Public Services Buffer Air Pollution Risks to Perinatal Health in China

A longitudinal investigation evaluating population-level data across thirty-one Chinese provinces over two decades has revealed that investments in public services and digital information resources significantly weaken the harmful relationship between ambient air pollution and perinatal health. Published by researchers Kexin Xie and Shuo Xu in the journal SSM – Population Health, the study analyzed a balanced panel dataset spanning from 2003 to 2022 to examine how regional institutional resources condition environmental health vulnerabilities. Using two-way fixed-effects regression models, the authors established a baseline finding indicating that higher population-weighted annual fine particulate matter exposure exhibits a statistically significant negative association with perinatal health, yielding a regression coefficient of minus 0.072 with a p-value of 0.013. Because perinatal health was quantified as the mathematical inverse of the perinatal mortality rate per one thousand births, this negative parameter demonstrates that provinces experiencing elevated ambient particulate concentrations suffered systematically worse perinatal survival rates even after accounting for province-specific unobserved characteristics and nationwide temporal trends.

The conceptual foundation of the investigation integrates environmental justice paradigms, the social determinants of health framework, and contemporary extensions of fundamental cause theory. Classical epidemiological inquiries into environmental health have predominantly treated fine particulate matter as an isolated biological hazard, focusing heavily on direct physiological pathways such as systemic inflammation, oxidative stress, immune disruption, and placental vascular dysfunction. While these biological mechanisms remain scientifically critical, the authors argued that a comprehensive population health perspective requires analyzing the contextual and institutional conditions under which physical hazards translate into tangible health harms. Fundamental cause theory posits that health inequalities are continuously reproduced through disparate access to flexible social and institutional resources. Xie and Xu extended this insight to macro-level institutional systems, arguing that regional public services do not necessarily have to eliminate ambient pollution concentrations at their emission sources in order to modify health consequences. Instead, well-developed civic systems provide vital institutional protection by supplying timely clinical care, health literacy, environmental surveillance, and organized institutional capacity to mitigate perinatal risks.

To rigorously evaluate these systemic dynamics, the research team constructed a comprehensive panel dataset utilizing multi-source atmospheric and demographic records. Exposure to ambient air pollution was primarily measured using population-weighted annual average fine particulate matter concentrations derived from satellite-based aerosol optical depth retrievals integrated with the GEOS-Chem chemical transport model by the Atmospheric Composition Analysis Group at Washington University. This spatial weighting technique accurately reflected human demographic exposure across diverse provincial topographies, yielding a sample mean log concentration of 3.705 with a standard deviation of 0.466 across the twenty-year observation window. Geographic-mean concentrations were also compiled as an alternative exposure metric, showing a mean log concentration of 3.531. Perinatal health outcomes were derived from the China Health and Family Planning Statistical Yearbook and provincial statistical yearbooks, defining perinatal mortality as fetal deaths occurring at or beyond twenty-eight completed gestational weeks plus neonatal fatalities within the first seven days of life per one thousand total births.

Evaluating public service provision as a multidimensional protective apparatus required the investigators to distinguish between traditional fiscal allocations and emerging digital information environments. Traditional public services were categorized into three core functional sectors: local budgetary expenditure on medical and healthcare services, spending on public education, and fiscal allocations dedicated to environmental protection. To ensure rigorous temporal comparability across the twenty-year study period, all monetary figures were standardized on a per-capita resident basis and converted into constant 2020 price indices using the national consumer price index chained across consecutive survey years. Across the panel, per-capita local fiscal expenditure averaged 0.081 units of ten thousand yuan for medical services, 0.177 units for public education, and 0.034 units for environmental protection. These fiscal metrics were explicitly interpreted by the researchers as operational proxies for local governance capacity and welfare resource availability rather than direct, individualized metrics of clinical treatment quality or patient utilization rates.

Recognizing the growing relevance of digital transformation within population health, the authors incorporated three provincial-level contextual metrics grounded in the digital determinants of health framework. Physical internet connectivity was captured through the provincial proportion of national IPv4 internet addresses, online application prevalence was measured via registered web domain names per ten thousand residents, and online informational richness was quantified by average webpage byte size. To minimize omitted variable bias and spurious confounding, the analytical models controlled for a extensive vector of time-varying provincial socioeconomic covariates, including log gross domestic product per capita, log population density, log urbanization rates, the population sex ratio, full-time equivalent research and development personnel, average schooling duration, and per-capita healthcare expenditure. Domain-specific adjustments were strictly enforced in moderation models: healthcare models excluded general healthcare spending covariates to avoid duplication, while public education models omitted provincial average schooling years to prevent conceptual overlap with educational investment pathways.

The initial phase of moderation modeling employed linear interaction equations to evaluate whether public service provision systematically attenuates the negative slope connecting particulate pollution with perinatal health. When standardizing the transformed service indicators across the full panel, a positive interaction coefficient denotes a protective buffering effect. The empirical estimates indicated that online information resources exhibited the most consistent and statistically significant linear attenuation effect, displaying an interaction parameter of 0.008 with a p-value of 0.018. This demonstrates that provinces with richer online informational content experienced a less severe negative slope between particulate pollution and perinatal outcomes. Local budgetary investments in healthcare and education also demonstrated positive buffering coefficients of 0.014, although these linear interaction terms achieved marginal statistical significance with p-values of 0.095 and 0.065, respectively. In contrast, environmental protection expenditures, physical internet address proportions, and online application densities showed no discernible linear moderating associations in the continuous interaction framework.

Hypothesizing that civic infrastructure might not operate as a uniform continuous buffer but may instead require discrete capacity thresholds before achieving protective efficacy, Xie and Xu implemented Hansen panel threshold regression techniques. This non-linear structural approach searched across three hundred candidate threshold values with one percent trimming, testing whether the negative regression coefficient of particulate matter shifted significantly between low-capacity and high-capacity service regimes. Statistical inference was conducted using 999 wild-cluster bootstrap replications with province-level Rademacher weights to rigorously address the unidentified threshold parameter under the null hypothesis of slope equality. Complementary Wald tests using province-clustered standard errors subsequently evaluated coefficient divergence at the estimated threshold parameters. The empirical bootstrap tests confirmed significant regime shifts across four of the six analyzed dimensions: local education investments, internet infrastructure, online application volume, and online informational resources, all exhibiting bootstrap p-values below the 0.001 threshold.

Within these four supported threshold domains, high-capacity institutional regimes demonstrated pronounced reductions in the severity of air pollution’s adverse perinatal associations. For provincial education expenditures, the estimated fiscal threshold was identified at 0.030 units of ten thousand yuan per person; below this cutoff, the particulate matter regression coefficient was minus 0.080, but it attenuated to minus 0.071 in the higher-spending regime, yielding a statistically significant Wald contrast p-value of 0.008. For online information resources, the estimated threshold occurred at 74.000 average webpage bytes; below this level, the pollution slope was minus 0.062, moderating to minus 0.055 above the cutoff with a Wald p-value of 0.039. Internet infrastructure and online applications similarly displayed dramatic regime moderation, shifting particulate matter slopes from minus 0.070 to minus 0.059 and from minus 0.073 to minus 0.063, respectively, with both Wald contrasts yielding p-values below 0.001. Healthcare expenditures displayed a suggestive threshold at 0.015 units with a bootstrap p-value of 0.053, whereas environmental protection investments showed no statistically supported regime structure.

To confirm the stability of their empirical conclusions, the authors conducted multiple rigorous sensitivity checks addressing measurement choices, data completion techniques, and alternative variable transformations. When restricting the panel sample by entirely excluding observations that required linear interpolation or endpoint extrapolation across outcome, exposure, or moderator variables, the baseline adverse particulate association persisted at minus 0.051 with a p-value of 0.073. Within this restricted non-interpolated subsample, the linear buffering interactions for education investment remained statistically significant at a p-value of 0.031, while healthcare spending and online information resources maintained positive buffering coefficients with p-values of 0.075 and 0.088. When the model was re-estimated using the natural logarithm of perinatal mortality rates as the dependent variable, particulate exposure consistently exhibited an adverse positive coefficient of 0.138, and interaction signs aligned fully with risk attenuation, although statistical precision was reduced. Furthermore, estimating threshold regressions using nominal current-price fiscal expenditures replicated the core structural regime divisions identified in constant-price specifications.

The study provides crucial empirical insights for environmental health governance while clearly acknowledging several methodological boundaries inherent to provincial macro-level data. The authors noted that aggregated provincial panel figures cannot measure individual-level maternal toxicological exposures, behavioral adaptations, clinical care quality, or within-province spatial disparities between rural villages and metropolitan centers. Additionally, high correlations among public investment sectors and potential unobserved time-varying confounders preclude definitive causal claims regarding specific biological mechanisms. Nevertheless, the findings demonstrate that environmental hazards do not impact human populations within an institutional vacuum. The presence of accessible digital health information, robust basic education, and adequate clinical capacity establishes an essential protective buffer that moderates vulnerability during sensitive developmental windows. Consequently, the authors concluded that while reducing industrial air emissions remains an indispensable global priority, targeted investments in maternal and neonatal support systems, environmental health literacy, and actionable digital guidance offer powerful institutional mechanisms to mitigate early-life health inequalities.

Subject of Research: The moderating role of public services and digital resources on air pollution and perinatal health

Article Title: Air pollution, public service provision, and perinatal health in China

Article References: Xie, K., & Xu, S. (2026). Air pollution, public service provision, and perinatal health in China. SSM – Population Health, 36, Article 101984. https://doi.org/10.1016/j.ssmph.2026.101984

Image Credits: AI Generated

DOI: 10.1016/j.ssmph.2026.101984

Keywords: air pollution, perinatal health, PM2.5 exposure, public service provision, digital determinants of health, panel threshold models, environmental health inequalities, perinatal mortality, pollution, public, service, provision