Artificial intelligence is reshaping the development paradigm of healthcare with transformative force (10). From AI-assisted interpretation of nuclear medicine in oncology and image generation services (11), to clinical AI readiness assessors enhancing diagnostic accuracy and operational efficiency while improving patient care (12), to the growing prominence of AI in veterinary medicine where numerous commercial veterinary AI products have already entered the market (13). The value of technology-enabled advancements is already evident in scenarios such as enhanced diagnostic and treatment efficiency (14) and healthcare cost control (15). The World Health Organization explicitly states in its Global Strategy on Digital Health (2020–2025) that AI serves as the core driver accelerating informatization upgrade in healthcare, with its development and regulation emerging as critical issues in global health governance. However, the governance lag risk underlying this technological leap forward must not be overlooked. AI in smart healthcare inherently involves data sensitivity, irreversible outcomes, and complex liability. Its governance extends far beyond mere technical regulation, constituting a multidimensional systemic endeavor encompassing legal, ethical, and societal dimensions. Against this backdrop, this paper analyzes core contradictions and explores implementation pathways for generative AI in smart healthcare across three dimensions: technological advancement, risk prevention and control, and value realization.

The Chinese healthcare AI policy system adopts the model of framework guidance integrated with local adaptation, rather than mere duplication. Higher-level policies such as national policies focus on top-level design. They establish the overall principles, strategic direction and core goals to form the basic framework of the policy system. In contrast, policies at lower levels such as provincial and municipal levels, take into account regional resource endowments, medical development levels, and actual needs to supplement the specific implementation details and operational arrangements. The overlapping of administrative levels is limited to the core strategic consensus, but there are significant differences in the implementation paths, division of responsibility entities, and direction of resource allocation at the execution level. Therefore, adopting the multi-level policy approach is of utmost importance. The multi-level coverage ensures the comprehensiveness of the policy system, thereby providing the solid empirical foundation for subsequent analysis of policy quality differences and the formulation of targeted optimization suggestions.

Policy texts primarily originate from the official websites of the Central People’s Government of the People’s Republic of China, provincial-level people’s governments, the National Health Commission, the National Development and Reform Commission, and other relevant departments, as well as the Peking University Law Database. The research set the subject term “smart healthcare” and “artificial intelligence in healthcare” to search for policy texts. Adhering to the principles of openness, validity, relevance, completeness, and uniqueness in policy selection (30). This study used the following principles in the filtration of the documents: (1) selection of healthcare AI policies at the national, provincial, and city level; (2) policy text directly related to AI in smart healthcare; (3) policy issued from January 2015 to October 2025; (4) removal of policy documents that are no longer effective and have been revised. We selected the most representative 77 policy texts closely related to AI in smart healthcare through information searches and manual reading, including 20 at the national level, 19 at the provincial level, 7 at the municipal level, and 31 at the prefecture-level city level, demonstrating broad coverage across administrative tiers and extensive scope.

The PMC-Index model, a text-mining-based evaluation framework, primarily measures policy consistency and effectiveness. It originates from Ruiz Estrada’s “Everything in Motion” hypothesis, which posits that all elements in the universe are in constant motion and interconnected, thus assigning equal importance to any single variable. Compared to other quantitative policy evaluation methods, the PMC-Index model’s key advantage lies in its multidimensional assessment. By establishing two hierarchical levels of variables, it captures the diverse characteristics within policies, enabling more accurate evaluation of their strengths and weaknesses (31). The parameters of the PMC-Index model are primarily extracted through text mining, significantly reducing human subjectivity and enhancing the scientific rigor and rationality of quantitative evaluation. Furthermore, within the PMC-Index model, all variables carry equal weight, and no specific numerical values are assigned to variables during the evaluation process, effectively minimizing subjectivity in the assessment. The PMC-Index model is constructed in four steps: (1) excavate text and design the variables; (2) construct a multi-input–output table; (3) compute the index of PMC variables; (4) construct a PMC-Surface. The specific analytical framework is shown in Figure 1.

Based on the construction principles of the PMC-Index model and in conjunction with the variable settings of the Smart Healthcare PMC-Index Model, a multi-input–output table was established, as shown in Table 3.

The calculation of the PMC-Index can be divided into three steps: (1) Set the secondary variable values based on Equations 1, 2, and construct a multi-input–output table; (2) Calculate the primary variable values based on Equation 3; (3) Calculate the PMC index based on Equation 4.

Among these,i is a first-order variable;jis a second-order variable;n is the number of secondary variables corresponding to the primary variable i, j,n = 1, 2, 3, 4, ….

PMC-Surface plots visually represent model calculation results, intuitively illustrating the strengths and weaknesses of policy samples. Their construction primarily relies on the PMC matrix, composed of nine primary policy variables as defined in Equation 5.

Through textual analysis of 77 smart healthcare policies, we selected 10 representative policies for analysis and evaluation. These policies are cutting-edge, all issued between 2024 and 2025. They cover a broad and comprehensive range of levels, involving the national government, various provinces, and multiple cities. The content includes medical support and security systems, data element evaluation, technology development and application, and other areas closely related to research in the smart healthcare field, as detailed in Table 4.

Following the calculation steps of the aforementioned PMC-Index model, content analysis and text mining methods were employed to assign values to the secondary variable parameters in the input–output tables of the 10 smart healthcare policies. This ultimately established the input–output tables for the 10 smart healthcare policies, as shown in Table 5.

Based on a multi-input–output table of 10 smart healthcare policies, the PMC-Index for each policy was calculated, as shown in Table 6.

Policy nature (X1): The mean value of policy attribute X1 is 0.90, indicating that the selected smart healthcare policies involve regulatory and guiding elements, while few policies lack predictions for the future development outcomes of smart healthcare.

Policy domain (X2): X2 have the highest average scores of 1, indicating that all 10 smart healthcare policies involve the fields of economy, health, and technology, taking into account the multidimensional aspects of smart healthcare in social development.

Policy timeliness (X3): X3 has the lowest score of 0.43, indicating that many policies fail to combine long-term, medium-term, and short-term goals, suggesting that most smart healthcare policies tend to prioritize immediate goals at the expense of long-term goals.

Policy objectives (X4): X4 has a score of 0.84, indicating that the comprehensiveness of objectives is generally considered in policy formulation, but there are deficiencies in promotion.

Policy content (X5): X5 has a score of 0.85, indicating that policy content covers infrastructure, service models, and security management.

Policy object (X6): The mean value of policy attribute X6 is 1, demonstrating that the policies involve a wide range of targets, fully leveraging the roles of various policy entities in smart healthcare.

Policy evaluation (X7): The mean value is the same as X1 and X6, which is 1, indicating that the policies are well-founded and have clear planning, demonstrating a certain degree of scientificity and feasibility.

Policy measures (X8): X8 has a score of 0.6, ranking second to last, indicating that efforts in financial support and legal protection are relatively low, with only a small number of policies covering these aspects. This suggests that these two aspects have not been fully considered in the formulation of smart healthcare policies and need to be strengthened.

Policy tools (X9): X9 has a score of 0.78, indicating that the selected smart healthcare policies focus on mandatory, innovative, and collaborative tools, but only three policies involve market tools, showing a lack of application.

Subsequently, based on the PMC index scores of each policy and referring to previous literature (32, 33), this study categorizes the policies into four grades: poor [0–2), fair [2–4), good [4–6), outstanding [6, 8) and excellent [8, 9), as shown in Table 7.

The average PMC-Index for the smart healthcare policy samples is 7.40, placing them in the excellent category of policy ratings. This indicates overall high quality with a degree of scientific rigor and rationality. Specifically, the ranking of the 10 smart healthcare policies is as follows: P7 > P3 > P2 > P9 > P5 > P8 > P6 > P10 > P1 > P4. Among these, one policy sample achieved a perfect rating, while nine policy samples attained an excellent rating.

To provide a more intuitive visualization of the overall scores for each smart healthcare policy, a radar chart was created. By comparing the average values of key variables across policies, it presents the overall score trends for 10 representative policies, as shown in Figure 3.

The primary variables exhibiting significant variation are policy objectives (X4), policy content (X5), policy measures (X8), and policy instruments (X9). Based on the score distribution, these four variables demonstrate marked differences across the ten policies. Some policies achieved near-perfect scores in these areas, while others scored well below the average. Taking policy objectives (X4) as an example, some focus on specific areas, such as smart healthcare sub-scenarios and regional medical collaboration policies. They score higher due to precise targeting and strong alignment with actual needs, whereas some policies with a broader scope score slightly lower because their objectives are relatively general. Variations in policy content (X5) were evident in the level of detail and specificity. Policies addressing local or industry pain points by refining technical applications and resource allocation received higher scores, while others leaning toward framework guidance with insufficient detail scored lower. Differences in policy measures (X8) primarily stemmed from their operational feasibility. Policies clearly defining responsible entities, implementation steps, and timelines demonstrated strong practicality, while others scored lower due to overly general measures lacking concrete implementation pathways. Changes in policy instruments (X9) are reflected in the diversity and adaptability of tool selection. Some policies comprehensively employ multiple instruments such as fiscal support, technical standards, and supervision and assessment, achieving high alignment with policy objectives. Conversely, certain policies overly rely on a single instrument, demonstrating insufficient flexibility in tool combinations and consequently receiving relatively lower scores. This significant variation fundamentally reflects the different policy adjustments made to address differentiated needs, as well as the distinct approaches taken by policymakers in setting objectives, planning content, designing measures, and selecting tools.

The policy nature (X1) and policy timeliness (X3) variables exhibit relatively minor variations. Most policies score near the average on these two metrics, showing no significant divergence in scores. Regarding policy nature (X1), this indicates that all policies maintain a high degree of consistency in their typological definitions. There are no substantial adjustments to their nature due to differences in policy objectives or application scenarios. This allows policy recipients to clearly grasp the core positioning and implementation orientation of the policies, avoiding execution deviations caused by ambiguous nature. The minimal variation in policy timeliness (X3) indicates relative uniformity in planning temporal dimensions such as validity periods and phase divisions. This avoids both the issue of overly short timelines causing rushed implementation and the problem of excessively long durations lacking flexibility. Such stability fosters a continuous and coherent policy implementation environment, enabling stakeholders to formulate long-term execution plans based on clear timelines. This reduces resource wastage and planning confusion caused by frequent adjustments to policy timelines. However, this minimal variation warrants scrutiny regarding alignment with dynamic practical demands. Subsequent adjustments should be made to the policy’s nature and the reasonableness of its timing in light of the industry’s development pace and external environmental changes, so as to better meet the actual development needs.

The primary variables that remained largely unchanged were policy domains (X2) and policy evaluation (X7). All ten policies achieved full marks on both variables X2 and X7, indicating that these policies consistently focused on core domains without deviating from their primary objectives. This also demonstrates that the policy evaluation system was well-designed, featuring clear evaluation criteria and standardized evaluation procedures.

The PMC-Surface plot is constructed using the PMC matrix Equation 5, forming an irregular three-dimensional uneven surface where different color blocks represent varying indicator scores. To compare the PMC-Index results across policy and regulatory texts, this paper utilizes MATLAB software to plot PMC-Surface diagrams showing peak PMC indices for each category, as illustrated in Figures 4–13. (Series 1: X1 policy nature, X2 policy domain, X3 policy timeliness; Series 2: X4 policy objectives, X5 policy content, X6 policy object; Series 3: X7 policy evaluation, X8 policy measures, X9 policy tool.) Here, the X-axis represents matrix columns, the Y-axis represents matrix rows, and the Z-axis represents individual PMC indicator scores. The color and position within the PMC-Surface diagram indicate that larger red areas and values closer to 1 signify higher scores for the corresponding evaluation indicator, reflecting a higher overall consistency level of the policy or regulatory text. Conversely, values closer to 0 indicate lower scores for the corresponding evaluation indicator, with purple representing a lower consistency level of the policy or regulatory text.

By combining the PMC-Surface plots with the specific scores for each representative policy, we further analyzed the strengths and weaknesses of the policies and then proposed potential improvement paths for policy optimization. The specific analysis results are as follows.

Policy P1, “Notice on the Three-Year Action Plan for Enhancing Smart Hospital Capabilities in Tianjin (2024–2026),” scored 6.98 in the PMC-Index, earning an “Excellent” rating and ranking ninth. Issued by the Tianjin Municipal Health Commission, this policy aims to build upon prior achievements in smart healthcare development and further address “disconnects” and “bottlenecks” in hospital informatization upgrade. While indicators like X1 scored above average, all others were at or below the mean. This may stem from P1’s robust framework and foundational core attributes, yet showing weaknesses in practical implementation and innovation execution. As a specialized local government implementation plan, it features clear authority and legal legitimacy, combining administrative guidance with enforceable requirements, resulting in a solid baseline score. The low or average scores in other indicators may stem from: Policy objectives lack quantifiable metrics, reducing measurability; Policy content lacks scenario designs tailored to Tianjin’s context, resulting in insufficient depth and specificity; Policy measures fail to specify technical support details and fiscal subsidy ratios; Policy tools rely on traditional approaches, struggling to stimulate multi-stakeholder engagement, thereby lowering indicator scores and contributing to the overall low ranking. Recommended optimization path for indicators: X3-X9-X4-X8-X5.

Policy P2, “Several Opinions on Promoting the Development of Digital Traditional Chinese Medicine,” scored 7.68 on the PMC-Index, earning an “Excellent” rating and ranking third. The National Administration of Traditional Chinese Medicine introduced this policy to innovate development pathways that deeply integrate the inheritance of traditional Chinese medicine with digital technology. Only two indicators, X3 and X9, scored below average, indicating its distinct characteristics and strong synergy, but with deficiencies in timeliness and detail. Therefore, the recommended optimization path for indicators is X3-X9.

Policy P3, “Guidance on Artificial Intelligence Application Scenarios in the Health Sector,” scored 7.73 on the PMC-Index. This policy was rated “Excellent” and ranked second. Issued by the National Health Commission, it provides specific guidance for the rational and standardized application of AI technology in the health sector. Indicators X1 and X8 scored relatively low, likely due to excessive emphasis on guidance while lacking sufficient promotion and safeguards. Therefore, the recommended optimization path for indicators is X1-X8.

Policy P4, “Shanghai Medical Artificial Intelligence Development Plan (2025–2027),” scored 6.90 on the PMC-Index, earning an “Excellent” rating and ranking tenth. Issued by the Shanghai Municipal People’s Government, this policy aims to advance comprehensive development in Shanghai’s AI in smart healthcare sector while enhancing application capabilities and innovation capacity. Notably, the scores for all ten indicators were below or equal to the average level. This indicates that while the policy achieved an overall excellent rating, it lacks outstanding advantages in individual indicator performance. Its balanced approach fails to demonstrate distinct characteristics. The recommended optimization path is X3-X9-X4-X8-X5-X1.

Policy P5, “Specifications for Information and Digitalization Development in Traditional Chinese Medicine Hospitals (2024 Edition),” scored 7.43 on the PMC-Index, earning an “Excellent” rating and ranking fifth. The National Administration of Traditional Chinese Medicine has designated this policy as a key guideline for information and digitalization development in TCM hospitals. While indicators X1, X5, and X9 scored below average, the remaining six indicators met or exceeded standards. This discrepancy may stem from ambiguous cross-departmental coordination details, insufficient rigidity in policy implementation safeguards, content relevance and depth falling short of comparable exemplary policies, and limited diversity and innovation in policy tools. Accordingly, the recommended optimization path for indicators should be X1-X9-X5.

Policy P6, “Chongqing Smart Medical Equipment Industry Innovation and Development Action Plan (2025–2027),” scored 7.18 on the PMC-Index, earning an “Excellent” rating and ranking seventh. Issued by the Chongqing Municipal People’s Government, this policy aims to drive innovation and development in Chongqing’s smart medical equipment industry, enhance its core competitiveness, and accelerate its digital and intelligent transformation. Its scores for indicators X3, X4, X5, and X9 fell below the average level. It is recommended to adjust the optimization path for these indicators in the sequence X4-X3-X5-X9.

Policy P7, “Implementation Plan for Digital and Intelligent Transformation of the Pharmaceutical Industry (2025–2030),” scored 8.22 in the PMC-Index, earning a “Perfect” rating and ranking first. Issued by the Central Committee of the Communist Party of China and the State Council, this policy carries high authority. Except for indicator X9, all other indicators scored above or equal to the average, indicating the policy’s overall strong performance, significant advantages, and sound rationality. The optimization path for indicators is X9.

Policy P8, “Beijing Action Plan for Accelerating Innovation in ‘Artificial Intelligence + Healthcare’ (2025–2027),” scored 7.23 on the PMC-Index, earning an “Excellent” rating and ranking sixth. Five of its indicators achieved scores at or above the average level, indicating strong performance in these areas that positively supported the policy’s overall effectiveness. Indicators X3, X4, X5, and X9 scored below the policy’s average. Therefore, it is recommended to optimize and adjust indicators X4-X3-X5-X9.

Policy P9, “Notice on the Action Plan for Accelerating the High-Quality Development of ‘Artificial Intelligence + Healthcare’ in Zhejiang Province (2025–2027),” scored 7.48 in the PMC-Index, earning an “Excellent” rating and ranking fourth. This policy document, issued by the Zhejiang Provincial Health Commission, promotes the efficient application and high-quality development of artificial intelligence technology within the local healthcare sector. While indicators X3, X4, and X9 scored below average, the remaining six indicators achieved scores at or above the average level. Therefore, the recommended optimization path for indicators should be X4-X3-X9.

Policy P10, “Hubei Province Implementation Plan for Accelerating the Application of Artificial Intelligence in Healthcare (2025–2027),” scored 7.13 on the PMC-Index, earning an “Excellent” rating and ranking eighth. Indicators X3, X4, and X8 scored below average. This may stem from the policy’s failure to leverage Chongqing’s strengths in manufacturing and its role as a western healthcare hub to define specialized tracks. It also lacks clarification on “the responsible entity for aligning corporate and hospital clinical needs,” “funding allocation ratios for core technology breakthroughs,” and the division of responsibilities among departments. Consequently, the measures lack concrete implementation mechanisms, hindering goal realization. Therefore, it is recommended to optimize the indicator sequence to X4-X3-X8.