Evolutionary Patterns and Advanced Strategies of Health Policies Based on Topic Modeling and Social Network Analysis

Abstract

We systematically analyze the evolutionary characteristics of China’s public health policies, focusing on the dynamic changes in policy content, stage-specific differences, and inter-subject collaborative relationships. Based on 137 public health policy documents issued by the central government, the analysis is conducted from a dual perspective: first, the BERTopic model is employed to identify prominent policy themes and track their evolutionary paths; second, Social Network Analysis (SNA) is utilized to deconstruct the collaborative mechanisms and network structural characteristics among policy actors, goals, and tools. The findings indicate: (1) Collaboration among core policy actors is close, yet inter-departmental transparency and collaborative inclusivity remain limited for certain organizations. (2) Policy goals show a diversifying trend, with the strategic focus shifting from infectious disease prevention and control to comprehensive public health services. (3) There are significant preferences in the selection of policy tools for balancing rapid emergency response with sustainable long-term health governance. These findings reveal the evolutionary laws of the public health policy system and provide a theoretical basis for optimizing the policy framework and enhancing governance efficacy.

1. Introduction

Against the macro-background of a profound restructuring of the global health governance architecture, the “triple crisis” constituted by the pandemic spread of emerging infectious diseases, complex environmental pollution syndromes, and the burden of chronic diseases is reshaping the international public health policy agenda. As a paradigmatic global public health crisis, the spread of COVID-19 has not only exposed the fragility of national public health systems [1] but also highlighted the imperative of perfecting mechanisms for major epidemic prevention and control, as well as strengthening national public health emergency management systems [2]. Constructing a policy framework that balances emergency response efficacy with long-term governance functions under resource constraints remains a pressing challenge in global health governance research [3]. As emphasized by van Reisen et al. [4], the intelligent processing and FAIR (Findable, Accessible, Interoperable, Reusable) management of health data are essential for monitoring crisis responses and optimizing resource allocation.

To address these global issues, China has promulgated a series of public health policies, undergoing a phased transition from an era prioritizing both revolution and hygiene, to one of universal welfare, followed by market-oriented reforms, and finally to the perfection of emergency and health systems [5]. China’s public health policy evolution can be broadly categorized into three stages: (1) The infectious disease prevention and control stage (pre-2003), which focused on the construction of epidemiological surveillance systems and the improvement of emergency response mechanisms; (2) The universal health-oriented stage (2003–2015), which realized a transformation toward health promotion strategies through the construction of a series of institutional systems; and (3) The Healthy China construction stage (2015–present), dedicated to building a multi-agent governance pattern involving “government–market–society.” This paradigmatic evolution is reflected in legislative milestones such as the Law on Basic Medical and Health Care and the Promotion of Health, and has been fully validated in practices such as the “dynamic zero-COVID” strategy, demonstrating the adaptability and resilience of the institutional system [6].

Despite extensive research on public health policy evolution, existing studies predominantly rely on static content analysis or single-method approaches, which often fail to capture contextual semantic nuances and dynamic multi-agent interactions. Traditional topic modeling techniques, such as Latent Dirichlet Allocation (LDA), operate on bag-of-words assumptions that neglect lexical order and contextual dependencies, thereby obscuring nuanced thematic shifts. Similarly, conventional social network analyses frequently treat policy collaboration as static, overlooking the temporal coupling between institutional restructuring and semantic evolution. The theoretical gap lies in the insufficient integration of advanced natural language processing techniques with network governance theories to elucidate the driving mechanisms behind policy adaptation. Furthermore, current methodologies frequently overlook the validation of topic stability and the structural inertia within collaborative networks, limiting their explanatory power regarding institutional logics. Therefore, this paper constructs a four-dimensional analytical framework comprising “Theme-Actor-Goal-Instrument,” integrating the BERTopic model and Social Network Analysis (SNA). Specifically, the BERTopic model is employed to systematically examine the diachronic evolution patterns of policy hotspots within 137 central-level policy documents. Subsequently, Social Network Analysis is utilized to quantify the degree centrality and betweenness centrality of policy actors, thereby deconstructing the evolutionary characteristics of the network structure. Finally, by examining the evolving relationships between policy actors and goals, as well as instruments and goals, this study elucidates the dynamic adaptation mechanism between governance tools and strategic orientation.

We aim to theoretically elucidate the evolutionary mechanisms of public health policy, expand the explanatory boundaries of policy process theory, and provide decision-making references for developing nations seeking to build resilient public health systems.

2. Literature Review

2.1. Institutional Evolution and Governance Logics of Public Health Policy

The development and evolution of China’s public health policy have consistently constituted a focal domain within public policy research. Empirical evidence indicates that changes in value consensus serve as the endogenous driving force for policy change, while governmental preferences directly constrain the selection of policy tools [7,8]. Building upon this premise, historical analyses demonstrate that the governing philosophy promoted a structural shift in public health policy from an economic orientation toward a value-driven, health-equity paradigm [9]. In crisis contexts, quantitative modeling has further validated the operational necessity of targeted containment strategies, providing methodological foundations for scientific decision-making during public health emergencies [10]. Recent network topology analyses have revealed the dynamic coordination patterns among multi-source governance subjects during emergency response [11].

Internationally, research on policy response mechanisms has deepened the understanding of organizational and cognitive dimensions in complex governance environments. Reviews of “wicked problems” have highlighted how bureaucratic structures and cognitive biases shape policy formulation and execution [12]. Following the onset of the global pandemic, econometric assessments confirmed that government intervention policies played a decisive role in modulating infection growth rates [13], while cross-national databases emphasized that dynamic policy adjustment remains critical for epidemic control [14]. Despite these contributions, a persistent limitation remains: existing literature predominantly focuses on either macro-level policy shifts or isolated crisis responses, failing to systematically integrate longitudinal thematic evolution with multi-agent collaborative networks.

2.2. Methodological Advances in Policy Text Analysis

Policy texts serve as crucial artifacts for comprehending public policy transitions. Contemporary scholarship has established three primary methodological pathways—structured content analysis, policy bibliometrics, and computational text mining—to examine content structure and evolutionary trajectories [15]. Content analysis emphasizes the systematic coding of keywords, policy actors, and implementation logic. Recent applications have successfully mapped evolutionary pathways in government data-sharing policies [16] and uncovered combinatory mechanisms among policy instruments in crisis scenarios [17]. Bibliometric approaches utilize quantitative mapping to reveal structural characteristics and actor relationships [18], with recent studies evaluating instrument matching deficiencies and inter-departmental synergies in talent evaluation frameworks [19], as well as exploring theoretical–practical interactions in public cultural service policies [20].

With advancements in Natural Language Processing (NLP), computational text mining has gained widespread application. While LDA models have effectively identified phasic characteristics in digital inclusion policies for the elderly [21], PPP policy volatility [22], and regional collaborative policy spillovers [23], their reliance on static word distributions limits contextual interpretability. Conversely, embedding-based models demonstrate superior capacity in capturing latent thematic shifts [24,25]. Recent validations in health informatics confirm that contextual topic modeling outperforms traditional methods in tracing research evolution [26,27], while cluster analysis of large-scale text data effectively identifies key health risk factors and thematic characteristics [28]. Furthermore, Social Network Analysis has emerged as a robust instrument for parsing collaborative structures and topic association networks [29,30]. Scholars have successfully applied co-word and inter-governmental networks to map standardization policy evolution [31], carbon emission reduction collaboration [32], and green shipping governance decentralization [33]. Nevertheless, the integration of contextual topic modeling with dynamic network analysis in public health policy remains underexplored, particularly regarding the validation of semantic stability and the structural mechanisms driving collaborative governance [34].

2.3. Synthesized Research Gaps

Current research targeting public health policy rarely combines advanced contextual topic modeling with longitudinal Social Network Analysis. This methodological silo results in an incomplete understanding of how semantic evolution intersects with network restructuring, leaving the institutional logics and driving mechanisms of policy adaptation inadequately explained. This paper addresses this gap by employing a dual-method approach to systematically examine thematic evolution, instrumental configuration, and actor interaction mechanisms, thereby contributing to the scientific formulation and synergistic governance of public health policy.

3. Research Design

3.1. Data Sources

We collected 137 public health policy documents issued at the central government level in China from 2003 to 2022. These policy documents were selected from Beida Law Database and the Chinese Government website. Keywords such as “public health,” “medical big data,” “vaccination,” and “health education” were used to retrieve relevant departmental and working documents. Invalid documents, including routine approvals, administrative notices, and bulletins, were excluded, yielding a final sample of 137 policy documents. The evolution of China’s public health policies from 2003 to 2022 is divided into two phases: S1 (National Health Promotion Phase, 2003–2015) and S2 (Healthy China Initiative Phase, 2015–2022).

As shown in Figure 1A visually depicts the annual number of policies issued in China from 2003 to 2022, with 2020 seeing the highest number of policy documents released. Figure 1B,C show that during Phase S1, a cumulative total of 85 policy documents were issued, accounting for 62% of all documents released. while Phase S2 there were 52 policy documents issued cumulatively, accounting for 38% of the total.

3.2. Research Methods

We employ the BERTopic model to preprocess, embed, and extract topics from policy texts [35]. Based on this, a co-occurrence relationship matrix is constructed for three key policy elements: policy actors, policy objectives, and policy instruments. Using UCINET software for social network analysis [36], then focuses on measuring indicators such as degree centrality, betweenness centrality, and structural holes within the network to depict the evolutionary structure of the policy network. Furthermore, a visual policy coordination network map was generated to analyze collaborative relationships among policy-making entities, identify core nodes and key pathways within the policy network, thereby revealing structural coordination and power concentration trends during public health policy evolution.

3.2.1. The BERTopic Model

The BERTopic model is employed to analyze the deep thematic structure of policy texts. Its selection is grounded in its capacity to overcome the bag-of-words limitations of traditional models by leveraging contextual embeddings [37]. The procedure involves three steps:

• Document embeddings are generated using a pre-trained Transformer model, representing each policy as a high-dimensional semantic vector.
• Dimensionality reduction is performed via UMAP (Uniform Manifold Approximation and Projection), followed by density-based clustering using HDBSCAN to identify cohesive, low-overlap topic clusters.
• Topic representation is refined using class-based TF-IDF (c-TF-IDF), and is calculated as:

$$W_{t,c}={tf}_{t,c}·\log \left(1+{\displaystyle \frac{A}{t{f}_t}}\right)$$

(1)

where $t{f}_{t,c}$ denotes word frequency within cluster $c$, $t{f}_t$ represents global frequency, and $A$ is the average document length per cluster. To ensure model reliability, we report a C_v coherence score of 0.62 and a topic diversity index of 0.82, confirming robust semantic interpretability and minimal cluster redundancy [38].

3.2.2. Social Network Analysis (SNA)

Drawing upon network governance theory [39,40], three multi-dimensional matrices are constructed:

(1)

Actor Co-occurrence Network (1-mode): nodes represent policy actors, and edge weights quantify joint issuance frequency;

(2)

Actor-Goal Association Network (2-mode): maps policy actors to strategic objectives;

(3)

Goal-Instrument Mapping Network (2-mode): links governance goals to implementation tools.

UCINET software calculates degree centrality, betweenness centrality, effective size, and constraint. Visualization reveals power concentration trends, collaborative pathways, and structural evolution across phases.

4. Results and Analysis

4.1. Thematic Evolution of Public Health Policy

(1)

The Universal Health-Oriented Stage

Thematic modeling was performed on policy documents from the Universal Health-oriented stage (stage S1), identifying a total of 26 topics. The resulting map is presented in Figure 2, where each circle represents a distinct topic. The size of the circle corresponds to the topic’s frequency across all documents. As for the distance between circles, it indicates their semantic similarity within the coordinate system. As it can be observed, Topic 0 (Disease Epidemics), Topic 1 (Food Poisoning), and Topic 2 (Mental Health) appear with the highest frequency.

To comprehend the latent hierarchical structure of these topics, the relationships between keywords were visualized (Figure 3). It intuitively displays associations at various levels. For instance, a direct and close link exists between Topic 1 (Food Poisoning) and Topic 7 (Drinking Water Safety). Similarly, Topic 19 (Health Education) is directly associated with Topic 22 (Health Records), but indirectly associated with Topic 7 (Drinking Water Safety). To illustrate document-topic distributions at a finer granularity, Figure 4 presents a color-coded mapping where each cluster denotes documents assigned to the same topic.

(2)

The Healthy China Construction Stage

By conducting topic modeling on policy documents from the Healthy China Construction stage (stage S2), a total of 16 topics were identified. The distribution map of policy text is shown in Figure 5, and the relationships between keywords are visualized in Figure 6. Results indicate that Topic 0 (Medical and Health Services) and Topic 1 (Food Poisoning) are the most frequent topics. Furthermore, Figure 6 reveals that Topic 4 (Mental Health) and Topic 7 (Health Education) are directly related, while sharing an indirect connection with Topic 9 (Novel Coronavirus).

(3)

Comparative Analysis of Stages S1 and S2

The thematic evolution reflects three structural patterns. First, major public health events (SARS in 2003, COVID-19 in 2019) catalyzed abrupt thematic realignment. Second, persistent themes (food safety, drinking water quality, mental health, port quarantine) demonstrate institutional path dependency and sustained governmental prioritization. Third, a strategic shift occurred: Phase S1 emphasized infectious disease control, whereas Phase S2 prioritized comprehensive medical and health services, reflecting a transition from crisis-driven response to proactive health governance. These findings align with prior research on policy lock-in [19], yet extend the literature by demonstrating how digital and service-oriented themes disrupt traditional administrative silos.

4.2. Evolution of Public Health Policy Networks

The evolution of public health policy constitutes a complex network structure. It involves multiple policy-issuing bodies and implementation agencies. Within this structure, policy actors tailor adaptive policy goals and deploy targeted policy instruments across different stages and environmental contexts. Together, these three elements—policy actors, policy goals, and policy instruments—form the dynamic evolutionary network of China’s public health policy. This section applies Social Network Analysis (SNA) to this dynamic network to elucidate the structural relationships between policy networks in different phases, identify critical nodes, and measure their functional roles.

(1)

Evolution of Policy Actors

As illustrated in Figure 7, during Phase S1, the Ministry of Health and the Ministry of Education accounted for the highest volume of independent policy documents. This underscores their pivotal status and robust functional roles in the institutional construction of China’s public health policy during this stage. In Phase S2, however, the State Council’s prominence increased significantly. This shift was primarily driven by the issuance of COVID-19 prevention and control policies. The State Council’s direct issuance of documents signaled a “speed-is-critical” attitude toward treatment during major public health events and reflected the state’s prioritization of life and health in response to the urgency of the crisis.

Figure 8 demonstrates a shift in authority for independent policy issuance from the Ministry of Health and Ministry of Education in S1 to the State Council in S2. This indicates an escalating level of state emphasis on public health undertakings. Quantitatively, the proportion of policies issued jointly increased from 24% in S1 to approximately 33% of the total in S2, suggesting that S2 placed a greater focus on inter-departmental collaboration.

(2)

Analysis of Collaborative Networks

Analysis of Collaborative Networks: In the policy formulation process, the functional limitations of single actors and the overlapping responsibilities among multiple actors can lead to implementation efficiency reducing or organizational redundancy. To optimize cross-departmental efficacy, national policy-making institutions frequently adopt a joint issuance mode, integrating multi-agent resources to achieve governance goals. We used UCINET software to construct a policy actor synergy matrix, analyzing actor interactions through the dual dimensions of Centrality (Degree Centrality, Betweenness Centrality) and Structural Holes (Effective Size, Constraint). The dual-centrality topology map in Figure 9 clearly reveals a paradigm shift in the network structure across the two phases. In Global Network Level, the Ministry of Finance consistently occupies the top position in dual-centrality indicators, leading multi-departmental coordination through financial resource allocation. This highlights the strategic leverage of financial assurance mechanisms in policy implementation. Notably, the National Administration of Traditional Chinese Medicine and the National Health Commission (NHC) witnessed a surge in degree centrality during Phase S2. That reflected a significant enhancement in their policy participation frequency and direct association capabilities. Meanwhile, the former Ministry of Health and the National Health and Family Planning Commission (NHFPC) maintained high betweenness centrality due to their administrative hierarchy. It indicated that traditional administrative departments retain the power of information control and resource deployment through institutional authority.

Phased structural observations for phase S1 (Monocentric Radiation) show that the collaborative network presented a typical “single-core radiation” structure. The Ministry of Health, with a degree centrality share of 15%, served as the absolute core node. This node implemented vertical control through a triangular support architecture formed with the Ministry of Finance and the NHFPC. 76% of independent issuance rate during this phase corroborates the characteristic of a policy process dominated by a strong centralization mode. As for phase S2 (Polycentric Nesting), the network structure evolved toward a “multi-center nested” morphology. Nodes increased from 12 to 19, and network density increased by 37%, indicating a deep expansion of collaborative governance. The National Administration of Traditional Chinese Medicine, with betweenness centrality of 4.333, emerged as a critical bridge node, propelling the transformation of health governance from single-disease prevention toward an integrated “Prevention–Treatment–Rehabilitation” Big Health paradigm.

Analysis of Structural Holes: Structural hole indicators of policy synergy network can reveal the dynamics of information control among key actors, as shown in

Table 1. Effective Size and Constraint of Joint Issuance Actors in China’s Public Health Policy (Partial Data).

No.	Policy Entity	Effective Size			Degree of Restriction
		S1-2	S1	S2	S1-2	S1	S2
1	Ministry of Finance (MOF)	3	3.75	3.556	0.535	0.503	0.706
2	National Administration of Traditional Chinese Medicine (NATCM)	1.667	1.528	4.095	0.84	0.918	0.64
3	National Health Commission (NHC)	2.333	-	3.256	0.611	-	0.777
4	Ministry of Health (MOH)	3	4.121	-	0.333	0.678	-
5	National Health and Family Planning Commission (NHFPC)	1	4.335	1	1.125	0.559	1.531
6	Ministry of Education (MOE)	1	1.152	1	1	0.888	1.125
7	State Council	-	1.267	1.571	-	0.837	1.156

, detailed data are listed

Table A1. Degree Centrality and Betweenness Centrality of Joint Policy-Issuing Entities in China’s Public Health Policy.

No.	Policy Entity	Degree Centrality			Betweenness Centrality
		S1-2	S1	S2	S1-2	S1	S2
1	Ministry of Finance (MOF)	30	9	21	177.367	109	5.833
2	National Administration of Traditional Chinese Medicine (NATCM)	25	2	23	34.55	-	7.833
3	National Health Commission (NHC)	18	-	18	22.55	-	4.333
4	Ministry of Health (MOH) [Historical]	15	15	-	147.75	137	-
5	National Health and Family Planning Commission (NHFPC) [Historical]	14	8	6	99.233	98	-
6	Ministry of Education (MOE)	6	4	2	18.25	-	-
7	State Council	5	2	3	12	-	-
8	National Development and Reform Commission (NDRC)	5	-	5	6.3	-	2
9	National Healthcare Security Administration (NHSA)	3	-	3	-	-	-
10	Ministry of Civil Affairs (MCA)	3	-	3	-	-	-
11	General Administration of Quality Supervision, Inspection and Quarantine (AQSIQ) [Historical]	2	2	-	-	-	-
12	National Population and Family Planning Commission (NPFPC) [Historical]	2	2	-	-	-	-
13	Central Committee of the Communist Youth League (CYLC)	2	2	-	-	-	-
14	Central Committee for Comprehensive Management of Public Security [Historical]	2	2	-	-	-	-
15	Ministry of Human Resources and Social Security (MOHRSS)	2	2	-	-	-	-
16	China Food and Drug Administration (CFDA) [Historical]	1	1	-	-	-	-
17	China Association for Science and Technology (CAST)	1	1	-	-	-	-
18	Ministry of Agriculture (MOA) [Historical]	1	1	-	-	-	-

and

Table A2. Effective Size and Constraint of Joint Policy-Issuing Entities in China’s Public Health Policy.

No.	Policy Entity	Effective Size			Constraint (or Degree of Constraint)
		S1-2	S1	S2	S1-2	S1	S2
1	Ministry of Finance (MOF)	3	3.75	3.556	0.535	0.503	0.706
2	National Administration of Traditional Chinese Medicine (NATCM)	1.667	1.528	4.095	0.84	0.918	0.64
3	National Health Commission (NHC)	2.333	-	3.256	0.611	-	0.777
4	Ministry of Health (MOH) [Historical]	3	4.121	-	0.333	0.678	-
5	National Health and Family Planning Commission (NHFPC) [Historical]	1	4.335	1	1.125	0.559	1.531
6	Ministry of Education (MOE)	1	1.152	1	1	0.888	1.125
7	State Council	-	1.267	1.571	-	0.837	1.156
8	National Development and Reform Commission (NDRC)	1	1	2.429	1	1	0.913
9	National Healthcare Security Administration (NHSA)	-	1	1.233	-	1	1.253
10	Ministry of Civil Affairs (MCA)	-	-	1.571	-	-	1.156
11	General Administration of Quality Supervision, Inspection and Quarantine (AQSIQ) [Historical]	1	1	-	1	1.125	-
12	National Population and Family Planning Commission (NPFPC) [Historical]	-	1	-	-	1.62	-
13	Central Committee of the Communist Youth League (CYLC)	-	1	-	-	1.531	-
14	Central Committee for Comprehensive Management of Public Security [Historical]	-	1	-	-	1.125	-
15	Ministry of Human Resources and Social Security (MOHRSS)	-	1	-	-	1.62	-
16	China Food and Drug Administration (CFDA) [Historical]	-	1.15	-	-	0.94	-
17	China Association for Science and Technology (CAST)	-	1	-	-	1	-
18	Ministry of Agriculture (MOA) [Historical]	-	1	-	-	1	-

. In phase S1, the Ministry of Health and the Ministry of Finance dominated with effective sizes of 4.121 and 3.75, respectively. Their constraint indicators (0.678 and 0.503) constituted a dual-information hub structure which enhanced execution efficiency. However, 81% of nodes exceeded theoretical constraint threshold (0.7). This indicated a severe insufficiency in network openness and significant institutional barriers between departments. In phase S2, the network exhibits a structural transformation. The National Administration of Traditional Chinese Medicine became primary occupant of structural holes with effective size of 4.095, while its constraint decreased by 30% compared to S1. Meanwhile, Cross-departmental collaborative formed with the NHC, which drove the transformation of the governance paradigm. Conversely, though the State Council had a 24% increase in effective size, its constraint decreased by 38%. From overall evolutionary trends perspective, network density increased from 0.134 to 0.274, indicating a significant expansion in range of cooperation among policy actors. But, mean constraint value decreased by only 8.6%, exposing a structural inertia characterized by “broad connection but low efficiency.” Core nodes, like Ministry of Finance and NHC, maintain resource control advantages through high constraint. However, peripheral actors, like Ministry of Education, are limited by a constraint value of 1.125, which struggling to breach structural constraints to participate in deep collaboration. This persistent “Core–Periphery” differentiation suggests that dissolution of information barriers requires in-depth institutional innovation.

(3)

Evolution of Policy Actor–Policy Goal Relationships

A two-mode network model of “Policy Actors–Policy Goals” was constructed based on multi-agent governance framework of Chinese public health policy. The network systematically elucidates the evolutionary of governance system from 2003 to 2022. Results are illustrated in Figure 10.

Regarding the actor dimension, the number of policy nodes expanded from 15 in Phase S1 to 23 in Phase S2, indicating continuous enhancement in governance system specialization, detailed data are listed

Table A3. Degree Centrality and Betweenness Centrality of Policy Entities in China’s Public Health Policy.

No.	Policy Entity	Degree Centrality			Relative Betweenness Centrality (%)
		S1-2	S1	S2	S1-2	S1	S2
1	Health Bureaus of Provinces, Autonomous Regions, and Municipalities [Historical]	0.108	0.116	0.065	8.123	20.047	2.161
2	Education Departments of Provinces, Autonomous Regions, and Municipalities	0.089	0.157	0.008	0.752	1.364	-
3	Health Bureau of the XPCC	0.074	0.096	0.042	12.161	26.814	2.161
4	Education Bureau of the XPCC	0.068	0.122	0.008	0.752	1.364	-
5	Health Commission of the XPCC	0.045	-	0.085	2.085	-	5.263
6	Health Commissions of Provinces, Autonomous Regions, and Municipalities	0.045	-	0.085	2.085	-	5.263
7	Finance Departments/Bureaus of Provinces, Autonomous Regions, and Municipalities	0.047	0.018	0.078	2.42	0.56	5.263
8	Administrations of Traditional Chinese Medicine of Provinces, Autonomous Regions, and Municipalities	0.039	0.007	0.079	0.919	-	1.57
9	Higher Education Institutions Affiliated with Ministries	0.027	0.053	0.003	0.752	1.364	-
10	Ministries and Commissions/Agencies Directly under the State Council	0.026	0.004	0.049	3.992	2.622	1.57
11	People’s Governments of Provinces, Autonomous Regions, and Municipalities	0.026	0.004	0.049	2.69	0.624	1.57
12	Financial Bureau of the XPCC	0.022	0.01	0.031	0.335	0.56	-
13	Chinese Center for Disease Control and Prevention (China CDC)	0.028	0.035	0.023	2.693	2.173	1.57
14	Finance Bureau of the XPCC	0.021	-	0.051	2.085	-	5.263
15	Administration of Traditional Chinese Medicine of the XPCC	0.019	-	0.046	0.919	-	1.57
16	Direct-Affiliated Inspection and Quarantine Bureaus	0.011	0.021	-	0.054	0.036	-
17	Population and Family Planning Commission of the XPCC	0.009	0.008	-	0.357	0.627	-
18	Food and Drug Administrations of Provinces, Autonomous Regions, and Municipalities	0.009	0.011	0.01	1.399	0.914	-
19	China CDC (Abbreviated)	0.008	-	0.02	0.919		1.57
20	Ministry of Health (MOH) [Historical]	0.006	0.014	-	1.292	1.617	-
21	Health Bureaus of Cities Specifically Designated in the State Plan	0.006	0.013	-	0.262	0.505	-
22	Population and Family Planning Commissions of Provinces, Autonomous Regions, and Municipalities	0.006	0.006	-	0.357	0.627	-
23	China Center for Health Education	0.006	0.012	-	2.286	4.33	-
24	Food and Drug Administration of the XPCC	0.006	0.005	0.01	0.234	-	-
25	Healthcare Security Administration of the XPCC	0.006	-	0.014	0.919	-	1.57
26	Civil Affairs Departments/Bureaus of Provinces, Autonomous Regions, and Municipalities	0.009	0.003	0.012	0.919	-	1.57
27	Ministry of Education (MOE)	0.005	0.012	-	-	-	-
28	Health News (Press/Newspaper)	0.005	0.011	-	0.45	0.221	-
29	People’s Medical Publishing House	0.005	0.011	-	0.45	0.221	-
30	China Population Communication Center (CPCC)	0.005	0.011	-	0.45	0.221	-
31	China Population News	0.005	0.011	-	0.45	0.221	-
32	China Population Publishing House	0.005	0.011	-	0.45	0.221	-
33	Development and Reform Commission of the XPCC	0.005	-	0.008	0.45		2.483
34	Joint Prevention and Control Mechanisms for COVID-19 of Provinces, Autonomous Regions, and Municipalities	0.005	-	0.008	0.45	-	2.483
35	Joint Prevention and Control Mechanism for COVID-19 of the XPCC	0.005	-	0.008	0.45	-	2.483
36	Health Supervision Center of the Ministry of Health	0.004	0.009	-	0.864	1.066	-
37	Education Commission of the XPCC	0.003	0.006	-	-	-	-
38	Decree of the Ministry of Transport of the PRC [Likely a citation error in original source, usually denotes a document]	0.003	0.006	-	0.215	0.345	-
39	Secondary Vocational Schools	0.003	0.006	-	-	-	-
40	Central Committee of the Communist Youth League (CYLC)	0.003	0.006	-	-	-	-
41	Health Administrative Departments of the State Council	0.003	0.006	-	0.042	-	-
42	Education Departments/Bureaus of Relevant Departments (Units)	0.003	0.008	-	0.193	0.914	-
43	Patriotic Health Campaign Committees (Offices) of Provinces, Regions, Municipalities, and the XPCC	0.003	0.003	-	-	-	-
44	China Center for Health Education	0.003	0.007	-	0.013	0.033	-
45	Education Departments (Bureaus) of Relevant Departments (Units)	0.003	0.006	-	-	-	-

and

Table A4. Degree Centrality and Betweenness Centrality of Policy Goals in China’s Public Health Policy for evolution of policy actor–policy goal.

No.	Policy Goal (Objective)	Degree Centrality			Relative Betweenness Centrality (%)
		S1-2	S1	S2	S1-2	S1	S2
1	Technical Standards	0.028	0.038	-	1.285	1.285	-
2	Epidemic and Disease Prevention and Control	0.164	0.189	0.14	8.74	8.74	24.444
3	Health Education	0.137	0.158	0.094	3.277	3.277	1.111
4	Public Health Services	0.129	0.076	0.209	1.856	1.856	21.111
5	Information Management	0.048	0.043	0.023	-	-	-
6	Project Subsidies	0.034	0.018	0.05	0.445	0.445	6.667
7	Food Safety and Hygiene	0.03	0.038	0.006	0.226	0.226	-
8	Supervision and Inspection	0.028	0.037	-	-	-	-
9	Emergency Response to Public Health Emergencies	0.02	0.024	-	5.479	5.479	-
10	Emergency Disaster Relief	0.019	0.025	-	-	-	-
11	Public Welfare in Public Health	0.016	0.022	-	2.569	2.569	-
12	Port Health Emergency Response	0.015	0.02	0.001	1.285	1.285	-
13	Medication Safety	0.012	0.013	0.013	1.856	1.856	-
14	Medical Big Data	0.012	-	0.026	1.856	1.856	8.889
15	Compensation and Distribution	0.011	0.013	-	1.856	1.856	-
16	Medical and Material Support	0.01	0.013	-	1.285	1.285	0
17	Meteorology and Public Health Safety	0.009	0.013	-	-	-	-
18	Vaccination	0.004	0.006	-	-	-	-
19	Institutional Improvement	0.002	-	0.013	-	-	-
20	Human Resources and Employment	0.002	-	0.013	-	-	-
21	Early Warning and Emergency Response	0.001	0.001	-	-	-	-
22	Encouragement and Guidance	0.001	0.001	-	-	-	-
23	Transportation Support	0.001	0.001	-	-	-	-

. In Phase S1, the network exhibited a “dual-core drive” pattern: provincial education departments advanced health promotion through multi-departmental collaboration, while health departments dominated agenda setting via policy resource control. Notably, although the Chinese Center for Disease Control and Prevention exhibited low collaborative frequency (degree centrality: 0.035), it retained critical discursive power in professional decision-making (betweenness centrality: 2.173). In Phase S2, the integration of the Joint Prevention and Control Mechanism for COVID-19 significantly reconfigured the network. Its average degree centrality (1.87) surpassed that of conventional governance actors, validating the coupling between bureaucratic administration and campaign-style governance during major crises.

From the goal dimension perspective, the Phase S1 network centered on epidemic prevention and control, radiating to other domains through institutional emergency response channels. In Phase S2, a structural leap occurred: “Public Health Services” ascended to the core position, and the network expanded to encompass the 15 specialized Healthy China Actions, covering emerging areas such as elderly care and medical-prevention synergy. This reflects a paradigm shift from “passive emergency response” to “proactive health governance.” Concurrently, emerging goals like medical big data and human resource development achieved strategic embedding, reflecting post-pandemic digital transformation and a renewed focus on health equity. However, 78% of nodes exhibited zero betweenness centrality, exposing structural contradictions characterized by core–periphery differentiation and low synergistic efficiency. This underscores the urgent need for institutional innovation to dismantle resource monopolies and enhance participatory governance.

(4)

Evolution of Policy Goal–Policy Instrument Relationships

Based on policy instrument theory, a “Goal–Instrument” two-mode network model was constructed. A relationship matrix was established by selecting 50 policy instruments and 24 policy goals from the period 2003–2022. The matrix systematically parsed the dynamic adaptation mechanism of China’s public health policy system, detailed data are listed

Table A5. Degree Centrality and Betweenness Centrality of Policy Goals in China’s Public Health Policy for evolution of policy goal–policy tool.

No.	Policy Goal (Objective)	Degree Centrality			Relative Betweenness Centrality (%)
		S1-2	S1	S2	S1-2	S1	S2
1	Epidemic and Disease Prevention and Control	0.189	0.191	0.176	0.634	0.673	-
2	Health Education	0.164	0.18	0.121	0.634	0.673	-
3	Public Health Services	0.135	0.07	0.222	0.634	0.673	-
4	Food Hygiene and Safety	0.054	0.068	0.021	0.634	0.673	-
5	Emergency Response to Public Health Emergencies	0.04	0.056	-	0.634	0.673	-
6	Information Management	0.037	0.039	0.032	0.634	0.673	-
7	Technical Standards	0.029	0.041	-	0.634	0.673	-
8	Port Health Emergency Response	0.022	0.023	0.018	0.019	0.029	-
9	Project Subsidies	0.02	0.016	0.028	0.634	0.673	-
10	Medical and Material Support	0.017	0.024	-	0.019	0.029	-
11	Public Welfare in Public Health	0.014	0.02	-	0.019	0.029	-
12	Medication Safety	0.013	0.013	0.014	0.019	0.029	-
13	Emergency Disaster Relief	0.01	0.014	-	0.634	0.673	-
14	Supervision and Inspection	0.01	0.015	-	0.019	0.029	-
15	Medical Big Data	0.01	-	0.035	0.019	-	-
16	Early Warning and Emergency Response	0.009	0.013	-	0.019	0.029	-
17	Transportation Support	0.009	0.013	-	0.019	0.029	-
18	Vaccination	0.007	0.009	-	0.019	0.029	-
19	Encouragement and Guidance	0.006	0.008	-	0.019	0.029	-
20	Institutional Improvement	0.006	-	0.018	0.019	-	-
21	Human Resources and Employment	0.004	-	0.013	0.019	-	-
22	Meteorology and Public Health Safety	0.003	0.005	-	-	-	-
23	Compensation and Distribution	0.003	0.004	-	-	-	-
24	Personnel Training/Talent Cultivation	0.001	0.002	-	-	-	-

and

Table A6. Degree Centrality and Betweenness Centrality of Policy Tools in China’s Public Health Policy.

No.	Policy Tool (Instrument)	Degree Centrality			Relative Betweenness Centrality (%)
		S1-2	S1	S2	S1-2	S1	S2
1	Supervision and Inspection	0.11	0.118	0.098	4.733	3.367	7.165
2	Publicity and Public Opinion	0.106	0.122	0.076	2.574	3.367	2.405
3	Information Release and Feedback	0.105	0.123	0.082	1.37	3.367	1.11
4	Technical Training and Capacity Building	0.071	0.076	0.058	4.733	3.367	4.759
5	Theoretical Knowledge Learning	0.06	0.085	0.018	1.37	3.367	0.198
6	Professional Personnel Training	0.055	0.052	0.061	2.574	3.367	2.405
7	Financial Support	0.047	0.03	0.07	3.334	3.367	3.187
8	Performance Appraisal	0.039	0.008	0.073	0.732	0.152	1.11
9	Maintenance of Hygiene Conditions	0.029	0.051	-	0.22	0.302	-
10	Government Assistance	0.028	0.035	0.025	0.865	0.849	0.238
11	Vaccination	0.026	0.024	0.032	0.865	0.784	0.728
12	Disinfection, Isolation, and Screening	0.024	0.021	0.023	0.22	0.444	-
13	Material Support/Supplies Guarantee	0.024	0.022	0.028	0.865	0.444	0.728
14	Informatization Construction (IT Infrastructure)	0.022	-	0.053	0.732	-	1.11
15	Infrastructure Construction	0.015	0.005	0.025	0.732	0.324	1.11
16	Personal Protection	0.013	0.015	0.01	1.219	1.74	1.11
17	Entry–Exit Quarantine	0.011	0.014	0.007	0.099	0.13	-
18	Research and Development (R&D)	0.011	0.008	0.014	0.865	0.302	1.11
19	International Cooperation	0.01	0.012	0.007	0.423	0.13	0.369
20	Psychological Counseling	0.008	0.013	0.002	-	-	-
21	Investigation and Research	0.008	0.01	0.006	0.732	0.615	0.198
22	Nucleic Acid Testing (PCR Testing)	0.008	-	0.019	0.54	-	0.728
23	Personnel Mobility Control	0.007	-	0.013	-	-	-
24	Transport Health Quarantine	0.006	0.012	-	0.099	0.13	-
25	Ventilation	0.006	0.008	0.003	0.104	0.13	-
26	Promotion of Resident Health Records	0.006	-	0.016	0.54	-	0.728
27	Simulation Drills	0.005	0.011	-	0.099	0.258	-
28	Establishment of Resident Health Records	0.005	0.001	0.013	-	-	-
29	Strict Punishment in Accordance with Law	0.005	0.001	0.01	0.994	-	2.486
30	Risk Detection	0.004	0.001	0.007	0.37	-	-
31	Pilot Demonstrations	0.004	-	0.014	0.01	-	0.024
32	Reward and Punishment Mechanism	0.003	0.004	-	0.246	0.324	-
33	Restriction of Outdoor Activities	0.003	0.006	-	0.104	0.13	-
34	Resource Sharing	0.003	-	0.01	0.551	0	0.764
35	Entry–Exit Declaration	0.002	0.004	-	-	0.113	0
36	Government Support	0.002	-	0.005	0.896	-	1.267
37	Risk Assessment	0.002	0.003	-	-	-	-
38	Risk Monitoring	0.002	0.001	0.006	0.919	-	0.764
39	Policy Support	0.002	-	0.005	0.945	-	1.346
40	Data Backup	0.002	-	0.005	0.01	-	0.024
41	Standardization/Standard Setting	0.002	-	0.005	0.01	-	0.024
42	Resumption of Work and Production	0.002	-	0.004	1.385	-	1.38
43	Re-examination and Retesting	0.002	-	0.003	-	-	-
44	Follow-up Visits	0.002	-	0.003	-	-	-
45	Suspension of Business Operations	0.002	-	0.003	-	-	-
46	Mask Wearing	0.002	-	0.003	-	-	-
47	Regulated Operation	0.001	-	0.001	-	-	-
48	Contracted Service Projects	0.001	-	0.004	-	-	-
49	Implementation of Guardianship and Care	0.001	-	0.004	-	-	-
50	Physical Exercise	0.001	-	0.002	-	-	-

. Results are shown in Figure 11.

In phase S1, the policy instrument network featured a core tool cluster include Supervision and Inspection, Propaganda and Mobilization, and Information Release and Feedback (mean degree centrality 0.105, relative betweenness centrality > 2.5%). In phase S2, instrument innovation exhibited a “Three-Track Parallel” characteristic. First is Integration and Proliferation of Digital Tools. In detail, the degree centrality of Informatization Construction rose to 0.053. And the embedding of resource sharing and data backup increased significantly, illustrate a transition in policy execution toward intelligent decision-making. Second is Specialization of Emergency Tools. Specifically, eight COVID-specific tools, such as Nucleic Acid Testing and Mask Mandates etc., formed independent functional modules, highlighting response capabilities for major public health events. Last is Supplementation of Livelihood Assurance Tools. It includes thirteen economic tools, like Work Resumption and Job Stabilization Subsidies etc., which filled traditional gaps. Their mean degree centrality of 0.04 reflects high sensitivity of policy system to socioeconomic demands.

The quantitative transformation of the policy actor network cannot be interpreted merely as a statistical expansion of collaborative ties; rather, it reflects a profound recalibration of institutional governance logic [41]. The transition from a monocentric to a polycentric architecture demonstrates the state’s strategic response to the functional limitations of departmental silos under complex health challenges. The persistence of high betweenness centrality and structural constraint among core entities indicates that policy coordination remains fundamentally anchored in bureaucratic resource allocation and institutional authority. Consequently, the evolution of the public health policy network adheres to a governance logic wherein collaborative efficiency is continuously negotiated against path-dependent administrative boundaries and information asymmetries. This structural inertia necessitates institutional innovations that transcend temporary crisis mobilization, transforming ad hoc coordination into standardized, rule-based inter-departmental governance mechanisms.

4.3. Case Study

The central-level policy framework provides a unified institutional foundation for the whole country, yet differences in economic development levels, resource endowments, and governance capacities across regions lead to significant heterogeneity in policy implementation. To reveal this phenomenon, we selected Guangdong, a developed coastal province, and Sichuan, a less developed inland province, as comparative cases. Focusing on the COVID-19 pandemic period (2020–2022), which falls within Phase S2 of this study, we compare the local implementation of policies in the two provinces across three dimensions: policy goals, policy instruments, and actor collaboration patterns, as shown in

Table 2. Comparative Case Analysis of Guangdong and Sichuan Provinces.

Dimension	Guangdong Province	Sichuan Province
Policy Instruments	Highly dominated by informatization tools	Balancing informatization with grid-based management
Actor Collaboration Patterns	Multi-center parallel collaboration type	Multi-center radial collaboration type
Policy Goals	Coordinated health and economic development	Primary health protection

. Data for the two provinces are derived from provincial-level COVID-19 policy documents issued between January 2020 and December 2022, primarily collected from the official websites of the two provinces’ health commissions, provincial government portals, and authoritative media reports.

In terms of policy instruments, the central-level analysis reveals that Phase S2 is characterized by the rise of informatization construction tools and the diversification of instrument types, exhibiting a “three-track parallel” feature. Both provinces demonstrate the application of informatization tools. However, leveraging its higher level of informatization infrastructure, Guangdong has achieved a technology-driven rapid transition dominated by digital tools such as the Yuekang Code, big-data contact tracing, and grid-based digital management. In contrast, constrained by limited resources, Sichuan relies on organizational mobilization, such as grid staff conducting door-to-door outreach and “knock-on-door actions”, as the main support for the implementation of digital tools, forming a pathway that balances informatization with grid-based management. Both align with the overall trend of instrument diversification, with the difference originating from regional disparities in informatization levels.

In terms of actor collaboration patterns, the central-level analysis finds that the joint issuance network of policy actors in Phase S2 evolved from the “single-core radiation” structure of Phase S1 toward a “multi-center nested” morphology. Following this trend, both provinces also exhibit multi-actor collaborative features at the local implementation level, yet with distinct patterns. Guangdong presents a “multi-center parallel collaboration type”, in which multiple actors participate in parallel within the network through mechanisms such as rural COVID-19 prevention task forces and “three-person groups”. on the other hand, Sichuan exhibits a “multi-center radial collaboration type”. At the institutional design level, Sichuan explicitly designates the health department as the lead, while embedding the responsibilities of public security, transportation, industry and information technology, commerce, education, and propaganda departments into the emergency response process, forming a regularized division of labor and collaboration list. Vertically, it strengthens the professional guidance and unified coordination of the provincial, municipal, and county command systems.

In terms of policy goals, the central-level analysis shows that the strategic focus of Phase S2 shifted from “prevention and control of infectious epidemics” in Phase S1 to “medical and health services”. Under this macro framework of “medical and health services”, the two provinces have developed differentiated sub-goal priorities. As a major export-oriented economy, Guangdong prioritizes “coordinated health and economic development”, implementing economic safeguards through measures such as facilitating work resumption and production via the Yuekang Code and introducing enterprise support policies. As a major labor-exporting province, Sichuan prioritizes “primary health protection”, focusing on grassroots control through measures such as the “four-level responsibility contracting” system and the “five haves and one grid” approach. These differences represent adaptive adjustments of the macro goal framework based on local endowments, and neither of them deviates from the overall strategic direction of Phase S2.

5. Conclusions and Outlook

5.1. Core Findings and Institutional Logics

This study employs a dual methodological approach combining BERTopic modeling and Social Network Analysis to systematically examine the evolutionary trajectory of China’s public health policy from 2003 to 2022. The empirical analysis reveals that China’s public health policy evolution is not a linear accumulation of regulatory documents, but a dynamic process governed by the interplay of exogenous shocks, endogenous bureaucratic restructuring, technological advancement, and institutional path dependence. To transcend descriptive summaries, this study constructs an analytical framework for the driving mechanisms of public health policy evolution, elucidating four core governance logics.

Crisis-induced adaptation serves as the primary exogenous driver catalyzing thematic paradigm shifts and institutional punctuated equilibrium. The recurrent prominence of infectious disease control and emergency response in the policy corpus demonstrates how macro-level public health emergencies disrupt institutional inertia, forcing rapid agenda reconfiguration and resource reallocation. The post-2020 emergence of digital health and medical big data topics illustrates a systemic transition from reactive crisis management to proactive, data-driven health governance [42]. This mechanism underscores that policy evolution in transitional contexts is fundamentally characterized by non-linear acceleration, where external shocks catalyze the institutionalization of latent policy preferences and compress the timeline for structural adaptation.

Bureaucratic resource allocation and power restructuring constitute the endogenous mechanism governing collaborative network dynamics and inter-agency synergy. The observed structural transformation from a monocentric radiation model to a polycentric nested architecture reflects the state’s strategic recalibration of administrative authority to overcome functional fragmentation. The persistence of high structural constraint and central resource control among core entities, such as the Ministry of Finance and the National Health Commission, reveals a governance logic wherein financial assurance and professional expertise function as critical leverage points for cross-sectoral coordination. Consequently, policy network evolution remains inherently bounded by bureaucratic politics, wherein collaborative expansion must continuously negotiate institutional path dependencies, authority distribution, and information control asymmetries.

Technological empowerment and instrument modernization drive the operational transformation of policy implementation and governance capability enhancement. The structural shift from traditional administrative commands toward intelligent, data-centric instruments highlights the state’s systematic pursuit of technical governance capacity. The integration of digital monitoring platforms, inter-departmental resource-sharing protocols, and targeted economic subsidies demonstrates a pragmatic alignment of policy tools with complex, multi-dimensional governance objectives. However, the enduring structural imbalance in instrument configuration indicates that technological adoption remains functionally compartmentalized; without systemic integration and normative standardization, digital instruments risk operating as supplementary measures rather than foundational governance infrastructure, thereby constraining long-term institutional resilience.

Institutional path dependence and bounded policy learning shape the longitudinal continuity and adaptive refinement of the public health governance system. The sustained policy emphasis on foundational public health services, health education, and food safety across both evolutionary phases evidences a strong path-dependent trajectory rooted in developmental state priorities and foundational welfare commitments. Simultaneously, the iterative adjustment of actor-goal-instrument alignments reflects a process of incremental policy learning, wherein institutional actors systematically refine governance strategies based on empirical performance feedback and evolving societal health demands. This dual dynamic of structural continuity and strategic adaptation ensures that policy evolution maintains institutional stability while accommodating emergent public health challenges.

5.2. Practical Implications

The evolutionary trajectory of China’s public health policy adheres to three fundamental governance laws: (1) the law of crisis-driven institutional punctuated equilibrium, wherein exogenous public health shocks accelerate structural transformation and compress policy learning cycles; (2) the law of resource-constrained bureaucratic synergy, which dictates that network expansion and collaborative density remain bounded by institutional capacity, fiscal leverage, and authority distribution; and (3) the law of instrument-goal dynamic congruence, emphasizing that long-term policy efficacy depends on the continuous, systemic alignment of governance tools with strategic objectives rather than scenario-specific improvisation. To optimize future public health governance, policymakers must institutionalize crisis-derived coordination mechanisms, dismantle structural barriers to cross-departmental data and resource sharing, and cultivate a balanced instrument portfolio that integrates short-term responsiveness with long-term systemic resilience.

5.3. Limitations and Future Research Directions

We acknowledge several limitations. First, the temporal scope (2003–2022) excludes post-pandemic policy normalization and long-term structural adjustments emerging after 2022. Second, the sample is restricted to central-level documents, omitting provincial heterogeneity and implementation variations. Future research should incorporate multi-level comparative studies to examine regional adaptation mechanisms. Third, while BERTopic and SNA provide robust semantic and structural insights, they do not capture implementation outcomes or grassroots feedback. Subsequent studies should integrate mixed-method designs, combining text analytics with field surveys and policy execution tracking. Finally, dynamic causal modeling and machine learning-driven policy simulation should be employed to forecast institutional evolution under varying crisis scenarios, thereby enhancing the predictive capacity and systemic resilience of public health governance frameworks.