Impact of artificial intelligence and digital upgrading on the sustainable development of public health services: An empirical study based on double machine learning

Abstract

Objective

With the global proliferation of chronic diseases and sudden infectious outbreaks, the use of artificial intelligence and digital technologies to enhance the sustainable growth of public health services has become a key research focus. This study examines the impact and transmission mechanisms of artificial intelligence and digital upgrading on the long-term development of China’s public health services.

Methods

It visualizes and performs regression analysis on panel data from 30 provincial-level units in China from 2012 to 2024, using kernel density estimation, standard deviation ellipse, and dual machine learning models.

Results

The following conclusions are drawn: (1) The sustainable development of public health services in China shows a regional distribution pattern, with higher levels in the east and lower levels in the west. Although overall levels have improved over time, regional disparities have widened. Hotspots for sustainable growth also show a spatial development trend toward the southeast. (2) Artificial intelligence and digital upgrading significantly positively impact the sustainable growth of public health services in China. A one-unit rise in artificial intelligence and digital enhancement results in gains of 0.014% and 0.080% in the sustainability of public health services, respectively. (3) The positive effects of artificial intelligence and digital upgrading on the sustainable development of public health services exhibit heterogeneity across economic zones, resource endowments, and the North–South regional division. (4) Digital upgrading and artificial intelligence significantly enhance the development of green technological innovation, green patent technology innovation, and green utility model innovation. Through this pathway, the sustainable development performance of public health services will be further improved.

Conclusion

Digital upgrading and artificial intelligence improve the sustainable development of public health services in China through multiple pathways, including spatial distribution dynamics, direct positive effects, heterogeneous regional impacts, and enhanced green technological innovation.

1. Introduction

Public health services hold a crucial role within the national health security system. Their potential for sustainable development influences the advancement of universal health coverage objectives and affects broader socioeconomic development. ¹ The worldwide proliferation of chronic noncommunicable diseases in recent years, Global public health service systems currently face several challenges: the unchecked trend of population aging, escalating costs of treating noncommunicable chronic diseases, and inadequacies in emergency response mechanisms for emerging and sudden infectious diseases^2,3; However, the rising public demand for health and well-being has highlighted the inherent deficiencies of traditional frameworks and the practical limitations of global disease control and treatment. There is an urgent need to seek transformative answers through technological advancement and innovation. ⁴ Against the backdrop of the continuous evolution of digital technologies, Artificial intelligence technologies are transforming public health early warning and decision-making systems by using their superior computational and predictive capabilities. Meanwhile, thorough digital upgrading has a substantial impact on the inclusiveness of public health care and the efficient allocation of resources.^5,6 In August 2025, the State Council stated in its Opinions on Deeply Implementing the “Artificial Intelligence Plus” Initiative that the positive effects of artificial intelligence and other digital technologies, particularly in supporting medical diagnosis and health management, should be fully leveraged to improve quality of life. In this context, Identifying the specific pathways and boundary conditions through which digital upgrading and AI applications impact the sustainable development of public health services is crucial for optimizing future resource allocation and guiding policy-making.

Scholars, both domestic and international, have investigated the integration of digital technology with public health services from three perspectives: technological application exploration, empirical outcome evaluation, and effect mechanism analysis. Regarding technological applications, Dankwa-Mullan ⁷ and King, Doueiri ⁸ assert that the deployment of artificial intelligence technology can enhance participatory science in public health and foster health equity. Nonetheless, while highlighting the extensive potential of integrating digital technology, it is crucial to mitigate the ethical and social hazards arising from the skewed dissemination of data and information. Yin ⁹ contended that the use of algorithms and models, including machine learning and deep learning, enables accurate processing of diverse data, thereby enabling dynamic forecasting of potential threats to public health services. Kong, Akpudo ¹⁰ found that artificial intelligence and big data analytics can help meet public health service demands in the Global South. Charalambides and Singh ¹¹ contend that AI-driven melanoma detection and prevention systems can enhance skin cancer diagnosis and offer technological assistance for clinical cancer research.

In terms of empirical evaluation, Zhao and Fu ¹² upon analysing data from 100 survey questionnaires, it was determined that the educational strategy employing an AI-based public service system achieved a score of 21.74%, which was 21.74% higher than that of conventional home health education approaches. Zhong, Wohlars ¹³ through a bibliometric study and the development of an artificial intelligence model that leverages multidimensional factors, they forecasted clinic closures for forthcoming major public health events. The model attained a prediction accuracy of 85.45%. This improves the effectiveness of governmental agencies’ emergency decision-making during public health emergencies, constituting a prerequisite for the sustainable advancement of public health services. Kumar, Singh ¹⁴ found that integrating AI into public healthcare services is hindered by four factors: technology, ethics, organization, and policy. Raza, Norin ¹⁵ developed a PLS-SEM model to analyse questionnaire data, and the results showed that the Internet of Things and the digital divide moderated the relationships among these variables.

In conclusion, current research primarily employs a bibliometric approach to assess the viability of digital technologies and artificial intelligence in enhancing the sustainable growth of public health services, with a particular focus on their social and ethical implications. However, most studies on the relationship between these technologies and public health sustainability have not used econometric models to detect causal impacts or uncover underlying mechanisms. Furthermore, there is hardly any research that integrates digital upgrading, artificial intelligence, and the sustainability of public health care into a cohesive analytical framework. On this basis, this study uses a double machine learning model to analyse the effects, heterogeneity, and action pathways of digital upgrading and artificial intelligence on the sustainable development of public health services, and the possible marginal contributions are as follows: First, at the theoretical level, by combining digital infrastructure and artificial intelligence applications into a unified research framework, it overcomes the limitations of previous studies that focus on a single technological dimension and offers a new theoretical perspective for systematically evaluating their synergistic effects and substitution relationships. Second, at the methodological level, this work innovatively employs a dual machine learning model for causal inference, thereby advancing prior studies. This method adeptly overcomes the constraints of conventional econometric techniques in managing high-dimensional confounding variables and model specification biases. Third, at the practical level, by elucidating essential transmission mechanisms and boundary conditions, it provides empirical evidence to bolster governmental initiatives in crafting differentiated, targeted digital health policies.

2. Mechanism analysis and research hypotheses

Resource constraints and the growth of health demand are two major challenges in the reform of public health service systems. ¹⁶ Science and technology can become advanced productive forces because they can effectively promote the organic integration and coordinated development of various production factors. The systematic upgrading of artificial intelligence and digital technologies can bridge service gaps and overcome efficiency bottlenecks, ¹⁷ thereby effectively empowering the sustainable development of public health services. In terms of resource optimisation, big data-driven prediction models can accurately identify epidemic trends in infectious diseases and the onset characteristics of chronic diseases,^18,19 which helps public health organisations formulate strategies in advance and allocate resources rationally, thereby effectively addressing resource mismatch. In terms of expanding service coverage, the introduction of frontier technologies, especially the application of digital technology platforms, can enable telemedicine and mobile health monitoring, ²⁰ breaking the geographical limitations of offline medical visits, extending medical resources to a wider population, and narrowing the medical service gap.^21,22 In terms of efficiency improvement, on the one hand, the application of online medical platforms can simplify traditional medical service processes such as registration, diagnosis, and payment, shorten consultation time, and improve medical service efficiency ²³ ; on the other hand, AI-assisted diagnosis and automated management systems can improve the accuracy of diagnosis and treatment, ²⁴ reduce medical costs, and enable medical resources to flow to a broader service scope. This represents a pathway through which service coverage is extended under efficiency improvement. Based on this, this study proposes the first hypothesis:

• H1: Artificial intelligence and digital upgrading can substantially promote the sustainable development of regional public health services.

Differences in resource endowments and geographical conditions across Chinese provinces lead to different regional characteristics in the upgrading of digital technology applications, artificial intelligence development, and the sustainability of public health services. Therefore, this study argues that the empowering effects of digital upgrading and artificial intelligence on the sustainable development of public health services exhibit regional heterogeneity. ²⁵ First, the eastern region has sufficient economic support, human resources, and well-developed digital infrastructure, ²⁶ and therefore holds a leading position in digitalisation and intelligent development. ²⁷ In contrast, the central and western regions lack inherent advantages and have large economic and technological gaps compared with the eastern region. The central region takes efficiency improvement as the core objective of digital upgrading in public health services, whereas the western region faces shortages of human resources and bottlenecks in building capacity from the ground up. Therefore, the integration goals of regional digital technologies and public health services differ, and their empowering pathways and effects also vary significantly.^28,29 Second, some cities with high resource endowments possess fossil resources such as coal and oil, which create path dependence during development and result in a relatively single industrial structure. After resource depletion, governments invest funds and labour in resource restructuring, leaving no additional funds for digital upgrading and artificial intelligence applications. Cities with low resource endowments lack traditional resource advantages, but their industrial structures are more diversified, giving them a stronger willingness to adopt and innovate with digital technologies. Industrial diversity also generates economic vitality, enabling governments to increase investment in public health services. ³⁰ Finally, China’s northern and southern regions differ significantly in resource endowments, marketisation levels, and business environments. The empowering effects of digital upgrading and artificial intelligence on the sustainable development of public health services may therefore differ between the northern and southern regions. Based on this, this study proposes the second hypothesis:

• H2: The enabling effects of artificial intelligence and digital upgrading on the sustainable development of public health services exhibit significant heterogeneity.

The emergence of disruptive environmental technologies is often accompanied by the application and innovation of AI tools. ³¹ At this stage, green invention patents such as wastewater treatment technologies, new clean energy technologies, and high-performance air purification materials have become effective tools for the sustainable development of the industry. ³² The application of AI-based advanced environmental protection technologies is conducive to addressing environmental problems, preventing environmentally induced diseases, relieving pressure on the medical system, and ensuring the healthy development of public health services. ³³ By contrast, digital upgrading can address substantive problems in public health services, and the green practical technologies it generates can optimise and improve service procedures. For example, the construction of green hospitals and smart hospitals is a prominent manifestation of digital upgrading in the medical service field. Intelligent lighting and air-conditioning control systems based on Internet of Things technology can help medical institutions conserve resources and save electricity costs ³⁴ ; medical waste treatment devices derived from digital applications can prevent cross-infection by viruses and bacteria. Environmentally friendly public health service processes help reduce hospitals’ own energy consumption and pollution, while the operating costs saved can be reinvested in medical services, thereby improving the operational efficiency and sustainability of the entire health system.^35,36 In summary, the driving effects of digital upgrading and artificial intelligence on the sustainable development of public health services are mainly reflected in shaping a macro-level healthy environment through green technological innovation and improving the efficiency of the internal system through green practical technological innovation pathways. Therefore, this study proposes the third hypothesis:

• H3: The enabling effect of artificial intelligence and digital upgrading on the sustainable development of public health services is primarily realised through the mechanism of green technological innovation.

3. Model specification and data description

3.1. Model specification

3.1.1. Double machine learning

In examining the impact of artificial intelligence and digital upgrading on the sustainable development of public health services, regression results are often confounded by nonlinear correlations among variables. The use of the double machine learning model not only leverages its strong fitting capability to mitigate estimation errors caused by high-dimensional confounding variables but also overcomes regression biases arising from overfitting in machine learning models through techniques such as sample splitting. Therefore, this study incorporates the double machine learning model in the empirical analysis, and its mathematical formulation is presented as follows:

$$Y_{it}={\theta }_{0}Main{X}_{it}+g\left(X_{it}\right)+{\epsilon }_{it}$$ (1)

$$Main{X}_{it}=m\left(X_{it}\right)+{\mu }_{it}$$ (2)

In the equation above, i denotes the city-level sample in Equation (1), t is the sample period. $Y_{it}$ and $Main{X}_{it}$ represent the indices for sustainable development of public health services and artificial intelligence and digital upgrading, respectively. Where ${\theta }_0$ represents the regression coefficient for artificial intelligence and digital upgrading, which reflects the degree of influence between variables. $X_{it}$ denotes the control variables selected for this study, namely urbanization rate, human capital level, science and technology expenditure level, education development level, healthcare expenditure level, and fiscal decentralization level. The value of $g\left(X_{it}\right)$ quantifies the extent to which the above variables affect the sustainable development of public health services. In Equation (2), $m\left(X_{it}\right)$ is the regression function of the core explanatory variable on a set of high-dimensional control variables, whose form should be measured through machine learning algorithms. ${\mu }_{it}$ is the random error term.

3.1.2. Kernel density estimation (KDE)

This study employs the kernel density estimation model as a preliminary econometric tool for spatial analysis, primarily because, as a non-parametric method, it offers the distinct advantage of not relying on prior assumptions. This characteristic prevents subjective bias from influencing regression results. It enhances the data-driven nature of the analysis, thereby providing stronger practical relevance and theoretical support in revealing the intrinsic patterns of the data. Accordingly, the model is adopted for the empirical analysis, and the mathematical formulation of the kernel density estimation is presented as follows:

$${\widehat{f}}_h\left(x\right)=\frac{1}{nh}\sum _{i=1}^{n}K\left(\frac{x-x_i}{h}\right)$$ (3)

In equation (3), x denotes the observation point, $x_i$ is a sample point, and the value of $x_i$ must be within the range 0 to N; $K\left(\left(x-x_i\right)/h\right)$ denotes the kernel function, whose result must satisfy the properties of non-negativity, symmetry, and normality (i.e., its integral equals 1). h represents the variance of the Gaussian kernel, i.e., the bandwidth, whose value has a decisive impact on data fitting. When the bandwidth of the kernel function remains greater than zero and is not set too small, the neighbourhood will contain a sufficient number of fitting points.

3.1.3. Standard deviation ellipse

Approaching the analysis from a spatial perspective enhances the persuasiveness of the regression results generated by the double machine learning model. This is because the implementation effects of policies aimed at promoting the sustainable development of public health services do not exhibit spatial uniformity. The application of the Standard Deviation Ellipse model helps identify spatial clustering and directional characteristics of the variables, thereby providing preliminary visual evidence of spatial influence effects and heterogeneity. Accordingly, this study employs the model for the preliminary analysis, and an overview of its computational procedure is presented as follows:

Mean Centre $\left(\overline{X},\overline{Y}\right)$ :

$$\overline{X}=\sum _{i=1}^m{q}_i{x}_i/{\sum _{i=1}^{m}q}_i;\overline{Y}=\sum _{i=1}^m{q}_i{y}_i\hspace{0.17em}/\hspace{0.17em}\sum _{i=1}^m{q}_i$$ (4)

Azimuth $\theta$ :

$$\tan \hspace{0.17em}\theta \left(\left(\sum _{i=1}^m{q}_i^2{\stackrel{\sim }{x}}_i^2-\sum _{i=1}^m{q}_i^2{\stackrel{\sim }{y}}_i^2\right)+\sqrt{{\left(\sum _{i=1}^m{q}_i^2{\stackrel{\sim }{x}}_i^2-\sum _{i=1}^m{q}_i^2{\stackrel{\sim }{y}}_i^2\right)}^2+4\sum _{i=1}^m{q}_i^2{\stackrel{\sim }{x}}_i{\stackrel{\sim }{y}}_i}\right)/\sum _{i=1}^{m}2q_i^2{\stackrel{\sim }{x}}_i{\stackrel{\sim }{y}}_i$$ (5)

where, ${\stackrel{\sim }{x}}_i$ , ${\stackrel{\sim }{y}}_i$ denote the coordinate deviations of each province or city from the mean centre, respectively.

$${\stackrel{\sim }{x}}_i=x_i-\overline{X},{\stackrel{\sim }{y}}_i=y_i-\overline{X}$$ (6)

The standard deviations of the X and Y axes are ${\sigma }_x$ , ${\sigma }_y$ :

$${\sigma }_x=\sqrt{\left(2\sum _{i=1}^m{\left(q_i{\stackrel{\sim }{x}}_i\hspace{0.17em}\cos \hspace{0.17em}\theta -q_i{\stackrel{\sim }{y}}_i\hspace{0.17em}\sin \hspace{0.17em}\theta \right)}^2\right)/\sum _{i=1}^m{q}_i^2}$$ (7)

$${\sigma }_y=\sqrt{\left(2\sum _{i=1}^m{\left(q_i{\stackrel{\sim }{x}}_i\hspace{0.17em}\sin \hspace{0.17em}\theta -q_i{\stackrel{\sim }{y}}_i\hspace{0.17em}\cos \hspace{0.17em}\theta \right)}^2\right)/\sum _{i=1}^m{q}_i^2}$$ (8)

Ellipse area S:

$$S=\pi {\sigma }_x{\sigma }_y$$ (9)

In equations (4)-(9), $\left(\overline{X},\overline{Y}\right)$ denotes the latitude and longitude coordinates of each city, $q_i$ represents the index value of the research object corresponding to each city.

3.2. Data description

3.2.1. Variable description

3.2.1.1. Dependent variable: Sustainable development of public health services (PHSSD)

The Outline of the Healthy China 2030 Plan, issued by the State Council in October 2016, states that by 2030, the institutional system for promoting health for all will be further improved, development in the health sector will become more coordinated, healthy lifestyles will be popularised, the quality of health services and the level of health security will continue to improve, the health industry will develop prosperously, health equity will be basically achieved, and major health indicators will reach the level of high-income countries. ³⁷ Therefore, this study argues that the sustainable development of public health services is influenced by multiple factors, and that interactions among systems also show complex and dynamic trends, while the interaction patterns among elements are often characterised by multidimensional and multilevel features. In constructing the evaluation indicator system, this study follows the principles of systematicity, scientific, and feasibility. Starting from five dimensions, namely public health resources, public health services, public health security, health care services, and medical and health expenditure, 19 tertiary indicators are selected to construct a comprehensive evaluation indicator system for the sustainable development of public health services, ³⁸ as shown in Table 1. Finally, the TOPSIS comprehensive evaluation method is used to measure the sustainable development of public health services index.

Table 1.

Sustainable development of public health services evaluation indicator system.

Primary indicator	Secondary indicator	Tertiary indicator	Indicator attribute	Weight
Sustainable development of public health services	Public Health Resources	Number of Medical Institutions	+	0.062
		Number of Beds in Medical Institutions	+	0.051
		Health Technical Personnel	+	0.049
		Licensed (Assistant) Physicians	+	0.052
		Registered Nurses	+	0.050
	Public Health Services	Medical and Health InstitutionsNumber of Medical Visits (person-times)	+	0.063
		Medical and Health InstitutionsNumber of Hospital Admissions (persons)	+	0.057
		Average Length of Hospital Stay	-	0.002
		Hospital Bed Occupancy Rate	+	0.008
	Public Health Security	Number of Urban and Rural Residents Enrolled in Basic Medical Insurance	+	0.087
		Basic Medical Insurance Fund Income for Urban Employees	+	0.086
		Number of Participants in Maternity Insurance at Year-end	+	0.077
	Health Care Services	Number of Health Education Institutions	+	0.119
		Premarital Medical Examination Rate	+	0.032
		Health Management of Children Under Seven	+	0.007
		Share of Medical and Health Care Expenditure in Consumption Expenditure	+	0.023
	Medical and Health Expenditure	Total Health Expenditure	+	0.060
		Total Assets of Medical and Health Institutions	+	0.056
		Total Revenue of Medical and Health Institutions	+	0.061

3.2.1.2. Core explanatory variable: Digital upgrading

In defining digital upgrading, this study takes information resource economy theory as its research basis and emphasises that digital information exists as a resource in economic activities and has economic value. In the process of digital upgrading, enterprises effectively utilise and develop data information through data mining, which can be used to optimise decision-making and reduce operating costs, thereby improving operational efficiency and promoting sustainable and high-quality socioeconomic development. Therefore, this study takes the Digital Development Index released by the National Bureau of Statistics as its foundation and refers to the research of Zhou, Qi ³⁹ Four dimensions, namely digital facilities, digital users, digital industries, and digital applications, are selected, and 14 tertiary indicators are used to construct a comprehensive evaluation indicator system for digital construction, as shown in Table 2. Meanwhile, the TOPSIS comprehensive evaluation method is used to measure the digital upgrading index.

Table 2.

Digital upgrading indicator system.

Primary indicator	Secondary indicator	Tertiary indicator	Indicator attribute	Weight
Digital upgrading	Digital facilities	Length of Optical Cables per Square Kilometre	+	0.043
		Number of IPv4 Addresses	+	0.122
		Number of Domain Names	+	0.126
		Internet Broadband Access Ports	+	0.055
	Digital users	Average Population Served per Business Outlet	-	0.005
		Number of Mobile Internet Users	+	0.050
		Number of Internet Broadband Access Users	+	0.061
		Telephone (Including Mobile Phone) Penetration Rate	+	0.026
	Digital industries	Total Volume of Telecommunications Services (100 million yuan)	+	0.107
		Revenue from Software Business (Including Information Technology Services)	+	0.189
		E-commerce Sales	+	0.138
	Digital applications	Number of Computers Used per 100 Employees and Number of Websites Owned per 100 Enterprises	+	0.036
		Number of Websites Owned per 100 Enterprises	+	0.013
		Share of Enterprises Engaged in E-commerce Transactions	+	0.030

3.2.1.3. Core explanatory variable: Artificial intelligence

Existing literature usually measures the level of artificial intelligence development from the perspectives of macro industries and micro enterprises. Macro-level measurement is based on input-output models, while micro-level measurement mainly relies on questionnaires and text analysis based on recruitment information. ⁴⁰ However, because patents, publications, and data usually have a publication lag of one to two years, the measurement results obtained through these methods may differ from actual conditions. Therefore, to measure the application level of artificial intelligence in the field of public health services, this study uses listed companies related to public health services in each region as the basic sample, where the definition of public health service enterprises refers to the study by Du and Lin ⁴¹ and enterprises falling within the category of “Health and Social Work” in the Guidelines for the Industry Classification of Listed Companies (2012 Revision) are defined as enterprises in the field of public health services enterprises. It also covers industries such as “pharmaceutical manufacturing”、“wholesale of pharmaceuticals and medical devices”、“specialised retail of pharmaceuticals and medical devices”、“warehousing of Chinese medicinal materials”、“operating leasing of medical equipment”、“medical research and experimental development and sports and health services”. These enterprises are mainly engaged in public health service-related products and production, wholesale, and social services. Therefore, based on the investment status of the above enterprises in artificial intelligence software and hardware, this study calculates the amount of artificial intelligence investment in each Chinese province from 2012 to 2024, and finally uses the logarithm of artificial intelligence investment to measure the level of artificial intelligence development in public health services across regions.

3.2.1.4. Mechanism variable: Green technological innovation

Sustainable development of public health services necessitates not only digital empowerment but also a significant emphasis on technology innovation and environmental sustainability. Traditional mechanism-focused research prioritizes resource distribution and efficiency. In contrast, the paradigm of green technology innovation offers a more comprehensive viewpoint, encompassing digital, environmental, and public health aspects. Innovation thereby serves as the central force driving reforms across these sectors and indicators. Consequently, this study employs green technology innovation as the mediating variable. This characteristic is assessed from three unique viewpoints: green technological innovation, green patent-based technological innovation, and green utility model technological innovation. This method creates a thorough, multifaceted analytical framework while simultaneously achieving seamless integration of the depth and breadth of green technology innovation. The indicators are measured as follows: green technological innovation is proxied by the number of green patents granted per 10,000 inhabitants. ⁴² Green invention-based technological innovation is proxied by the number of green invention patent applications per 10,000 inhabitants. In contrast, green utility-model innovation is measured using the number of green utility-model patent applications per 10,000 inhabitants. ⁴³

3.2.1.5. Control variables

To disentangle the net effects between the core explanatory variable and the dependent variable, and to investigate in greater depth the transmission mechanism of green technological innovation, this study draws on relevant literature and incorporates the following control variables^44–46: Urbanization rate: Measured by the proportion of urban population to the total population; Human capital level: Measured by the number of college students per 100 people; Science and technology expenditure level: measured by the proportion of government spending on science and technology to total general public budget expenditure; Educational development level: measured by the proportion of fiscal expenditure on education in the general public budget; Healthcare expenditure level: Calculated using the formula for the proportion of healthcare fiscal expenditure to general public budget expenditure; Fiscal decentralisation level: measured by the ratio of general fiscal budget expenditure to general fiscal revenue.

3.2.2. Data source

After removing provinces with significant data gaps, this study chose 30 provinces from 2012 to 2024 as its research sample. Data for the dependent variables were mainly obtained from the China Health Statistics Yearbook and the China Statistical Yearbook. The primary sources of data on digital upgrading among the basic explanatory variables include the China Statistical Yearbook, provincial statistical yearbooks, city statistical bulletins on national economic and social growth, and the China Economic Information Network statistics database. The investment data on artificial intelligence software and hardware of public health service enterprises are mainly obtained from the CSMAR database. The control variables are predominantly derived from data obtained from the China Health Statistical Yearbook, the China Statistical Yearbook, the China City Statistical Yearbook, and the China Statistical Yearbook on Science and Technology. Table 3 presents the descriptive statistics of the variables. The statistical results facilitate a preliminary comprehension of their real-world developmental trends. The significant disparity between the maximum and minimum values indicates substantial variation among the indicators. No anomalies are identified in the remaining data, suggesting their appropriateness for subsequent empirical research.

Table 3.

Descriptive statistical analysis.

Variable type	Variable	Symbol	Obs	Mean	SD	Min	Max
Dependent variable	Sustainable Development of Public Health Services	PHSSD	390	0.289	0.134	0.079	0.840
Core explanatory variable	Artificial Intelligence	AI	390	2.917	1.139	0.000	6.412
	Digital upgrading	DIG	390	0.202	0.110	0.063	0.689
Mechanism Variable	Green technological innovation	GTI	390	7.767	1.335	3.466	10.735
	Green invention-based technological innovation	GPTI	390	6.212	1.456	1.609	9.747
	Green utility-model technological innovation	GUTI	390	7.498	1.312	3.135	10.572
Control Variables	Urbanisation rate	URB	390	0.612	0.118	0.354	0.942
	Human capital level	HUM	390	0.022	0.006	0.009	0.044
	Science and technology expenditure level	TEC	390	0.163	0.028	0.095	0.226
	Educational development level	EDU	390	0.023	0.016	0.005	0.071
	Healthcare expenditure level	HEA	390	0.082	0.015	0.044	0.139
	Fiscal decentralisation level	FIN	390	2.418	1.042	1.074	6.604

4. Empirical analysis

4.1. Spatiotemporal evolution characteristics of sustainable development of public health services

4.1.1. Spatial distribution characteristics of sustainable development of public health services

This research initially analyses the regional distribution patterns to elucidate the current status of sustainable development in China’s public health services. Utilizing ArcGIS software to visualize sample data, we ultimately generated the spatial distribution pattern map depicted in Figure 1. This map uses colour coding to visually illustrate the spatial distribution of sustainable development and its growth trends in China’s public health services. Figure 1 illustrates that China’s overall sustainable development in public health services was comparatively low in 2012, with a regional distribution pattern with elevated values in the east and diminished values in the west. Qinghai registered the lowest readings in the country, ranging from 0.093 to 0.124. The majority of provinces fell within the 0.124–0.185 range, although eastern coastal areas such as Zhejiang, Jiangsu, and Shandong exhibited somewhat elevated values, ranging from 0.225 to 0.362. The spatial distribution patterns identified in 2016, 2020, and 2024 consistently demonstrate an east-high, west-low tendency, with a notable strengthening of this pattern. This is because, by 2024, the developmental levels of eastern coastal regions like Zhejiang had increased, whilst those of northwestern and northeastern regions, including Jilin, Gansu, and Shaanxi, had diminished. Regarding resource allocation, eastern coastal areas like Zhejiang have strong economic foundations, with per capita spending in public health care significantly surpassing that of western regions. This allows for greater tolerance for trial and error while using internet technology to improve public health services. The driving force for government-led public health service improvements remains insufficient, owing to the inherently weaker economic foundation in western regions and the increasingly severe population outflow in recent years. These factors are the key drivers of the east-high, west-low spatial distribution of public health services in China.

Figure 1.

Spatial distribution pattern in the sustainable development of public health services.

4.1.2. Spatiotemporal agglomeration characteristics in the sustainable development of public health services

This study extends the temporal analysis of sustainable development in China’s public health services by applying kernel density estimation to the data. The results are displayed in Figure 2. Figure 2 shows the kernel density curves for the entire territory and for eastern, western, and central China. The centre of the kernel density curve shifts to the right over time across the entire region, accompanied by a broadening of the peak width and the formation of a trailing phenomenon. This suggests that the overall level of sustainable development in China’s public health care is improving, while regional discrepancies persist and are gradually widening over time. Observing density curves for various economic zones demonstrates that the nuclear density curve exhibits characteristics similar to those of the other curves, such as a rightward shift in the centre, a trailing tail, and a broader peak. However, the kernel density curve of the western region briefly shifts leftward but then shows a rightward trend. This indicates that the overall level of sustainable development of public health services in the western region continues to improve, but internal differences are significant, development gaps among provinces within the region continue to widen, and the development level shows a trend of polarisation. To prevent polarisation, the Western area should invest more in policy support and digital applications.

Figure 2.

Kernel density estimation of the sustainable development of public health services.

4.1.3. Spatial evolution characteristics of the sustainable development of public health services

Figure 3 is a standard deviation ellipse diagram created by visualising public health service sustainability data with GIS software. The geographical evolution characteristics of China’s public health care sustainability can be determined by examining the movement trend of the ellipse’s mean centre of gravity. Figure 3 indicates that the mean centre of gravity of the standard deviation ellipse has shifted from the northwest to the southeast, specifically moving from Henan towards Hubei. This suggests that the focal regions for the sustainable development of China’s public health services are shifting towards the southeast. This is due to Hubei’s status as a key transit hub in central China, which provides a solid foundation for medical development. With the help of central government programs, its primary healthcare network has grown more tightly integrated, raising overall public health service standards. Meanwhile, Henan Province, a populous region, has struggled with population loss in recent years. As a result, when public health service resources exceed their current allotment, they are often transferred to Hubei, shaping the migratory pattern of the hotspot area within the standard deviation ellipse. Table 4 shows that the centroid longitude is growing by 8.88 km while the centroid latitude is decreasing by 26.64 km. China’s sustainable public health development hotspot region has migrated 8.88 km eastward and 26.64 km southward, resulting in a decrease of 220497.35 km² in the circular area. This is a clear tendency towards concentration in the long-term development of China’s public health services, with dispersion levels substantially reduced.

Figure 3.

Standard deviation ellipse of sustainable development of public health services.

Table 4.

Standard deviation ellipse parameters for sustainable development of public health services.

Year	Centroid longitude	Centroid latitude	Semi-major axis (km)	Semi-minor axis (km)	Area (km2)	Azimuth (°)
2012	112.84	33.05	1297.00	1009.31	4112598.83	70.04
2024	112.92	32.81	1274.58	972.00	3892101.48	68.30

4.2. The impact of artificial intelligence and digital upgrading on the sustainable development of public health services

4.2.1. The direct impact of artificial intelligence and digital upgrading on the sustainable development of public health services

This work utilised machine learning models for regression analysis, initially introducing data sets without control factors, then including both first- and second-order control variables. This finally produced the direct effect results displayed in Table 5. Firstly, when artificial intelligence is treated as the primary explanatory variable, the regression results yield coefficients of 0.011, 0.008, and 0.014, all of which are statistically significant at the 5% level. This suggests that advances in artificial intelligence can significantly drive the sustainable development of public health services, with each unit increase in artificial intelligence resulting in a 0.014% increase in sustainable public health development. Secondly, when digital upgrading is utilised as the primary explanatory variable, the estimated coefficients are 0.742, 0.694, and 0.800, all of which remain statistically positive at the 1% level. This suggests that digital upgrading has a substantial positive effect on the long-term development of public health care. A one-unit rise in digital upgrading results in a 0.800% increase in the level of sustainable public health service development. Advances in artificial intelligence and digital upgrading have a clear, favourable impact on the long-term development of public health care. This is consistent with the findings reported in Hypothesis 1 of this study, indicating empirical support for the hypothesis.

Table 5.

Direct impact effects.

Variables	Artificial intelligence			Digital upgrading
	(1)	(2)	(3)	(4)	(5)	(6)
AI or DIG	0.011***	0.008**	0.014***	0.742***	0.694***	0.800***
	(2.74)	(2.49)	(4.16)	(10.91)	(9.72)	(10.49)
Constant term	-0.037***	-0.025***	-0.020***	0.001	-0.008*	-0.008**
	(-8.64)	(-7.24)	(-4.83)	(0.22)	(-1.90)	(-2.04)
First-order terms of control variables	No	Yes	Yes	No	Yes	Yes
Quadratic terms of control variables	No	No	Yes	No	No	Yes
Individual fixed	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes	Yes
Sample size	390	390	390	390	390	390

Note. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively; the values in parentheses represent the z-statistics.

4.2.2. Robustness test

This research evaluates the original conclusions from the following dimensions to ascertain their robustness. Given that outlier values in the data on digital technology and artificial intelligence application levels in developed cities may affect the final empirical results, possibly rendering them unrepresentative, this paper conducted an additional regression analysis on the remaining provincial data after omitting figures for Beijing Municipality, Shanghai Municipality, and Guangdong Province. The results are respectively shown in Table 6 of Panel A and Panel B column (1). If the causal effects between variables are authentic, the model’s applicability should not be limited to particular sample periods. Therefore, this research modified the sample period to 2013–2024 and re-executed the regression analysis. The results are respectively shown in Table 6 of Panel A and Panel B column (2). To assess the validity of the original conclusions under varying partitioning criteria, the training and test sets were also split at a 1:3 ratio. The regression analysis results are shown in Table 6 of Panel A and Panel B column (3). Table 6 in Panel A and Panel B column (4) shows the regression results after the model is re-implemented. Following the reassessment of the sustainability of public health services utilising the entropy technique in this study. To eliminate any intrinsic correlation between sample selection and model specification that could introduce variability into regression outcomes, the original support vector machine was replaced with a random forest to reassess the initial findings. The results are shown in Table 6 Panel A and Panel B, column (5). Based on the regression results presented in the table, the coefficients for artificial intelligence and digital upgrading remain significantly positive at the 5% level across various robustness tests. This indicates that, even after conducting robustness checks, both digital upgrading and artificial intelligence continue to exert a significant promoting effect on the sustainable development of public health services. These findings are consistent with the original results, thereby confirming the representativeness and reliability of the initial conclusion.

Table 6.

Robustness tests.

Panel A: Robustness test of artificial intelligence
Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Core explanatory variables	0.010***	0.012***	0.016***	0.019***	0.010**	0.010**	0.012**	0.014***
	(2.69)	(2.80)	(4.59)	(4.77)	(2.11)	(2.55)	(2.37)	(4.16)
Constant term	-0.015***	-0.018***	-0.019***	-0.015***	-0.006	-0.021***	-0.102	-0.020***
	(-3.34)	(-3.99)	(-4.25)	(-3.26)	(-1.41)	(-5.30)	(-1.18)	(-4.83)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Individual fixed	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Sample size	351	360	390	390	390	360	390	390

Note. ***, **, and * denote significance levels of 1%, 5%, and 10% respectively; the values in brackets represent z-values.

Meanwhile, considering that the analysis may not fully account for lagged effects, serial correlation, or province-specific trends, further robustness tests are conducted. The lagged term of the explained variable is introduced into the double machine learning model, and the results are shown in Table 6 Panel A and Panel B, column (6); To exclude the interference of serial correlation in the model, generalized least squares is used to replace the original double machine learning model to further verify the robustness of the model, and the results are shown in Table 6 Panel A and Panel B, column (7); finally, considering that each province may have specific trends, the model further adds an “individual x time trend term” on the basis of individual and time fixed effects, and the results are shown in Table 6 Panel A and Panel B, column (8). The further robustness test results show that the regression coefficients of digital upgrading and artificial intelligence are both significant at the 5% level, indicating that the regression results remain robust and valid.

4.2.3. Endogeneity test

Although the above robustness tests have verified the reliability of the conclusion that artificial intelligence and digital upgrading can significantly empower the sustainable development of public health services, the model may still suffer from endogeneity problems caused by potential reverse causality and omitted variables. Specifically, although artificial intelligence and digital upgrading promote the sustainable development of public health services, improvements in the sustainable development level of public health services may also increase the industry-wide demand for intelligent digital technologies, thereby promoting the application and iteration of artificial intelligence-related technologies in the public health field. This may lead to estimation bias caused by reverse causality. To address this endogeneity problem, this study selects the terrain relief degree of each province, the number of post offices in 1984, and the number of fixed-line telephones in 1984, and multiplies these three variables by the number of Internet users in the previous year as instrumental variables for artificial intelligence and digital upgrading, followed by two-stage least squares estimation. On the one hand, terrain relief affects the construction cost and coverage quality of regional network communication infrastructure. Regions with lower terrain relief are more likely to achieve the popularisation of digital infrastructure, satisfying the relevance requirement for instrumental variables. Meanwhile, regional terrain is a natural geographical condition and does not directly affect the current sustainable development level of public health services, thereby satisfying the exogeneity requirement. On the other hand, the historical level of postal and telecommunication infrastructure affects the early diffusion path of Internet technology and further imposes long-term constraints on subsequent regional artificial intelligence and digital industry development, satisfying the relevance requirement. The numbers of post offices and fixed-line telephones in 1984 are historical economic data and do not directly affect the current development quality of public health services, thereby also satisfying the exogeneity requirement. Therefore, the results of the endogeneity test are shown in Table 7. After controlling for endogeneity factors, the core explanatory variables of artificial intelligence and digital upgrading still show significant positive effects, which is consistent with the previous regression results. This indicates that after addressing endogeneity problems, the core conclusion that artificial intelligence and digital upgrading promote the sustainable development of public health services still holds.

Table 7.

Endogeneity test results.

Panel A: Endogeneity test of artificial intelligence test
Variables	(1)	(2)	(3)
Core explanatory variables	0.400***	0.380***	0.394***
	(2.74)	(4.83)	(4.75)
Constant term	0.020	0.018	0.020
	(1.09)	(1.20)	(1.30)
Control variables	Yes	Yes	Yes
Individual fixed	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes
Sample size	390	390	390

4.3. Further empirical testing

4.3.1. Heterogeneity analysis

This study categorises the sample data to assess if diverse resource endowments and economic divides have distinct effects on the first findings. The sample is categorised into eastern, central, and western economic regions, and into southern and northern geographical locations. Furthermore, based on the degree of regional resource endowment, the sample is categorised into regions with high and low artificial intelligence resource endowments, as well as regions with high and low digital upgrading resource endowments. Regression analyses are performed for each subsample, and the results of the heterogeneity test are displayed in Table 8.

Table 8.

Results of the heterogeneity analysis.

Panel A: Heterogeneity analysis among the three major economic regions
Variables	Artificial intelligence			Digital upgrading
	Eastern region	Central region	Western region	Eastern region	Central region	Western region
	PHSSD	PHSSD	PHSSD	PHSSD	PHSSD	PHSSD
Core explanatory variables	0.027**	0.008*	0.013***	0.661***	1.204***	1.114**
	(2.48)	(1.86)	(2.63)	(6.92)	(23.89)	(13.90)
Constant term	-0.023***	-0.013*	-0.020***	-0.011	0.006**	-0.001
	(-3.32)	(-1.74)	(-2.89)	(-1.58)	(1.99)	(-0.07)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Urban Fixed	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes	Yes
Sample size	143	104	143	143	104	143

Note. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively; the values in parentheses represent z-statistics.

The heterogeneity test data for economic regions indicated that the regression coefficients for artificial intelligence’s impact on the sustainable development of public health services were 0.028, 0.008, and 0.013 for the eastern, central, and western regions, respectively, all significant at the 10% level. This suggests that artificial intelligence has a more significant enabling function in the sustainable growth of public health services in the eastern region. In contrast, its enabling impacts in the central and western regions are similar. This principally arises from the high levels of informatisation and digitisation in eastern medical facilities, which facilitate the collection and accumulation of comprehensive, reliable medical data. This data is crucial for training artificial intelligence models. The concentration of digital technology talent in eastern regions is significantly greater than in central and western areas. The heterogeneity test results for digital upgrading across economic regions are 0.661, 1.204 and 1.114, all of which meet the significance criterion. The beneficial impact of digital enhancement on the sustained advancement of public health services is especially evident in the central area. The central area prioritises optimising resource allocation and enhancing process efficiency, given its comparatively extensive medical resources. Thus, the marginal gains derived from digital upgrading are more significant than those in the eastern area, which emphasises innovation, and the western region, which is currently pursuing fundamental enhancement through digital efforts.

In resource-endowed heterogeneity testing, the enabling impacts of artificial intelligence and digital upgrading on the long-term development of public health services are more evident in low-resource-endowed regions than in high-resource-endowed regions. This pattern can be attributed to the fact that regions with high resource endowments typically have stronger economic foundations and already perform relatively well in public health service provision, leaving little room for further improvements driven by digital upgrading and artificial intelligence. Low-resource regions, on the other hand, have larger gaps in both digital technology application and public health care capacity, indicating greater opportunities for progress. As a result, the marginal contributions from digital upgrading and artificial intelligence increase significantly in these places.

Notably, in the heterogeneity analysis between the southern and northern regions, artificial intelligence and digital upgrading have a more significant promoting effect on the sustainability of public health services in the southern region than in the northern region. This is mainly because the southern region has a stronger digital industry foundation, richer application scenarios for artificial intelligence technologies, and higher market acceptance and application maturity of digital technologies, enabling it to transform technological advantages more quickly into improvements in the quality and efficiency of public health services. By contrast, the digital transformation of traditional public health service systems in the northern region is still being deepened, and the incremental effects of technological empowerment have not yet been fully released; therefore, the overall empowering effect has not yet reached the level observed in the southern region.

The heterogeneity results across different groups clearly show that the promoting effects of artificial intelligence and digital upgrading on the sustainable development of public health services are constrained by existing regional development foundations, resource endowments, and geographical location conditions, and that the empowering effects differ significantly across different types of regions. In summary, the findings confirm the validity of Hypothesis 2.

4.3.2. Mechanism test of green technological innovation

To clarify how green technological innovation fits into the pathway connecting artificial intelligence and digital upgrading to the long-term development of public health services, this study includes indicators of green technological innovation, green patent innovation, and green utility-model innovation in the regression analysis. The complete statistical results of the mechanism examination are shown in Table 9. Analysis of the transmission mechanism of artificial intelligence reveals regression results of 0.173, 0.198, and 0.168 for the aforementioned indicators, all significant at the 1% level. Under these conditions, the regression coefficients for these indicators regarding the sustainable development of public health services are 0.061, 0.052, and 0.061, respectively, likewise significant at the 1% level. This indicates that artificial intelligence exerts a significant positive influence on the aforementioned indicators. Within the same research framework, these indicators in turn exert a significant positive influence on the sustainable development of public health services. As a result, it is clear that in the process of artificial intelligence empowering the sustainable development of public health services, green technological innovation, green patent technological innovation, and green utility model technological innovation are critical mechanisms and transmission pathways. This is because green invention patents reduce pollution at the source and reduce disease burdens by developing fundamental environmental technologies. In contrast, green utility model patents improve the efficiency and sustainability of healthcare systems’ internal operations by optimising energy consumption and pollutant levels.

Table 9.

Mechanism test results for green technological innovation.

Panel A: Mechanism test results of artificial intelligence for green technological innovation
Variables	(1)	(2)	(3)	(4)	(5)	(6)
	GTI	PHSSD	GPTI	PHSSD	GUTI	PHSSD
Artificial Intelligence	0.173***	​	0.198***	​	0.168***	​
	(3.19)	​	(3.50)	​	(3.11)	​
Green Technological Innovation	​	0.061***	​	0.052***	​	0.061***
	​	(14.13)	​	(13.78)	​	(13.90)
Constant term	-0.109**	-0.014***	-0.052	-0.018***	-0.103**	-0.014***
	(-2.32)	(-4.43)	(-1.01)	(-5.48)	(-2.21)	(-4.57)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Urban Fixed	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes	Yes
Sample size	390	390	390	390	390	390

Note. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively; the values in parentheses represent z-statistics.

In the mechanism tests for digital upgrading and green technological innovation, the core explanatory variable shows regression coefficients of 9.109, 10.501, and 8.439 for green technological innovation, green patent innovation, and green utility-model innovation, respectively, all of which are significant at the 1% level. The regression coefficients of these innovation indicators for the sustainable development of public health services are likewise significantly positive at the 1% level. These results indicate that digital upgrading significantly enhances the development of green technological innovation, green patent innovation, and green utility-model innovation, and that, through this pathway, it further promotes the sustainable development performance of public health services. Digital technologies offer robust tools for the essential development of sustainable technologies. The advancement of novel eco-friendly materials, efficient pollution-control technology, and clean energy solutions is clearly evidenced by the increase in green invention patents. Digitalisation directly propels application-focused green innovation. Solutions developed on digital platforms—such as advanced energy management systems and sophisticated medical waste tracking and treatment apparatus—are generally represented as green utility model patents. When implemented in healthcare facilities, these technologies can significantly reduce energy and water use while facilitating the safe, environmentally friendly disposal of medical waste, thereby promoting the establishment of sustainable hospitals. This not only mitigates the healthcare system’s adverse environmental effects but also enables the reinvestment of operational cost savings into medical services, thereby improving the system’s operational efficiency, resilience, and sustainability. In summary, Hypothesis 3 has been validated.

5. Discussion

Using panel data from 30 Chinese provinces from 2012 to 2024 as the research sample, this study employs a double machine learning model to examine the effects and action pathways of digital upgrading and artificial intelligence on the sustainable development of public health services. The findings are as follows: First, digital upgrading and artificial intelligence can significantly promote the sustainable development of public health services. Ghanem, Moraleja ⁴⁷ found that the application of artificial intelligence in the field of public health is developing rapidly, and that artificial intelligence has the potential to improve the effectiveness, precision, decision-making capacity, and scalability of public health programmes. Iyamu, Haag ⁴⁸ argued that digital transformation can bridge the perceived disconnect between digital technologies and public health functions, thereby better supporting public health practitioners in integrating digital technologies into their work. Therefore, the conclusions proposed in the above studies support the first hypothesis of this study. Second, Chen, Ding ⁴⁹ the coordinated development of the digital economy and public health services is crucial for integrating the Digital China and Healthy China strategies and accelerating the modernisation of the public health system; however, significant differences exist across regions. Therefore, as stated in Hypothesis 2 of this study, in the process of promoting the sustainable development of regional public health services, digital upgrading and artificial intelligence are affected by regional resource endowments, geographical differences, and other characteristics, resulting in significant heterogeneity in their effects. Third, Zhang, Zhang ⁵⁰ confirmed the potential of artificial intelligence-oriented policies to promote sustainable industrial transformation. By promoting green technologies that reduce emissions and industrial pollution, these policies generate a dual return: advancing environmental sustainability and supporting public health. Therefore, the above conclusions are consistent with Hypothesis 3 of this study, namely that digital upgrading and artificial intelligence development promote green technological innovation, which in turn significantly promotes the sustainable development of regional public health services.

Based on the above analysis, although this study is relatively complete in terms of its theoretical framework and analysis, and its conclusions are consistent with existing research, it still has the following limitations. First, the research sample used in this study analyses the effects of digital upgrading and artificial intelligence on the sustainable development of public health services only at the macro level, and does not further examine their action pathways for public health institutions, front-line practitioners, and other actors at the micro level. Thus, the depth of the research can still be expanded. Second, when exploring the mechanism, this study focuses only on green technological innovation as a mediating channel. In fact, digitalisation and artificial intelligence may also affect public health services through multiple pathways, such as optimising resource allocation and improving response efficiency. This study has not yet comprehensively tested these potential pathways, and future research can further expand and improve upon these limitations. Third, regional heterogeneity is mainly examined through group comparisons, without further constructing more precise quantitative models to identify the marginal contributions of different sources of heterogeneity. As a result, the deeper driving factors behind the differences are not sufficiently explored, and more refined evidence for differentiated policy formulation cannot be provided. Therefore, future research will further expand the sample dimensions, conduct analyses from the macro, meso, and micro levels, systematically identify the multiple pathways through which digital upgrading and artificial intelligence affect the sustainable development of public health services, test the effects of different pathways one by one, and combine more refined quantitative methods to deeply explore the driving logic behind heterogeneity, thereby addressing the limitations of existing research and providing more solid and precise research support for the digital transformation of China’s public health system.

6. Conclusions and recommendations

6.1. Research conclusions

This study explores the effects and underlying mechanisms of artificial intelligence and digital upgrading on the long-term growth of public health services in China. This is accomplished by building a twofold machine learning model and using panel data from 30 Chinese provincial-level administrative units spanning 2012-2024. Regression analysis and data visualisation lead to the following findings from the study: (1) China’s sustainable development of public health services exhibits distinct spatiotemporal patterns. Specifically, the level of sustainable development shows a spatial distribution characterised by higher performance in the east and lower performance in the west. Over time, the overall level of sustainable development has continued to improve; however, regional disparities persist and are widening. The disparities are particularly pronounced in the western region, where signs of polarisation are emerging. In addition, the hotspots of sustainable public health service development display an evolutionary trend of shifting towards the southeast, accompanied by a clear pattern of spatial agglomeration. (2) The sustained growth of public health services in China is greatly aided by artificial intelligence and digital upgrading. In particular, a one-unit increase in digital upgrading results in a 0.014% gain in sustainable development, whereas a one-unit increase in artificial intelligence results in a 0.080% gain. (3) Across economic zones, resource endowments, and the north-south divide, the enabling effects of digital upgrading and artificial intelligence on the sustainable development of public health care in China vary significantly. While the benefits of digital upgrading are more noticeable in central China, artificial intelligence is more important for promoting the sustainable development of public health services in the eastern part of the country. Digital upgrading and artificial intelligence both offer greater facilitation benefits in low-resource endowment areas than in high-resource endowment areas. Digital upgrading and artificial intelligence are more significant in southern China than in northern China. (4) Digital upgrading and artificial intelligence have a significant impact on the development of green technological innovation, including green invention patents and utility model innovations. This method improves the sustainability performance of public health services.

6.2. Research recommendations

Based on the above findings, this study proposes the following recommendations:

• (1) Increasing grassroots deployment of digital and intelligent public health resources to address regional inequities and resource gaps. We will strategically deploy public health digital infrastructure resources, focusing on the risk of polarisation in western regions and the untapped potential of digital applications in resource-constrained places. This involves building low-cost intelligent public health monitoring systems for primary healthcare facilities, executing targeted talent development programs, and working with high-quality medical resources from eastern regions to provide digital remote collaboration support. Through balanced resource allocation, we will gradually minimise development gaps between eastern and western regions.

• (2) When implementing digital and intelligent tools, consider regional disparities and amplify diverse driving influences. Customise the use of digital and intelligent applications to align with regional developmental attributes: The eastern regions should enhance AI applications, focusing on critical areas such as intelligent epidemic early warning and AI-assisted primary-level diagnosis. The central regions should bolster digital infrastructure development by establishing provincial-level interoperability systems for public health data. The northern regions must rectify deficiencies in digital service coverage. The southern regions should broaden AI applications in the allocation of public health resources. This method fully leverages the region-specific advantages of digital and intelligent technologies.

• (3) To construct a practical transmission pathway linking digital intelligence, green innovation, and the sustainable development of public health services, and to fully unleash the effectiveness of this mechanism, digital tools should be utilised to support R&D in green medical technologies. Market actors should be encouraged to leverage digital platforms to tackle key low-carbon technologies, such as energy-efficient disinfection equipment and low-energy primary healthcare facilities. Green patents and utility-model technologies should be promoted for translation and application in the public health sector, with the outcomes of green technology adoption incorporated into the public health service evaluation system. This will ensure that the green-innovation dividends generated through digital upgrading continue to reinforce the sustainable development of public health services.

ORCID iD

Luxin Zhang https://orcid.org/0009-0009-9932-7698