Unlocking AI’s Potential in China’s Healthcare System: Overcoming Structural and Economic Barriers

Abstract

With the development of artificial intelligence (AI) in recent years, AI has become an important tool to improve medical efficiency and lower the costs. However, in China, structural and economic barriers exist for the adoption of AI technologies in hospitals, making the progress slower than expected. This paper aims to use ideas from governance and political economy to study if and how factors such as institutional structure, budget, and investment can affect medical applications of AI in China. The study is based on a semi-structured interview with the head of a medical technology company as well as an analysis of policy documents and market data. The findings show a number of possible explanations as to why the AI development in China is limited. For example, public hospital budgets, data barriers between institutions, uneven digital development, and uncertain returns on investment are the main limits. This paper also concludes that long-term funding plans, better data-sharing policies, and ethical supervision are in need to support meaningful growth in this field. In summary, this study provides a new view on the obstacles for the adoption of AI in a non-Western healthcare system and discusses what policy changes are required to address these issues.

Keywords: medical AI, governance, China healthcare, data governance, public hospitals, budget rigidity

Introduction

Background and Problem

The World Health Organization (WHO) estimates that the global population aged 60 or above will be doubled by 2050, as compared to 2030. The population aged 80 or above will be tripled, in which approximately two thirds will live in low- or middle-income countries¹. Such a major demographic shift will inevitably place high pressure on healthcare systems worldwide, particularly at a time when international conflicts and pandemics are not uncommon. This is a worrisome trend. As a response, some have suggested developing and integrating artificial intelligence (AI) into healthcare to help reduce the burden on healthcare systems, improve access to resources, and raise overall living standards. However, the implementation in this field remains difficult.

Without doubt, with the recent developments in technology, AI has become an important tool that can reshape medical diagnosis, treatment, and government decision-making and regulation². Around the world, AI is now used in areas such as medical-image recognition, disease-prediction models, remote monitoring and surgery, and digital health management. These technologies are evolving quickly. That being said, technical progress alone does not guarantee that AI will be widely adopted in healthcare systems, and thus the benefits to mass population are still questionable. Governance structures, economic incentives, and institutional conditions also play major roles³.

In China, the government has introduced national strategies such as the “Health China 2030” plan and the “New Generation Artificial Intelligence Development Plan” (2017), both of which encourage the use of AI in healthcare. Still, the use of AI in hospitals and public health systems remains slow. Studies show that China’s healthcare system faces challenges in aspects such as institutional design, resource distribution, and financial mechanisms⁴. The key issue is that, under a system dominated by public hospitals (meaning that budget cycles are short and data governance is fragmented), AI development faces deep structural barriers and challenges⁵. In other words, even when the technology itself is ready to be deployed, it cannot be easily transformed into real-world applications. Governance and economic factors matter just as much.

Although countries like France and Sweden have already implemented AI into healthcare management, the use of AI in clinical settings is still under development. In non-Western countries such as China, strict regulations and budget systems create even more barriers (as one can imagine – red tape). Therefore, it is important to re-examine older policy frameworks and clear the obstacles that limit AI implementation.

While existing studies focus on AI in Western healthcare systems, this paper takes a different approach. It draws on an interview with Jun Xia, the chairman of Goodwill Medical Group in Beijing, a company that is at the forefront of AI adoption in Chinese hospitals. The objective of the paper is to understand the unique challenges faced during China’s local development of medical AI techniques and to explore how different social environments shape the spread of these technologies.

The interview reveals several major obstacles that Chinese entities face when developing and applying AI models. These include unreliable medical datasets, digital divide between big and small hospitals, weak collaboration between institutions, restrictions from China’s budget system, among others.

Research Gap and Purpose

Today, most research on medical AI focuses on places where AI has already been deployed or tested, or on studies done in Western healthcare systems⁶. For example, some papers discuss the benefits and ethical issues of using AI in medical imaging and personalized treatment². However, fewer studies have looked at why AI has not been widely adopted in non-Western, state-led healthcare systems.

This research aims to fill that gap. Its main purpose is to examine how institutional structures in China’s healthcare system shape the AI adoption. More specifically, this research combines qualitative interviews with analysis of policy documents and market data. The objective is to connect the “promise of technology” with the “reality of institutions”, and to better understand why AI implementation is still slow in China.

Theoretical Framework

This study uses two main perspectives, governance and political economy, to build the analytical framework. From a governance perspective, it focuses on how government agencies, hospital managers, and private technology companies interact with each other. This includes how they manage data, approve new projects, and allocate resources. From a political-economy perspective, it focuses on the annual budget system, investment-return cycles, market incentives, and the internal motivation structure of public institutions.

In other words, this study submits that technological readiness is not enough. Governance structures and economic mechanisms also play key roles. This point is especially applicable to China’s state-led healthcare system. Strong governance oversight and centralised budgeting can create policy stability for AI implementation. On the other hand, these features also reinforce long approval timelines, rigid funding rules, and weak cross-departmental coordination. These issues continue to shape how AI is adopted in practice.

Literature Review

The Development of Medical AI Around the World

Research from various countries shows that AI can substantially improve medical diagnosis and how resources are allocated. Topol² introduced the idea of “high-performance medicine”, indicating that AI can help doctors make better decisions by finding complex patterns in data or images. Jiang et al.⁷ and Shaheen⁸ also pointed out that AI creates economic benefits in drug development, personalized treatment, and hospital operations. During COVID-19 pandemic, AI played a notable role in predicting outbreaks, sorting patients, and managing resources^9‘10.

However, technical progress does not remove ethical or institutional risks. Anawati et al.¹¹ and Murdoch¹² showed that AI still faces challenges in privacy protection, algorithmic bias, and explainability. Celi et al.¹³ alerted that global medical datasets could have geographic and population biases, which then worsen health inequalities. Overall, AI has become more influential in global healthcare, but institutional barriers get in the way of utilizing the full potential of AI.

Structural Barriers to Medical AI

Studies across various countries showed that the main obstacles to using AI in healthcare systems usually come from governance instead of technical issues. Petersson et al.⁶ found that European hospitals struggle with AI adoption because of the lack of clarity in law and the lack of coordination between departments. In Canada, Anawati et al.¹¹ indicated that data silos and slow regulation reduce innovation in the public healthcare sector.

In China, the structural issues are equally severe, if not worse. Chen et al.¹⁴ explained that centralised data control and the division of responsibilities across agencies weaken interoperability. Meng et al.⁴ found that top-tier hospitals hold most of the resources, while lower-tier hospitals lack resources, especially technology support. The WHO report¹⁵ added that various local “smart health” projects in China operate as isolated information systems without national- or provincial-level connections. As a result, AI development in China appears as scattered experiments instead of a coordinated system-wide transformation.

Economic and Budget Constraints

In addition to governance issues, China’s financial system can limit AI development over the long term. According to the OECD¹⁶ and China Health Economics¹⁷, Chinese public hospitals rely on an annual budget system. This means that funds cannot roll over into the next year, so hospitals do not have incentives to make long-term investments. AI projects typically require several years of expenditure, but hospital performance evaluations focus on short-term financial results. This contradictory situation makes it difficult to support high research costs or slow returns.

Meanwhile, medical data cannot be freely commercialized, and private investment faces such barriers. This creates an investment environment that is “high-risk yet low-return”. In contrast, private healthcare systems in Western countries can use capital markets and insurance systems to share such risks, thereby allowing more flexibility¹⁶. This explains why China produces strong AI research, but has not yet turned the fruits into sustainable hospital-level applications.

Research Gap and Problem Statement

In view of the above, scholars have already explored the benefits and risks of AI in healthcare, but few studies examine how structural barriers and economic mechanisms interact within specific governance systems. This gap is particularly noticeable in China’s state-led healthcare system, where AI adoption is shaped not only by technology itself, but also by fiscal rules, regulatory practices, and coordination between different departments.

Therefore, this study asks the following questions: 
What institutional and economic mechanisms limit AI adoption in China’s healthcare system? How do these mechanisms interact, and what governance reforms could help overcome them in the future?

Methodology

Research Design

This study uses a qualitative exploratory research design. It combines a semi-structured interview with analysis of secondary data to examine the structural and economic barriers that affect AI adoption in China’s healthcare system.

This method was chosen because adopting AI in China is not only a technical issue, as described above. It also involves complex interactions between institutions, financial systems, and policy governance. Unlike quantitative studies that test specific hypotheses, qualitative studies are more adept at uncovering how institutions actually work, how organizations actually behave, and how policies are actually carried out in real life. They help build conceptual understanding in areas where large statistical datasets are not available.

This exploratory design aims to answer the question, “Why is AI adoption limited?” rather than “How effective is AI adoption?” For this reason, this study focuses on understanding processes and structural factors, not on making causal claims or evaluating specific policies.

Participants / Sample

The primary participant in this study was the chairman of Goodwill E-Health Info Co., Ltd., a company with extensive experience in hospital information systems, electronic medical records, and smart-health architecture in China. This individual was selected because of his long-term involvement in the digitalisation of hospitals and his familiarity with policy implementation and AI-related projects.

No other individuals were interviewed, and the study did not involve patients, hospital staff, or government officials. The sample is therefore intentionally narrow, yet consistent with exploratory qualitative research where depth of insight from a knowledgeable expert is prioritized over breadth of participation.

Data Collection

(1) Interview Data

The main primary data for this study came from one in-depth interview with the chairman of Goodwill E-Health Info Co., Ltd. Goodwill (A-share code 688246), which is a major medical-information company that has worked for many years on hospital information systems (HIS), electronic medical records (EMR), and smart-health solutions. According to the company’s public reports, it has provided services to more than 1,300 hospitals across many provinces in China (goodwillcis.com). Because of this background, the company can be seen as a representative actor in HIS and smart-hospital projects in China.

The interview followed a semi-structured outline and covered several topics, such as:

• how hospitals make decisions and go through approvals for AI projects;
• the main barriers in data sharing and privacy protection;
• how the hospital budget system affects technology investment; and
• the policy environment and the situation of industry capital.

The interview lasted around 90 minutes. With written consent from the interviewee, the conversation was recorded and later transcribed. All sensitive personal or institutional information was anonymized during analysis.

(2) Secondary Data

This study also collected national policies, statistical documents, international reports from the WHO and OECD, and academic and industry publications related to medical digitalisation and healthcare investment. These materials were used both for validation and for building comparative context.

Variables and Measurements

Because this is a qualitative study, the research does not use numerical variables or quantitative measurement tools. Instead, the core analytic units are themes and concepts emerging from the interview data and policy documents.

The “variables” in this context refer to institutional factors, such as data governance, funding structures, and collaboration mechanisms, that influence AI adoption. These were identified during coding and categorization of the transcript and secondary sources.

Data Analysis

After organising the interview content and the policy documents, the materials were analysed using a thematic analysis approach. First, the interview transcript was read several times to highlight the points that were most related to the research questions. Then, similar ideas and examples were grouped together. Over time, this helped form broader themes that reflect common problems.

While reading policy papers and industry reports, their information was compared with what the interviewee mentioned. This helped check whether the findings were reliable. After going back and forth several times, five main themes were identified:

• data quality and interoperability;
• digital divide across hospital tiers;
• weak collaboration and limited coordination between institutions;
• rigid budgets and long approval delays; and
• uncertainty in the investment environment.

These themes show the main institutional and economic barriers that AI faces in China’s healthcare system. They will be discussed in more detail in Chapter Results and Analysis.

Ethical Considerations

Ethical considerations were addressed by obtaining written consent from the interviewee before the conversation was conducted. The participant was informed of the purpose of the study, the intended use of the information, and the voluntary nature of participation. Recording was conducted only after consent was granted.

All personally identifiable information and any sensitive institutional details were anonymized to protect confidentiality. The study does not involve medical patients, personal health data, or clinical records, so risks to privacy were minimal. All procedures follow standard ethical practices for qualitative research involving human subjects.

Research Limitations

Because this study is exploratory, it has several limitations.

First, the research includes only one interview with an expert from the medical-information industry. The interview gave useful insights, but the study would have been more complete if more types of participants would have been involved, such as hospital managers or government officials.

Second, due to confidentiality rules in companies and hospitals, I could not access internal financial or budget records. So, the parts about investment and funding are mostly based on public information and the interviewee’s own experience.

Finally, this study mainly uses qualitative methods to understand the problem, not to produce exact statistical findings. The numbers used in the analysis are only for supporting the discussion, not for precise measurement.

Even with these limitations, the study still helped understand the challenges for AI adoption in China’s healthcare system from both an institutional and economic perspective. It also offers a useful starting point for future, larger-scale research.

Results and Analysis

Data Quality and Interoperability

During the interview, Jun Xia said, “If the model relies entirely on internet-based data, it can only reach a superficial stage before hitting the technical bottleneck. So, during the actual diagnosis and treatment period, in reality, the real world data from previous hospital diagnosis and treatment processes is actually very helpful for large model”. This shows one of the biggest barriers to medical AI in China: poor data quality and weak interoperability.

In recent years, China’s healthcare system has become more digitalised, but data from different hospitals and regions is still very scattered. Many hospitals use their own information systems (such as HIS, EMR, and PACS), which were developed separately and followed different technical standards. This means that even if hospitals hold a huge amount of medical records, their data formats and system interfaces are not compatible with each other; AI systems cannot directly combine or analyze them.

According to the China Digital Health Strategy Report (WHO, 2022), the number of local “smart health” platforms in China is growing fast. However, national data standards and sharing rules are still not fully established. Because of this, hospitals can hardly exchanging data with each other, and AI systems cannot learn or validate models across institutions.

In addition, China has a large imbalance in how medical resources are distributed. A study based on surveys from 2010 to 2020 found that the adoption rate of electronic medical records in top-tier (AAA tier) hospitals is above 85%, and in some places even higher than 95%¹⁸. On the other hand, lower-tier hospitals and county medical centres are still behind in digitalisation. Among community health centres, approximately 58% have electronic medical records, but most of these systems cannot connect with others¹⁹. This means, high-quality data is still concentrated in big-city hospitals, while smaller hospitals lack a strong digital foundation. For nationwide AI training and testing, this creates both coverage and standardisation problems.

Another study covering 2011 to 2020 reported that the average efficiency of medical-resource use across 31 provinces (measured using total factor productivity (TFP) from a digital-economy perspective) is only 0.938²⁰. This suggests that digital transformation has not yet improved system efficiency as much as expected. One main reason is poor data quality and the existence of “data islands”.

For AI model training, having more accurate and more diverse data can usually lead to better performance. However, in reality, hospital data often has inconsistent formats and many missing or incorrect entries. Researchers and companies need to spend a lot of efforts to clean and fix the data before using. This increases development costs and slows down the use of AI in real hospitals.

The result is apparent: AI models cannot generalize well across different hospitals. When promoting AI tools, developers often need to retrain or fine-tune the model for each new hospital. This takes more time and requires more funds. In the long term, without national standards and a strong system for data sharing, the overall progress of medical AI in China will continue to be limited.

From this perspective, data quality and interoperability are not just technical issues. They are institutional and governance issues. Building unified standards, improving data-sharing across hospitals, and protecting patient privacy at the same time will be one of the most important tasks for AI development in China’s healthcare system.

Digital Divide across Hospital Tiers

In the interview, Jun Xia mentioned, “Chinese enterprises at this stage will primarily choose top-tier hospitals because they have access to a substantial amount of budgets given by the government. Although lower- to mid-tier hospitals will be key institutions of the future application of AI in healthcare, their limited budgets make it more practical to focus on top-tier hospitals for the current stage”. This comment reflects a long-standing problem in China’s healthcare system: uneven digital development.

For years, medical resources in China have been concentrated in large, urban tertiary hospitals. County hospitals and grassroots clinics often face shortages in equipment, funding, and skilled workers. This gap appears not only in medical services but also in digital infrastructure.

According to one study²¹, large-city hospitals have much higher adoption rates of information systems compared to smaller hospitals. In remote regions, the use of online medical services is also much lower²². This means that smaller hospitals are less involved in high-quality, digital, and shareable data resources where AI depends on most.

Unequal funding is another important cause of this digital divide. Research shows that in 2018, direct government funding made up only about 10.1% of total revenue in public hospitals²¹. As a result, hospitals mostly rely on service fees and drug sales to cover daily operations. However, these funds usually go toward basic needs, not toward developing new digital systems or AI projects. In contrast, top-tier hospitals receive more government support and can also use research grants and private investment, allowing them to adopt AI technology early on.

This situation leads to a clear structural consequence: smaller hospitals, limited by their budgets and technical staff, struggle to join early AI pilots or contribute to model training. This widens the digital gap between urban and rural areas, and between top-tier and lower-tier hospitals.

If this gap continues, AI will not reduce healthcare inequality. It may even make it worse. Big hospitals will keep improving their AI tools, while grassroots hospitals and clinics will fall further behind because of weaker and outdated data systems.

Therefore, reducing the digital divide between hospitals is an important public governance issue. Future policies need to provide structural support to grassroots hospitals in areas such as budget funding, data-sharing mechanisms, and personnel training such that AI can be used fairly across the whole country.

Weak Institutional Collaboration

In the interview, Jun Xia said, “Collaboration between hospitals is a highly challenging topic to discuss. In China, for example, hospital collaboration is primarily facilitated by the government, medical consortia, or top-tier hospitals that help implement such partnerships. However, for an enterprise to initiate this collaboration, it is significantly more difficult, as the investment required far outweighs the potential returns, making it an irrational endeavor”. This points to one of the challenges in promoting medical AI in China: low collaboration efficiency and weak governance coordination.

China’s healthcare system is supervised by several government agencies, such as the National Health Commission (NHC), the National Healthcare Security Administration (NHSA), and the National Development and Reform Commission (NDRC). Each has its own priorities: the NHC manages hospitals and industry standards, the NHSA handles insurance payments and reimbursement rules, and the NDRC oversees major project approvals and public funding. But in practice, these departments do not always share information smoothly or make decisions in a unified way.

For example, one AI medical project may involve data-security approval, changes to insurance payment standards, and budgeting for new equipment, all at the same time. This “multi-agency management” can cause long approval processes, unclear responsibilities, and delays.

Research also shows that the collaboration problem is not only because of the lack of management, but also a structural problem. A study published in BMC Public Health²³ found that the distribution of medical resources across different regions in China is substantially unequal. The geographic Gini coefficient is between 0.6 and 0.7, which is considered high. Because of this unevenness, some areas have too many resources while others have too few; this makes cross-regional policy coordination even more complicated.

Cooperation between hospitals and companies is also constrained by organizational culture and incentive systems. Many public hospitals use an administrative style of management, with many layers of approval and slow decision-making. Tech companies, on the other hand, focus on fast testing and market-driven growth. These two working rhythms are often incompatible.

A study in Frontiers in Public Health²⁴ found that large hospitals in Beijing often maintain “slack resources”, meaning that extra resources kept aside for administrative assessments or unexpected tasks. This can help with stability in the short term, but in the long run, it reduces the motivation for innovation, as too much approval and resource reservation limits the flexible implementation of new projects.

Therefore, in a state-led healthcare system, medical AI collaborations face governance-related challenges. To improve collaboration among hospitals, companies, and government agencies, several things are in need: (1) clearer division of duties across departments, with unified rules for data approval and sharing; (2) cross-department coordination bodies for reducing repeated reviews; and (3) more room for hospitals for carrying out innovation under policy frameworks.

Only when institutional cooperation improves can AI move from small pilot projects to wider adoption.

Budget Rigidity and Delayed Approval

In the interview, Jun Xia said, “things like AI, while promising in the long term, are not the top priority for hospitals given the government’s current requirements in the short term … Chinese hospitals operate on a budget system – next year’s budget is set this year – so if there was no budget allocated last year, then no new budget was planned, meaning the pressure on procurement this year would be significant”. This shows a common problem for promoting AI in hospitals: budget rules are not flexible, so new projects struggle to receive steady funding.

In China, most public hospitals rely on yearly government funding. This means hospitals make a budget plan at the start of the year, and by the end of the year they must finish spending and settle all accounts. Any leftover funds cannot be carried into the next year.

According to²¹, government funding makes up only about 10.1% of total revenue in public hospitals. Most hospitals rely on their own service income to operate. This short-term budget system works well for daily operations, but it is not helpful for AI projects that require large and long-term investment.

AI projects often need several years of continuous work. For example, developing an AI-assisted diagnosis system can take 2-3 years. After that, it still needs regular updates to the algorithm and training data to stay accurate. If the hospital has to re-apply for money every single year, the approval process becomes slow and tiring, and projects can easily be interrupted due to policy changes or personnel shifts.

Research also shows that budget rigidity affects how well medical resources are used. Study²⁰ found that, from 2011 to 2020, the average total factor productivity (TFP) of healthcare resources across 31 provinces was only 0.938, with large year-to-year changes. This suggests that the current fiscal system does not support stable long-term investment, especially for digital and AI-related projects.

Slow approval processes are another part of the problem. Purchasing large medical equipment or software may require many layers of review: from the hospital’s finance office, from health authorities, and from local finance departments. Each step involves paperwork, expert evaluations, and budget confirmations. Sometimes this process can take months or even more than a year. These procedures may help prevent waste and corruption, but for the fast-moving AI industry, they are simply too slow. 

From an institutional point of view, this mismatch between short-term budgets and long-term investment makes it difficult for AI projects to succeed in the healthcare system. On one hand, short funding cycles make hospital leaders unwilling to take on long-term financial risks. On the other hand, AI projects usually need 2-5 years before showing real economic benefits.

Because of this, both hospitals and companies lack motivation to invest steadily in AI under the current budget system.

This situation also affects China’s overall progress in smart healthcare. In the future, if policy reforms can introduce multi-year funding mechanisms to allow some innovation projects to use funds across different years and simplify approval processes, the adoption of medical AI may become more promising and more widespread.

Investment Uncertainty

Based on the interview, it is also known that investors are still very cautious, and the reasons may include unclear approval rules, unclear charging models, and long payback periods. This shows another main issue limiting the development of medical AI in China: the uncertainty of the investment environment.

In the healthcare industry, investors care about two basic questions: Can the project make money? and when will it make money? But for medical AI projects, both answers are unclear.

First, AI systems cost a lot to develop, clean data for, and deploy. Yet, their revenue depends on whether hospitals are willing to buy and continue using them. At the moment, AI-assisted diagnosis or management tools still lack clear pricing standards. They are also not widely included in the national insurance system. As a result, the payback cycle becomes lengthy. 

On a broader level, the commercialization of healthcare in China is already complicated. Research shows that since China began reforms to drug and medical-service pricing in 2015, drug prices have gone down, but overall hospital operating costs and inpatient expenses have not decreased much²⁵. This means hospitals still face a lot of financial pressure.

In this situation, hospitals become more careful with AI-related investments, especially if the AI system does not directly generate income or is not covered by insurance reimbursement. Hospital managers tend to prioritize spending on things with short-term and visible benefits.

In addition, Chinese public hospitals face strict limits on financing channels. Studies noted that public hospitals have many policy restrictions on adjusting their capital structure; they have low access to outside financing, and most of their funds come from government budgets and their own revenue²⁶. In contrast, private hospitals and foreign healthcare institutions can use equity investment, insurance partnerships, or commercial loans to share risks. Because Chinese public hospitals operate within a more closed funding system, AI companies often struggle to build stable business models with them.

The cautious attitude of the capital market also appears in industry data. According to a 2023 report by the Qianzhan Industry Research Institute, the number of investment deals in medical AI has dropped for three consecutive years. Early-stage financing (Series A and B) fell by about 40%, and late-stage financing (Series C and above) almost came to a stop²⁷. This shows that investors are losing confidence due to policy uncertainty and long-term return risks.

In summary, several economic factors affecting the investment in AI development and deployment in China includes: (1) AI diagnostic services are not yet covered by the insurance system; (2) healthcare pricing reforms leave hospitals with limited profit margins; and (3) long investment cycles and slow approval processes make returns hard to predict.

Because of this, even though AI healthcare is seen as an important direction for the future, in reality its progress still relies mostly on government policy and public funding. Market investment remains relatively limited.

If future reforms can create clearer charging rules, more flexible insurance policies, and more open financing channels for hospitals, the investment environment for medical AI may improve in a real and lasting way.

Discussion

Linking the Findings: The Dual Role of State-Centered Governance

The study shows that China’s state-centred governance system plays a “double-edged” role in the development of medical AI. On one side, the government gives strong policy support and strategic direction. National plans such as “Healthy China 2030” and the New Generation Artificial Intelligence Development Plan (2017) give medical AI a clear, high-level position. This centralised approach helps gather resources, lowers the barriers for early pilot projects, and provides policy stability.

On the other side, the same centralised system also slows down the spread of innovation. Because of rigid budgets, layered approval procedures, and weak cross-department coordination, AI projects often struggle to move beyond the pilot stage. As mentioned in the interview, Chinese public hospitals tend to rely on administrative orders rather than market competition when making decisions, which reduces their motivation to innovate.

This pattern can be better understood through a political-economy lens. In a state-led healthcare system, a structural tension exists between policy control and market forces. Government oversight ensures fairness and safety, but it also limits flexibility and reduces the willingness of investors to take risks. This is a different situation from Western countries, where innovation is driven more by market incentives²⁸. In short, the adoption of AI in large scale depends not only on the maturity of the technology but also on the deeper institutional logic and economic motivations of the system.

Cross-National Comparison: Centralised vs. Decentralised Governance

When comparing China with other countries, an interesting contrast is observed. In places like Sweden and Canada, the healthcare system is also public, but policy making is more decentralized. This allows local health authorities to introduce digitalisation or AI projects more quickly based on their own needs. For example, in Ontario, Canada, the “AI for Health” program uses a regional funding system. Hospitals and universities can apply for support on their own and receive multi-year grants²⁹.

Sweden takes a different but related approach. The national digital health agency (E-hälsomyndigheten) sets unified standards for electronic health records; local medical regions carry out the projects. This creates a balance between standardised data and local flexibility in decision-making⁶.

In contrast, China has a more centralised policy system. While centralisation can ensure consistent policy direction, local implementation varies a lot because resources and capabilities differ across regions. This leads to uneven progress across the country. The comparison suggests that institutional flexibility and a sense of policy competition are important for deploying AI more quickly.

Similar challenges also appear in other developing countries such as India, Brazil, and South Africa.
In India, major AI health projects, like the Ayushman Bharat Digital Mission, are slowed down mainly by uneven digital infrastructure and a lack of strong privacy laws³⁰. In Brazil, the public healthcare system (SUS) supports AI pilots, but local governments have little financial independence, so many projects stay at the research stage³¹. In South Africa, AI in public health depends heavily on international aid programs (such as the WHO-AI4Health Africa initiative). However, limited domestic funding and weaker digital infrastructure make long-term sustainability difficult³².

These countries share some common problems: strong policy ambition but relatively weak financial, governance, and technical support. Compared with them, China performs well in top-level planning and technological development. However, there is still room for improvement in institutional implementation and market-based incentives.

A summary table comparing these countries is shown in Table 1 below.

Country	Governance Model	Policy/Funding Mechanism	Primary Barriers	Sources
China	Centralized, state-led governance	Annual budget system; top-down pilot programs; limited private funding	Budget rigidity, fragmented data governance, weak market incentives	14‘21‘23‘26
Sweden	Decentralized public health governance	National data standards; local health regions implement and fund projects	Coordination costs between agencies; lengthy ethics approval	6
Canada	Federal decentralized system	Provincial funding (e.g., Ontario AI for Health initiative) with multi-year grants	Regional disparities; limited continuity of funding	29
India	Mixed public-private decentralized system	National digital health mission (Ayushman Bharat Digital Mission)	Infrastructure inequality; lack of privacy regulations	30
Brazil	Decentralized under public SUS system	Local governments execute with restricted fiscal autonomy	Fiscal constraints; projects remain research-oriented	31
South Africa	Hybrid governance with international aid	Supported by WHO-AI4Health Africa and donor-funded programs	Insufficient local funding; weak technical capacity	32

Policy and Ethical Implications

Based on the findings and the cross-national comparison, China will need to focus on three main policy directions to support long-term development of medical AI.

(1) Multi-year Funding Mechanism

Currently, hospital budgets follow a one-year cycle. This makes it difficult to support long-term research, development, and maintenance for AI projects. A multi-year funding mechanism should be added to the fiscal system so that innovation projects can use funds across several years. Experiences from Canada and Sweden show that long-term funding greatly improves the survival rate and maturity of digital-health projects¹⁶.

(2) National Data-Sharing Framework

AI training needs high-quality data to be shared. China could learn from the European Union’s “European Health Data Space (EHDS)” and build a national data-governance framework that balances privacy protection with scientific innovation. Such a system would help AI models generalise better and also support fairer distribution of medical resources across regions³³.

(3) Institutional Innovation and Market Incentives

Besides funding and data systems, future policies should also improve the coordination between public-sector regulation and market-driven innovation. At present, many AI medical projects rely heavily on government subsidies and policy support, while market mechanisms are still underdeveloped. Companies do not have stable expectations about future returns.

Following examples from Canada, Sweden, and India, for example, the government could use a model involving both policy-guided fund and private capital participation. This approach keeps healthcare fair and accessible, but also brings in competition and encourages technology adoption.

This mixed model could create a more resilient innovation ecosystem. With such support, AI projects would have a better chance to grow on their own and scale up within the healthcare system.

Conclusion

This study shows that although AI has strong potential in China’s healthcare system, its adoption is still limited by institutional habits and rigid financial rules. Yearly budgets, long approval processes, and fragmented data governance make it hard for AI projects to move forward in a stable way. These problems are even more serious in mid- and lower-tier hospitals, in which digital systems are weaker and funding is limited.

The research also shows that governance plays a deciding role in how new technology is adopted. A state-led system can gather resources quickly and support pilot projects. But at the same time, it reduces local flexibility and weakens market incentives. In other words, having advanced technology alone is not enough. Whether AI can truly enter the healthcare system depends more on institutional design and economic mechanisms.

Looking ahead, the development of medical AI in China ought to focus on three areas: creating multi-year funding mechanisms, improving data-sharing rules and governance standards, and setting clear performance indicators for AI use in hospitals. Only when technological innovation moves forward together with institutional reform can AI become a real driving force for upgrading China’s healthcare system.

Appendix I: Interview with Xia Jun, April 26th,2025 – May 1st, 2025.

1. What type of AI model do you use?

Original Version:

我们现在使用的AI模型，早期是基于小型模型开发的。现在有了大模型后，我们将小型模型与大模型相结合，应用于辅助诊断。具体来说，有以下几个应用场景：第一，医疗辅助诊断，也就是临床决策支持系统（CDSS）；第二，诊疗过程中的质量控制；第三，电子病历的自动生成。这三个方向是我们目前正在开发并已具备商业化应用能力的领域。

Translated Version:

We used to work with small models, but now, with the introduction of larger models, we have combined the original models with large models to perform auxiliary diagnostics. To be more exact, the application scenarios include three main areas: the first is medical aided diagnosis, known in medical terms as CDSS(Clinical Decision Support System);the second is the control of the diagnosis and treatment process; and, the third is the automatic generation of electronic medical records. To conclude, these are the three applications we are currently developing and are also ready for commercial sales.

Original Version:

医疗大数据对AI的影响主要体现在几个方面。首先，数据质量是首要因素。对于大模型来说，仅依靠互联网数据会遇到瓶颈，难以进一步提升。而来自医院内部、诊疗过程中的真实世界数据对大模型的优化至关重要。同样，小模型虽然对数据量的要求较低，但对数据质量的要求依然很高。其次，数据的数量也很重要。只有提供足够多的优质数据来训练模型，才能显著提升模型的性能，比如在自动诊断场景中提高诊断的成功率。

Translated Version:

I think the impact of medical big data on AI primarily circulates around a few key aspects. First of all, the accuracy of AI, especially for large models, depends first on accessing a sufficient amount of real, high-quality data. If the model relies entirely on internet-based data, it can only reach a superficial stage before hitting the technical bottleneck. So, during the actual diagnosis and treatment period, in reality, the real world data from previous hospital diagnosis and treatment processes is actually very helpful for large model.Similarly, small models may have lower data volume requirements, but they still demand very high data quality. In the end, I believe data quality comes first followed by quantity, and only by feeding the model with enough high-quality data can we truly enhance its diagnostics capabilities. In the case of automatic diagnosis, by doing so, the success rate of the diagnostic function can be significantly improved.

2. From your point of view, comparing the old system with the new system, how much revenue or economic value has the application of AI improved for the company overall?(percentage data available?)

Original Version:

我感觉AI优化了刚才说的三个产品。 比如，自动生成病例主要在于提升诊疗效率，同时确保准确性不下降。我认为，自动生成病例绝对是未来一个非常有前景的方向。其次，临床辅助决策系统（CDSS）虽然无法替代医生，但能为医生提供非常精准的建议。这些建议对中低层医生尤其有价值，可以显著提升他们的诊断能力，甚至可能超越他们自身的水平。当然，对于顶级医院的资深专家来说，他们的经验可能使得他们的判断暂时优于AI生成的建议。但对于中低层医生，CDSS能够提供非常高质量的辅助建议，大幅提升诊疗效果。”

Translated Version:

I believe AI adoption in healthcare significantly improves diagnostic and medical efficiency to a certain extent.For example,automated case generation definitely enhances the efficiency of diagnosis and treatment, making it a key trend for the future. Furthermore, the CDSS, while highly beneficial, is unlikely to replace doctors entirely.  While it provides accurate recommendations and guidance for middle to lower-skilled doctors,improving their diagnostic capabilities, highly skilled doctors, with their extensive experience, may rely less on CDSS, as their decisions and insights are often more professional than the system’s recommendations.

Original Version:

AI目前仍处于早期阶段，特别是在中国，推广速度可能不如大家预期的那样快。至于经济效益能提升多少，目前还难以量化，没法用具体的百分比来衡量。效率提升的幅度也需要经过一段时间的实际应用，才能得出真正有价值的统计数据。

Translated Version:

AI, as I mentioned, is still in its early stages, particularly in the context of China, so its adoption rate is not as high as the public might have expected ideally.  As a result, it’s challenging to accurately estimate the total revenue increase AI will generate. Moreover, the impact on the efficiency is not quantifiable as well at the current stage. However, after a longer term usage, I believe there will be more reliable statistics to emerge that determine the exact extent to which efficiency has been improved.

Original Version:

从收入角度看，AI应用肯定会带来提升。但中国的医院与国外不同。国外多是私立医院或医疗集团，而中国主要是政府开办的医院。这些医院的预算机制决定了它们通常是今年制定明年的预算，明年制定后年的预算。因此，尽管AI应用今年开始加速推进，但当年的收益可能不会显著

Translated Version:

In terms of income, there is certainly an improvement due to AI adoption. However, the situation in China differs from that in foreign regions. Overseas, private hospitals and medical groups are the primary focus, while in China, most hospitals are run by the government. In other words, our  budget system operates on a multi year cycle, with this year’s time determining next year’s budget allocation. Consequently, despite the seemingly high adoption rate of AI, the revenue generated this year will not be as substantial in response. 

3. Quantifiable data?

Original Version:

AI相关的数据现在还没有，至于真正用AI诊断，中国也是刚刚开始。

Translated Version:

No data yet, since China only recently began to adopt AI in their medical care system. 

4. Do hospitals have a budget management plan?

Today the government has apparently increased the budget for AI development, but will this cost be passed on to consumers? Will it remain free when consumers encounter more complex problems that require advanced AI solutions? Will this upgrade process remain free, or will there be a fee?

Original Version:

嗯，一些AI诊断模型需要收费，比如影像诊断。不过，在中国，收费需要经过漫长的审批流程，首先要通过国家发改委的收费条件审批，还要经过医保局的审核等步骤。我认为收费是大趋势，但部分AI服务可能会持续免费。不过，现在讨论这个问题还为时过早。正如前面提到的，AI在医疗领域的应用表面上很热，但实际投入与预期之间仍存在较大差距。比如，中国的A股上市公司中，真正从事AI医疗的企业大多仍处于亏损状态。

Translated Version:

 Some AI diagnostic models do require payments, such as those for imaging diagnostics. However, the pricing process involves a lengthy approval procedure – First, the National Development and Reform Commission must approve it, confirming it meets the conditions for charging; then, the National Healthcare Security Administration must also grant approval, and so on.  That being said, I believe charging will be a major trend in the future. Additionally, while some services might remain free continuously, it’s too early to determine this as the conclusion. As I mentioned earlier, AI’s applications in healthcare, despite their popularity, still face a significant gap between public perception and actual investment. A clear example is China’s A-share listed companies, where those that focused on medical AI are currently experiencing a substantial loss.

5. How to solve the digital divide? (the sense that access to AI is uneven/differential depending on class, background etc.) 

Original Version:

第六个这种事情其实它是一个全世界的问题啊，就是就是数字鸿沟这件事情，它是因为优质数据，各家医院的都是都是都是有很强的保护意识啊。你说要开放数据，最终实现互联互通，那是一个长期的过程。在美国也一样啊，美国的数据都是在一个保险公司圈儿内打通，或者在一个医疗集团圈儿内打通，很少是跨集团这个。就是政府能够统一的这个其实在美国也没有人去这么做啊，也没有实现啊，所以在中国也是一样啊，这个虽然政府它是这几年着力这个提升互联互通的水平。嗯，但是这是需要一个漫长的过程。

Translated Version:

The digital divide is a global issue. Due to the high quality nature of user’s data, hospitals worldwide have that strong protection sense, making it a long-term process to achieve data sharing and interconnectivity among them. Similarity, this is the same in the U.S., where data is typically shared only within insurance companies or medical groups, and it is rare for transnational companies to establish interconnections across different oversea organizations. In fact, even if the government were to unify data systems, widespread adoption and implementation would likely remain limited still. Similarly, in China, despite government’s efforts over recent years to enhance interconnectivity between institutions, significant challenges still remain, and achieving full integration, at the end of the day, will still take a considerable amount of time. 

6. What are the selection criteria for partner hospitals? How to start cooperation between hospitals when applying medical AI?

Original Version:

合作医院的目标，那肯定其实肯定还是跟优质医院，前期跟优质医院合作能够。在医疗在在AI的这个产品开发上能够共同探索，然后但是未来的这个使用最频率最高的肯定是在中基层的医疗机构啊。

Translated Version:

The goal of hospital collaboration is to initially focus on high quality hospitals. To be more exact, in the early stages, medical AI application development will be explored with these top-tier hospitals, but the actual future implementations will be primarily targeted to the mid-level medical institutions. 

Original Version:

医院间的合作这个话题很难说啊。院医院间的合作，这不同不同国家，你像比如中国医院间的合作更多的是通过政府撮合，或者通过医联体、医共体，或者是有头部医院啊。这个这个这个协同等等等等啊，这些个很难企业去撮合医院之间的合作，这个是比较难啊，因为这个这个投入和产出是完全不成正比的。

Translated Version:

Collaboration between hospitals is a highly challenging topic to discuss. In China, for example, hospital collaboration is primarily facilitated by the government, medical consortia, or top-tier hospitals that help implement such partnerships. However, for an enterprise to initiate this collaboration, it is significantly more difficult, as the investment required far outweighs the potential returns, making it an irrational endeavor. 

Original Version:

再补充一点，关于合作医院的选择标准，现阶段中国企业主要倾向于与头部医院合作，因为这些医院有能力提供合理的预算。基层医院虽然是未来重要的应用场景，但由于预算有限，目前合作仍以头部医院为主。

Translated Version:

To add on, regarding more on the selection criteria for partner hospitals, Chinese enterprises at this stage will primarily choose top-tier hospitals because they have access to a substantial amount of budgets given by the government. Although lower- to mid-tier hospitals will be key institutions of the future application of AI in healthcare, their limited budgets make it more practical to focus on top-tier hospitals for the current stage. 

7. How can we ensure that the service is available in all areas of China? Currently it seems to be focused only on Guangzhou or Beijing. In addition, are all the hospitals in Beijing and Guangzhou in the same degree of AI medical application?

Original Version:

广东、广州、北京、上海等地在AI医疗领域的宣传非常踊跃，但实际应用水平远低于外界普遍的认知。要确保AI在医疗中的有效使用，首先需要在产品有效性上真正体现出高使用价值。同时，中国的公立医院主要依赖政府预算支持，很难通过其他方式实现盈利。如果采用面向消费者的收费模式（To C），至少在中国的历史经验中，还没有成功的先例。

Translated Version:

In Guangdong, Guangzhou, Beijing, and Shanghai, AI healthcare initiatives are heavily promoted, but their actual implementation is lower than what the public perceives it to be. To ensure an equitable amount of access, the first thing to prioritize is probably the effectiveness of the AI product itself – whether it’s usable and beneficial for the public. In addition, for public hospitals, government budget support is the only means of sustaining these initiatives, because In China, if we’re developing a consumer paid(to-C) charging model, there haven’t been any successful previous cases. 

Specific Questions

1. The partners

Original Version:

成功案例其实不少，比如301医院、协和医院和北医三院都取得了成果。从数据角度看，这些医院已经开展了大量工作，但实际价值可能不像外界想象的那样显著。AI更多是为特定目标服务的，比如科研等领域，而非通用的高价值应用。

Translated Version:

There are many successful cases, such as Chinese People’s Liberation Army General Hospital, Peking Union Medical College Hospital, and Peking University Third Hospital, all of which are examples of successful implementations of AI into the healthcare industry. From a data perspective, a lot has been done, but it’s not quite what people imagine – something with obvious value and significance.Alternatively, it often serves through a specific purpose, and by this, I mean data services are tailored to support specific tasks, such as research, and so on. 

Original Version:

你要说数据的应用，比如基于真实世界数据形成的临床辅助决策系统（CDSS），在北京的协和医院和北医三院已经得到应用。这种方式利用真实世界数据为医生提供精准的诊疗建议。

Translated Version:

If you’re talking about using the data – for example, real world data to form clinical decision support – hospitals like Peking Union Medical College Hospital and Peking University Third Hospital in Beijing are adopting this approach.

2. Are there any specific unmentioned collaborating companies or any future plans that you can tell us about?

Original Version:

我们现在正在与华为和Deepseek合作，实际上就是硬件与软件的结合。DeepSeek主攻大模型，而华为则专注于提供算力支持。二者的结合很大程度上是政策性的。目前，国内许多大模型应用厂商已转向使用DeepSeek，因为相比其他国内大模型，DeepSeek的使用量最大。华为在算力方面的优势主要体现在其强大的市场能力和与政府的良好关系。因此，在To B（面向企业）或To G（面向政府及国有企业）的行业中，客户更倾向于选择华为。对于民营企业来说，使用英伟达的可能性较大，不过国内其他厂商也在逐渐突破。我们目前主要接触过英伟达和华为的算力解决方案，之前用过英伟达，现在主要用华为。

Translated Version:

Deepseek and Huawei are the two companies that we’re currently collaborating with. In particular, Huawei mainly handles the computing power, while Deepseek focuses on the software aspects of things. Their partnership is largely driven by policy considerations. Indeed, Huawei and Deep can collaborate, especially since the application of large language models in China has shifted significantly toward Deepseek after its first introduction – previously, there were many large model providers in China, but now Deepseek has the largest user base. On the other hand, Huawei’s strength lies in its robust market capabilities and strong government ties, making it a preferred choice for industries serving business(B2B) or government entities(G2G). For private enterprises, NVIDIA is also a common choice, though China has other manufactures gradually making breakthroughs. We haven’t engaged with them, though; our experience is mainly with Huawei and NVIDIA, currently using Huawei.

3. Based on the AI small models you discussed above, could you provide some names of standard models you use?

Original Version:

病例内涵质控，临床辅助诊断，病例自动生成。

Translated Version:

case content quality control, clinical diagnostic assistance, and automated case generation.

4. If possible, could you provide a few examples of how small AI models are applied in clinical trials or other fields(such as data collection, precision surgery, etc…)?

Original Version:

专科专病多模态辅助诊断。

Translated Version:

Specialty specific disease multimodal auxiliary diagnosis.

5. You previously mentioned that ‘data quality takes precedence over quantity’, so what does ‘quality’ mean here, and what does ‘quantity’ mean? Could you provide examples to illustrate ‘quantity’ and ‘quality’ separately?

Original Version:

质量是指头部医院的正确的诊断治疗数据，数量是指基层医院的海量数据。

Translated Version:

Quality refers to the accurate diagnostic and treatment data from top-tier hospitals, while quantity refers to the massive data from primary hospitals.

6. You also mentioned some observable trends and records; where specifically can I find these important statistics? Additionally, is there any internal data available that could potentially benefit my research?

Original Version:

首先，医疗数据涉及高度隐私问题，我们无法直接从医院获取这些数据。另外，公开网站上能找到的资源主要是SCI学术文章，但我认为这些内容对你来说可能太复杂，且难以具体引用。”

Translated Version:

The data is highly private, and we can’t get it from the hospital. As for the other, on public websites, it’s just some SCI articles, but I think this is useless to you, as you can’t cite them specifically during your essay writing.

7. You also discussed how some services remain free; from the perspectives of hospitals and patients, which services are provided free of charge?

Follow up 1:As technology advances, will consumers want to know if they can subscribe to these advanced services? What services are currently free or paid for consumers? Overall, how informed are consumers about these services?

Follow up 2:You mentioned that “paid services may be a future trend”; from the current hospital perspective, what are some clear examples that support this statement?

Original Version:

免费是指中国的医院都是公立医院，都是公立医疗机构，运营主要依赖政府预算。这些预算通常优先满足短期必需开支，而AI这类技术虽然长期前景看好，但在短期内往往不是医院或政府预算的首要考虑。因此，目前AI应用在医院中更多以免费形式推广。由于中国医院采用预算制，今年制定明年的预算，如果上一年未纳入AI相关预算，今年的采购压力就会很大，这使得收费模式推广面临挑战。此外，面向患者个人的订阅模式可能是一个未来趋势，但目前中国的互联网医疗服务仍以免费为主。”

Translated Version:

Most public hospitals provide free service, because the existing budget can only cover short term necessities. In other words, this means that the existing budget can only cover short term necessities. As such, things like AI, while promising in the long term, are not the top priority for hospitals given the government’s current requirements in the short term. So, for now, the focus is likely on free service, because Chinese hospitals operate on a budget system – next year’s budget is set this year – so if there was no budget allocated last year, then no new budget was planned, meaning the pressure on procurement this year would be significant. Therefore, if they want to quickly introduce AI into hospitals, charging for it currently faces some pressure. 

Original Version:

付费的成功案例虽然存在，但非常少见，个别案例并不能说明整体趋势。关键还是要看政府的政策导向。中国的医疗机构与海外有明显差异，尤其是在当前医改的关键时期，医疗机构普遍面临较大的收入压力。

Translated Version:

Additionally, I think personal subscriptions for patients are also a trend. Currently, though, health services in China are still mostly free. Likewise, for obvious examples of paid services, they are very rare – individual cases don’t really prove much, as it mainly depends on which direction the government’s leading work is heading. Chinese healthcare institutions are clearly different from those overseas as well. Moreover, China is in a critical period of healthcare reform, meaning the revenue of healthcare institutions is under enormous pressure. 

8. Regarding what you mentioned about the NRDC (National Development and Reform Commission), how long does it take, specifically, and what is the usual range(e.g., between a few to several hours, or other time units)?

Follow up1:Are there any real world examples of complex cases and simple cases?

Original Version:

AI收费审批的过程相当漫长。目前，部分影像AI应用，比如肺结节检测，已经获批收费，但实际市场交易量并不大。审批周期通常不会以天计算，至少需要一年时间。

Translated Version:

Well, the approval process for charging for AI actually takes quite a long time. Currently, some imaging AI, like those for pulmonary nodules and such, is allowed to be charged for, but in reality, the transactions aren’t that many. Furthermore, this charging approval probably won’t be counted in days; it’s definitely at least counted in years. 

9. Will patients have direct access to any of the AI’s capabilities throughout the diagnostic and treatment process?

Original Version:

从长远来看，AI医疗的发展前景可能包括基于全球数据库直接解析病例，匹配最合适的医生和治疗方案，但这需要若干年甚至更长时间才能实现。至于预算方面，在中国通常需要在年底前确定，集中在第四季度。具体时间因申报类型而异，比如国债、财政资金、中央财政或地方财政的申报时间各有不同，但大体上都在第四季度。

Translated Version:

In terms of the vision, it may be possible to parse the cases directly from the global database to match not the right doctor, to match the right treatment plan, that’s possible. That’s the vision, that’s a number or many years down the road. Well, the budget is basically we have to establish by the end of the year in China, about 4 quarters. there is also about next year’s budget is set this year, there is no specific budget set date, for example, this year September 30th must be set after next year’s budget, that is, there is no similar date There are a variety of different examples:declare the national debt, declare the financial resources, the central financial, local financial time is also different, but basically in the four seasons.

References

1. World Health Organization. (2024, October 1). Ageing and health: Fact sheet. Retrieved from https://www.who.int/news-room/fact-sheets/detail/ageing-and-health [↩]
2. Topol, E. J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44–56. https://doi.org/10.1038/s41591-018-0300-7 [↩] [↩] [↩]
3. Hassan, M., Kushniruk, A., & Borycki, E. (2024). Barriers to and facilitators of artificial intelligence adoption in health care: Scoping review. JMIR Human Factors, 11(1), e48633. https://doi.org/10.2196/48633 [↩]
4. Meng, Q., Fang, H., Liu, X., Yuan, B., & Xu, J. (2019). Consolidating the primary health care system in China. The Lancet Public Health, 4(12), e601–e608. https://doi.org/10.1016/S2468-2667(19)30164-4 [↩] [↩]
5. Wang, Y., Castelli, A., Cao, Q., & Liu, D. (2020). Assessing the design of China’s complex health system – Concerns on equity and efficiency. Health Policy Open, 1, 100021. https://doi.org/10.1016/j.hpopen.2020.100021 [↩]
6. Petersson, L., Larsson, I., Nygren, J. M., Nilsen, P., Neher, M., Reed, J. E., Tyskbo, D., & Svedberg, P. (2022). Challenges to implementing artificial intelligence in healthcare: a qualitative interview study with healthcare leaders in Sweden. BMC Health Services Research, 22(1). https://doi.org/10.1186/s12913-022-08215-8 [↩] [↩] [↩] [↩]
7. Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., Wang, Y., Dong, Q., Shen, H., & Wang, Y. (2017). Artificial intelligence in healthcare: Past, present and future. Nature Biomedical Engineering, 1, 1–11. https://doi.org/10.1038/s41551-017-0069 [↩]
8. Shaheen, M. Y. (2021). AI in healthcare: Medical and socio-economic benefits and challenges. ScienceOpen Preprints, 1, 1–9. https://doi.org/10.14293/S2199-1006.1.SOR-.PPRQNI1.v1 [↩]
9. Balasubramanian, S., Jain, R., & Patel, P. (2023). Applying artificial intelligence in healthcare: Lessons from the COVID-19 pandemic. International Journal of Production Research, 63(2), 596–606. https://doi.org/10.1080/00207543.2023.2263102 [↩]
10. Abdulkareem, M., & Petersen, S. E. (2021). The promise of AI in detection, diagnosis, and epidemiology for combating COVID-19: Beyond the hype. Frontiers in Artificial Intelligence, 4, 652669. https://doi.org/10.3389/frai.2021.652669 [↩]
11. Anawati, A., Borycki, E., & Kushniruk, A. (2024). Artificial intelligence and social responsibility in Canada’s healthcare system: A rapid literature review. PLOS Digital Health, 3(9), e0000597. https://doi.org/10.1371/journal.pdig.0000597 [↩] [↩]
12. Murdoch, B. (2021). Privacy and artificial intelligence: Challenges for protecting health information in the digital age. BMC Medical Ethics, 22, 122. https://doi.org/10.1186/s12910-021-00687-3 [↩]
13. Celi, L. A., Cellini, J., Charpignon, M. L., Rahimian, F., Saberian, M., & Cai, T. (2022). Sources of bias in AI that exacerbate healthcare disparities: A global review. PLOS Digital Health, 1(3), e0000022. https://doi.org/10.1371/journal.pdig.0000022 [↩]
14. Chen, Y., & Esmaeilzadeh, P. (2022). Data governance and interoperability in China: Opportunities and challenges for digital medicine. Nature Digital Medicine, 5(6), 520–529. [↩] [↩]
15. World Health Organization. (2022). China digital health strategy report. Geneva: WHO Regional Office for the Western Pacific. [↩]
16. Organisation for Economic Co-operation and Development (OECD). (2023). Health policy studies: Rethinking health investment models. Paris: OECD Publishing. https://doi.org/10.1787/healthpol-2023-en [↩] [↩] [↩]
17. China Health Economics. (2022). Analysis of public hospital financial management under annual budget system. China Health Economics Journal, 41(8), 12–18. [↩]
18. Zhang, Z., Gao, J., Wang, Q., & Liu, X. (2022). A decade of hospital electronic medical record adoption in China: Patterns, barriers, and implications. JMIR Medical Informatics, 10(1), e33600. https://doi.org/10.2196/33600 [↩]
19. Xia, Z., Wang, D., Li, H., & Huang, C. (2020). Implementation of electronic medical records in community health service centers in China: Progress and challenges. International Journal of Environmental Research and Public Health, 17(22), 8510. https://doi.org/10.3390/ijerph17228510 [↩]
20. Zhang, Y., & Wang, L. (2023). Evaluation of the medical resource allocation based on digital economy perspective. In Proceedings of the 9th International Conference on Social Science and Higher Education (ICSSHE 2023) (pp. 321–326). Clausius Scientific Press. https://doi.org/10.23977/icsshe.2023.078  [↩] [↩]
21. Wang, Y., Xu, X., Zhang, J., & Liu, C. (2022). Financial sustainability of public hospitals in China: Government subsidies and internal revenue structure. International Journal of Environmental Research and Public Health, 19(23), 15507. https://doi.org/10.3390/ijerph192315507 [↩] [↩] [↩] [↩]
22. Zhou, M., Zhao, L., Kong, N., Campy, K. S., & Wang, S. (2019). The digital divide and health disparities in China: Evidence from a national survey and policy implications. Journal of Medical Internet Research, 21(9), e13631. https://doi.org/10.2196/13631 [↩]
23. Yin, D., Li, J., Zhang, Q., & Wang, L. (2025). Spatial inequality and coordination challenges in China’s healthcare resource distribution: Evidence from Gini coefficient analysis. BMC Public Health, 25(1), 22567. https://doi.org/10.1186/s12889-025-22567-w [↩] [↩]
24. Qiu, L., Zhang, J., & Chen, X. (2022). Slack resources and hospital performance: Evidence from public hospitals in Beijing, China. Frontiers in Public Health, 10, 982330. https://doi.org/10.3389/fpubh.2022.982330 [↩]
25. Chen, J., Xu, L., & Zhou, Y. (2020). Hospital cost structures and pricing reform in China: Evaluating the impact of medical service price adjustments. BMC Health Services Research, 20(1), 584. https://doi.org/10.1186/s12913-020-05849-4 [↩]
26. Huang, Q., Zhang, W., & Li, H. (2022). Capital structure and financial constraints of public hospitals in China: Policy implications for sustainable financing. PLOS ONE, 17(10), e0275580. https://doi.org/10.1371/journal.pone.0275580 [↩] [↩]
27. Qianzhan Industry Research Institute. (2023). China Artificial Intelligence in Healthcare Industry Research Report 2023. Beijing: Qianzhan Industry Research Institute. [↩]
28. Evans, P. (2019). Embedded autonomy: States and industrial transformation. Princeton University Press. [↩]
29. Government of Ontario. (2022). AI for Health Ontario: Strategy and funding framework. Toronto: Ontario Ministry of Health.
    [↩] [↩]
30. NITI Aayog. (2021). Responsible AI for All: Strategy for India. New Delhi: Government of India. [↩] [↩]
31. Pereira, R., Costa, C., & De Lima, F. (2023). Artificial intelligence in Brazil’s public healthcare: Opportunities and governance gaps. Frontiers in Public Health, 11, 1256–1264. [↩] [↩]
32. WHO Regional Office for Africa. (2023). AI for Health in Africa: Regional Outlook Report. Geneva: World Health Organization. [↩] [↩]
33. European Commission. (2022). European Health Data Space (EHDS) proposal. Brussels: European Commission. [↩]