The Hezhou team argues that this assumption obscures precisely the dynamics that matter most for policy, because the drivers of modest baseline health budgets may behave very differently from the drivers of ballooning expenditures in periods of elevated demand.

To capture that heterogeneity, the authors deployed a battery of quantile-based econometric tools rather than conventional ordinary least squares approaches. The analysis begins with quantile ADF and quantile KPSS stationarity tests, which check whether the stochastic properties of the variables hold across different quantiles of their distributions rather than merely at the mean. Establishing stationarity at the quantile level is a critical precondition: if the series are not stationary in the relevant regions, regression estimates can be spurious, producing relationships that are statistical artifacts rather than genuine economic signals. Once the stationarity properties were confirmed, the researchers moved to their central instrument, a multivariate quantile-on-quantile regression framework, supplemented by conventional quantile regression as a robustness check.

The quantile-on-quantile method is the technical heart of the paper, and its logic is worth unpacking. Unlike standard regression, which asks how the mean of a dependent variable responds to changes in an independent variable, quantile-on-quantile regression examines how a specific quantile of the dependent variable’s distribution responds to a specific quantile of the independent variable’s distribution. This produces a matrix of estimates: the effect of artificial intelligence on the tenth percentile of health expenditure may bear little resemblance to its effect on the ninetieth percentile, and the relationship may even flip sign across the distribution. In doing so, the technique captures tail dependence and asymmetric interactions that average-based models are structurally incapable of detecting. The authors report that this approach revealed clear nonlinear patterns, stable conditional distributions, and strong quantile-specific differences in how the drivers of health spending operate.

Perhaps the most striking result concerns artificial intelligence, which the study characterizes as having a genuinely mixed, double-edged effect. On one side, AI may push healthcare spending upward: building digital infrastructure, deploying diagnostic algorithms, and integrating machine-learning systems into hospitals require substantial upfront capital investment, and those costs feed directly into expenditure figures. On the other side, AI can relieve cost pressure through efficiency gains, earlier disease detection, and better allocation of scarce medical resources. An AI system that flags disease at a treatable stage, for instance, may avert the far costlier hospitalizations that advanced illness demands. The quantile framework reveals that these opposing forces do not cancel into a single tidy average; instead, they manifest differently across the spending distribution, with cost-increasing channels dominating in some quantile regions and cost-reducing efficiency effects dominating in others.

The findings on green growth and air quality reinforce the study’s central claim that China’s health expenditure sits at the intersection of a complex technology–energy–environment nexus. Green economic growth, in the authors’ framing, is not simply an environmental ambition but a health-finance variable: cleaner energy production and more sustainable growth pathways can reduce pollution-related illness and thereby relieve pressure on health budgets, although the relationship again varies across quantiles rather than holding uniformly. This distributional sensitivity echoes earlier literature the authors cite, including work showing that China’s clean-air actions have alleviated health-expenditure inequality, and international studies linking environmental quality, energy consumption, and healthcare costs in contexts ranging from Malaysia and the G7 economies to Southeast Asia. What the new study adds is the demonstration that these relationships in China are not merely dynamic but contingent on where in the spending distribution an observation falls.

The methodological shift from averages to quantiles also has a substantive interpretation beyond econometric elegance. Low-spending regimes may correspond to periods or regions where healthcare demand is relatively contained and where efficiency gains from digitalization can genuinely bend the cost curve. High-spending regimes, by contrast, may reflect surges driven by pollution episodes, aging populations, or expensive diagnostic investment, in which case the marginal effect of any given driver changes character entirely. By modeling these regimes separately, the study effectively treats health expenditure as a nonlinear outcome shaped simultaneously by artificial intelligence, green growth, energy production, and air quality, rather than as a passive sum of independent average effects. The robustness checks using conventional quantile regression confirmed that the quantile-on-quantile results were not sensitive to the specific estimation technique.

For policymakers in Beijing and provincial capitals, the implications are sobering and prescriptive at once. The authors conclude that long-term health-fiscal stability in China will require coordinated policy across at least five domains: artificial intelligence deployment, green growth strategy, clean energy transition, urban planning, and air-quality improvement. None of these levers operates in isolation. An aggressive national AI program that expands digital health infrastructure without parallel investment in clean energy could, according to the study’s quantile evidence, raise rather than reduce healthcare costs in certain spending regimes. Conversely, pairing AI-driven diagnostic efficiency with sustained reductions in particulate pollution could compound savings, particularly in the upper quantiles of expenditure where costs are most burdensome. The study thus argues against siloed policymaking and for an integrated framework in which technology policy and environmental policy are evaluated partly through their joint effects on health budgets.

The research also speaks to a rapidly expanding international literature on AI’s economic footprint in health systems. Recent systematic reviews of medical AI’s technological maturity and cost-effectiveness, studies of AI’s role in reducing carbon abatement costs in China’s industrial sector, and analyses of AI applications in public expenditure management in low-income countries all point to a technology whose fiscal consequences are contingent on context and implementation. The Hezhou study contributes a distinctive perspective by situating AI within the same analytical frame as energy production and air quality, and by insisting that all of these effects be measured across the full distribution of outcomes. In a country where healthcare spending is growing faster than many comparable economies and where hundreds of millions of urban residents remain exposed to elevated pollution levels, that distributional lens is more than a statistical refinement; it determines which policies will actually deliver relief, and for whom.

The study, published in Volume 19 of Air Quality, Atmosphere & Health as article number 159, was supported by the Research Start-up Fund for Doctoral Professors at Hezhou University under grant number 2023BSQD05, and was accepted on 16 June 2026 before appearing in its version of record on 6 July 2026. The authors note that the datasets analyzed are available from the corresponding author upon reasonable request. As China continues its twin transitions toward a digital economy and a greener growth model, the research offers a quantitative warning and a quantitative opportunity: the same technologies and policies that will define the country’s economic future are already writing themselves into its health ledger, and only by reading that ledger across its full distribution, rather than at its average, can policymakers understand what those pathways will cost, and where they will pay off.