key: cord-0919975-hhmmmzst
authors: Cheng, Quan; Zheng, Shulin; Xiong, Zheng; Lin, Minwang
title: Characterizing the Dynamic Evolution of Interagency Collaborative Decision-Making Networks in Response to COVID-19 in China: A Policy Document Analysis
date: 2022-03-21
journal: Healthcare (Basel)
DOI: 10.3390/healthcare10030590
sha: aecbab58581bda3a634e4d7f21c80d0f389a91c4
doc_id: 919975
cord_uid: hhmmmzst

Collaborative decision-making across multiple government agencies is considered a critical and effective strategy to combat public health crisis; however, we know little about how the collaborative decision-making works and evolves during periods of crisis. To fill this lacuna, this study uncovers the structure and evolving dynamics of the network by employing a policy document analysis. Based on the policy documents, jointly issued by the agencies of Chinese central government in four phases regarding COVID-19 control, we first constructed a co-occurrence matrix of policy-issuing agencies to outline the network structure, then drew a breadth–depth matrix to identify the role evolution of agencies, and lastly built a two-mode network consisting of policy topics and agencies to determine the evolution mechanisms of policy attentions for each agency. It was found that the network structure of interagency collaboration involves three forms: discrete structure in the early phase, subgroup structure in the middle phase, and connected structure in the latter phase. Agencies embedded in the network can be categorized into three types: leading agencies, key agencies, and auxiliary agencies, with their constituent members changed as the pandemic risks are gradually becoming under control. Furthermore, each type has its own primary policy attentions, but shares some common foci in all four phases and shifts attention in the emergency management process. This study contributes to shedding light on the formation of and variations in collaborative networks in health emergencies and provides policy implications for other countries that have struggled against COVID-19.

Public health emergencies, characterized by suddenness, publicness, harmfulness, and complexity, demand interagency cooperation, especially in the policy-making process [1, 2] . The formation of emergency response policies involves multiple government agencies. Incomplete consideration is one of the obstacles to emergency response decision-making, which leads to inadequate governance measures [3] . In the context of pandemic emergencies, a holistic view is necessary for returning people's lives to normal. To this end, policymakers should consider many aspects of post-pandemic recovery, but merely target the pandemic du jour. In a traditional hierarchical bureaucratic system, different agencies have their own separate responsibilities or missions; however, each of them is loyal to the specialized function so that they are not authorized to make emergency decisions without cooperating with other agencies [4] . However, collaborative efforts in policy-making, such as interagency information sharing and joint coordination, are inevitable due to the dynamic environment of disasters [5] . This kind of collaboration is definitely helpful to optimize health emergency responses. In practice, when multiple agencies participate in are the carriers of policy, which implicitly reflect policymakers' ideologies and provide a channel to observe the policy process [16, 17] . Moreover, the policy document analysis method has been widely used in research on prior public health emergencies [18, 19] as well as COVID-19 [1, 8] . For example, Wu et al. (2021) applied policy document analysis to characterize the patterns of China's policies against COVID-19 [1] . Cheng et al. (2021) used policy document analysis to assess the evolution of Chinese central government attention in response to COVID-19 [8] . Therefore, the policy document analysis method can be used not only to identify the policy topics, but also to analyze the evolution of policies. Considering that policy documents published by government agencies are authoritative and can directly reflect the work dynamics of government agencies, we decided to carry out policy document analysis based on China's policies regarding combating COVID-19 to examine the research questions. In detail, three specific studies were conducted. First, we applied a social network analysis method to show the network structure by constructing co-occurrence matrixes of policy-issuing agencies. Second, we drew a two-dimensional "breadth-depth" matrix to reveal the role changes of policy-issuing agencies. Third, we built a two-mode "agency-topic" network to unveil the policy focus of each agency and then mapped a Sankey diagram to work out the evolution mechanisms of policy topics belonging to the policy-issuing agencies.

The remainder of this article is structured as follows. Section 2 presents a review of the literature on interagency collaborative decision-making and policy document analysis in the public health emergencies. The data, methodology, and empirical analysis results are described in Sections 3 and 4. A subsequent discussion of the results is shown in Section 5. The final section concludes with theoretical and practical implications, as well as their limitations.

Policy capacity has become a significant indicator for measuring the performance of national-level emergency management [20] [21] [22] . In terms of policy capacity required by emergency management, the state's capacity to determine what responses are chosen ranks first. In other words, better leadership and flexible decision-making are both expected in emergency management [23, 24] . The dynamic environment of disasters requires multiple agencies participating together in cooperation and coordination in decision-making procedures to achieve a common goal [5] . In nature, collaborative decision-making in emergencies is thus a multifaceted phenomenon because it requires transboundary collaboration between policymakers with different expertise [25] . To define the nature of collaborative decision-making, the literature generally encompasses membership structures which range from an individual level to the organizational level [5] . The variation which can be summarized as a continuum is collaborative decision-making between politicians or policymakers at the one extreme, and at the other extreme is collaborative decision-making across different sectors. Therefore, interagency decision-making can be recognized as one type of collaborative decision-making.

In addition to focusing on the nature of collaborative decision-making, crisis management scholars also pay much attention to factors affecting collaborative decision-making process. These factors include political and institutional legitimacy [22] , socio-technical structures [26] , information sharing [5, 26, 27] , communication and interaction [28] [29] [30] , actor characteristics (i.e., number of actors, level of interdependency, level of mutual trust) [5] , etc. The items mentioned above cover the environmental level, cross-sector level, and organizational level. All the influencing factors can be derived from collaborative governance theory because collaborative decision-making is the beginning of collaborative governance processes. As for the advantages, collaborative decision-making plays an increasingly significant role in improving the policy capacity of emergency response [31] , and positively affect the emergency response outcome [32] . The prior disasters have made it clear that traditional crisis management characterized by hierarchy system has proved to be ineffective, especially in the policy-making process. Collaborative decision-making in emergencies brings its own advantage to the table. Moreover, the quality of collaborative decision-making has become a performance indicator to evaluate emergency response systems [33] .

In terms of techniques and guidelines that provide support for decision-making in emergencies, the literature focuses on auxiliary tools for improving the analysis capacity [34] , virtual reality techniques for assisting decision-making [35] , collaborative mapping engines for helping with decision-making [36] , knowledge management support for storing and disseminating the contextual knowledge [37] , decision support tool-set for supporting multi-agency decision-making [38] , socio-technical instruments for supporting social learning between different participants [39] , and principles for guiding the effective collaborative decision-making [40] . These techniques and guidelines aim to connect all the parties and improve the efficiency and quality of decision-making. The literature also presents many modeling methods to work out the best collaborative decision-making strategy and forecast the consequence achieved by the decision, such as agent-based modeling [41, 42] and the stochastic Petri net [43] .

Although the literature on collaborative decision-making in emergencies is considerable, few studies have focused on how collaborative decision-making works. In fact, a well-coordinated network is significant for emergency decision-making. However, scholars are nearly impossible to involve in the political area; thus, inquiring the government decision-making process becomes difficult. Thus, we turn to focus on studies on emergency management networks, which can provide theoretical implications for understanding the networks of collaborative decision-making. Among these studies, two research tracks can be traced. The first one focuses on the process level and aims to explore the factors that influence the building and sustainability of collaborative decision-making networks, including interagency communication [30] , information exchange [26] , mutual trust, and commitment to common purpose [44] . Scholars have also devoted themselves to develop tools or methods, such as information communication technology [45, 46] and task-adaptive information distribution systems [47] , to ensure the networks are successful. The second one centers on the structure level, which involves network structures and their performance. In detail, some of them conduct social network analyses to identify the key actors and structural attributes in the emergency management networks [6, [48] [49] [50] [51] , and some of them examine the relationships between structural attributes and network performances [50, 52, 53] .

Regarding collaborative decision-making networks in emergencies, the relative literature is limited. As mentioned before, Wu et al. (2021) once deconstructed the network of collaborative decision-making in emergencies; unfortunately, they failed to consider the network variation in dynamic emergency situations. Investigating emergency management networks should not neglect the dynamic developments present in disaster responses. For instance, Abbasi and Kapucu (2012) applied social network analysis to investigate the evolution of inter-organizational response networks over time [54] ; Robinson et al. (2013) used media reports and emergency operation plan data sources to track the evolution of collaborative emergency networks [55] . However, the unit of analysis of these studies is executive network. In this article, we take a closer look at the network of collaborative decision-making from a dynamic perspective by analyzing policy documents developed in response to public health emergencies, jointly issued by multiple government agencies.

Policy documents, reflecting governments' decisions, have become analytical objects increasingly used in many topics of policy research, such as policy topics evolution [56] , policy change [16] , policy diffusion [57] , and policy evaluation [58] . On the basis of policy documents, scholars have developed a unique research method by drawing on content analysis and text analysis, which is generally called policy document analysis. Based on the adopted analytical approaches, policy document analysis can be categorized into two types: qualitative document analysis and quantitative document analysis. The former features contextual interpretations and thick descriptions, based on the researcher's knowledge [59, 60] ; the latter is more prevalent in policy document research because it mainly employs bibliometric methods, which could be more objective. Two types of policy document analysis have been applied to the health research field. For instance, some scholars used qualitative document analysis to discuss the equity embedded in public health policies [61, 62] . Other scholars have applied bibliometric analysis, a quantitative method widely used in literature review research, to characterize the evolution of healthrelated policies [63, 64] .

Regarding public health emergency, policy document analysis has been highly recommended due to its advantages on analyzing policy outputs and their roles play in the emergency management. Prior to the outbreak of COVID-19, scholars had applied policy document analysis to study emergency response polices of other pandemics, such as H1N1 [19] and Ebola [65] . Policy document analysis also has been used in research about national health emergency policies. For instance, Sprogis et al. (2021) analyzed the structure and processes of a rapid response system in Australia by using document analysis [66] . Regarding COVID-19, a considerable number of studies have paid attention to emergency response polices against COVID-19, and most of them employed quantitative and qualitative document analysis methods in the policy research. Some studies have focused on the evolution of COVID-19 emergency management policies during the control and prevention periods by using some topic modeling methods [1, 8] . Except for quantitative analysis, researchers also have qualitatively analyzed COVID-19 control polices from the document perspective. Specifically, Yoo et al. (2020) compared the government guidelines regarding COVID-19 control across six countries and identified their major differences and similarities [67] . Benítez et al. (2020) reviewed the response policies for COVID-19 in five Latin American countries and argued that their impacts on health outcomes [68] . In addition, some studies have focused on some specific policy-making relating to the COVID-19 pandemic, such as child protection [69] and workforce governance [70] .

Through reviewing the literature of policy document analysis on COVID-19, it was found that although considerable research emphasizes the importance of policy study and also conducts policy document analysis to evaluate the public health emergency policies, but few studies apply quantitative document analysis to characterize the evolution of policy-making, let alone the dynamic network structure of collaborative policy-making. Therefore, it is necessary to extend quantitative methods to clarify how government agencies collaborate in policy-making processes in public health emergencies, which can not only enrich the literature of policy document analysis, but also contribute to studies on public health emergencies.

This article focuses on China's COVID-19 prevention and control practices and aims to explore the structure of the collaborative decision-making network and agency-topic evolution network. Figure 1 shows the research methods design. 

This study explores the dynamic evolution mechanism of collaborative decisionmaking networks constructed by multiple Chinese central government agencies in response to COVID-19 (Table A1 in Appendix A shows the abbreviated name of each agency). The policy documents jointly issued by multiple Chinese government agencies can be regarded as a manifestation of collaborative decision-making. The official policy document data were extracted from the official website of the State Council of China (http://www.gov.cn/, accessed on 10 September 2020), which are available in the public domain and exempt from ethical approval. Specifically, "COVID-19" and "pandemic prevention and control" were used as the search keywords, and all the data were retrieved from 20 January 2020 to 7 August 2020. Through manual review of the crawled data, we obtained a total of 584 valid data points. The data frame included the document title, publication time, policy-issuing agency, subject classification, body content, etc.

This article focuses on the practice of COVID-19 prevention and control in China, and explores the structure of collaborative decision-making networks and agency-topic evolution networks. If we use the same time interval to divide the phases based solely on the time span, it will not be able to adequately explain the characteristics of China's fight against the COVID-19 pandemic, which are the adopt of measures according to the time and adjusting the strategy in different phases. To effectively reveal the dynamic evolution of interagency collaborative decision-making networks, the development timeline of government efforts to combat the COVID-19 pandemic should be considered. Thus, this article refers to the White Paper "The China Action against COVID-19 Pandemic" released by the State Council Information Office of the People's Republic of China on 7 June 2020, in which it divides the timeline of Chinese government's efforts to combat the COVID-19 pandemic into five periods. However, the Chinese central government issued the earliest policy document concerning combating COVID-19 in the second period. Thus, we only focus on the last four periods and renamed them as "Phase T1", "Phase T2", "Phase T3" and "Phase T4", respectively (Table 1) . Phase T1 preliminary features the containment of COVID-19; a milestone was the lockdown of Wuhan city. The government issued a total of 152 policy documents, but most of them were issued separately by each policy-issuing agency; only 38 policy documents were jointly issued by multiple agencies. Phase T2 features the initial control of COVID-19. A symbolic event of this stage is the number of new local COVID-19 infections dropping to single digits, and the number of documents jointly issued by multiple agencies accounted for 26.3%. Phase T3 features the overall control of COVID-19. On 18 March 2020, China reported zero new local cases for the first time, and the number of documents jointly issued by multiple agencies accounted for 34.8%. Phase T4 features regular COVID-19 prevention and control processes. On 29 April 2020, the Chinese central government made new arrangements on implementing regular COVID-19 prevention and control measures and fully advancing work resumption, and the number of documents jointly issued by multiple agencies accounted for 37.9%. In summary, the proportion of jointly issued policy documents gradually increased at each phase.

Social network analysis is a quantitative analysis method developed by sociologists based on mathematical methods and graph theory, which can be used to analyze social networks from many different aspects, such as network density analysis, centrality analysis, and core-periphery analysis [71] . The social network can intuitively display a structure composed of multiple subjects and reflect the relationships between various nodes through various indicators.

The core of social network analysis includes constructing a co-occurrence matrix of actors in the network, drawing a collaborative network, and using three centrality types-betweenness centrality, closeness centrality and eigenvector centrality-to analyze a single node. Betweenness centrality is a measure of the ability of an actor in the network to act as a network intermediary, and it reflects the degree of control of a node over resources. Closeness centrality is used to evaluate the closeness between one node and other nodes in the network. Eigenvector centrality is used to determine the core nodes of the network. The basic idea is that the centrality of a node is a function of the centrality of adjacent nodes. It is believed that the importance of a node depends on the number of neighbors and the importance of neighboring nodes.

Text mining is based on computational linguistics and statistical mathematical analysis, combined with machine learning and information retrieval technology, to discover and extract hidden knowledge from text data in a document set independent of user information needs. It is a process of text information description to selection and extraction modes, and finally the formation of user-understandable information knowledge.

We employed the LDA (latent Dirichlet allocation) topic-training model to extract the topics of the policy documents to explore the evolution of their foci. LDA is a document topic generation model proposed by David M. Blei [72] . In essence, it is based on the three-layer Bayesian probability model of "document-topic-word" and the algorithm adopting the bag-of-words model method. Each document is regarded as a word frequency vector; each document represents the probability distribution of some topics; and each topic is the probability distribution of many words.

We selected TF-IDF (term frequency-inverse document frequency) to calculate feature weights. TF-IDF is a method of calculating feature weights in text classification. TF stands for word frequency, which is used to count the frequency of each word in the text and reflect the ability of the word to describe the content of the document. IDF is called the inverse document probability, which is used to evaluate the universality of each word to the corpus.

Selecting an appropriate number of topics has a greater impact on topic extraction and topic strength. For the LDA model, the two most commonly used evaluation methods are Perplexity and Corre. Perplexity (PP) is used to evaluate the quality of a language model. Perplexity can be understood as how uncertain the trained model is about which topic the document d belongs to for an article d. This uncertainty is the perplexity. The lower the perplexity, the better the clustering effect. For models trained on different topics, its perplexity is calculated. The topic corresponding to the minimum perplexity is the optimal number of topics. This study chose perplexity as the basis for determining the optimal number of topics. The formula for calculating perplexity is as follows:

where M is the number of documents, N d is the set of entries in each document, and P(w d ) is the probability of generating each entry.

A Sankey diagram, also known as a Sankey energy split diagram or a Sankey energy balance diagram, is a specific type of flow chart that shows the use of thermal energy and the flow of energy in a very vivid way. Sankey diagrams were originally used in energy management and the chemical industry. With the development of society, they have also been widely used in some emerging industries and financial institutions. The complex data can be displayed more clearly through the use of a Sankey diagram, and the connection strength relationship between the co-words of different keywords at different stages can be more clearly displayed, the time series information of the data can be more effectively conveyed, and the potential information can be mined and visualized.

This article refers to the time series text visualization method based on a Sankey diagram. It adopts the probability distribution of the topic-word in the LDA to calculate the similarity between the topic-words and reveals the topic evolution in different stages. First, we determined all the high-frequency feature terms and selected the representative terms of each phase, and then converted them into bow vectors. Second, the bow vectors were converted through the TF-IDF model. Finally, a cosine similarity was used to calculate the topic similarity among adjacent phases. The calculation formula is as follows, where two attribute vectors A and B are given, attribute vectors A and B are usually word frequency vectors in the document, and A i and B i represent the components of the vectors A and B, respectively.

The number of policy-issuing agencies involved in each phase can change, and the type of collaborative decision-making networks can vary in different phases of emergency governance [73, 74] . To explore the structural variation of collaborative decision-making networks, a social network analysis is employed. Based on the data of the policy-issuing agencies involved in each phase, this study constructed co-occurrence matrices of policy-issuing agencies to outline the network structures and then used social network analysis software to draw the collaborative decision-making networks in each phase. The result is shown in Figure 2 . Among them, the blue squares represent policy-issuing institutions, the size of the squares represents the number of times the institution has issued policies, and the lines between the squares indicate that there is a joint policy-issuing relationship between institutions. The network densities of collaborative decision-making in the four phases were 0.3043, 0.1954, 0.2323, and 0.5164, which indicated that the interagency collaborative decision-making network is dynamically changing in the emergency management process.

In the first phase, the core agencies involved are relatively fixed, and they are connected to peripheral agencies to form a relatively stable but discrete network structure. The National Health Commission (NHC), Ministry of Finance (MOF), Ministry of Transport (MOT), Ministry of Commerce (MOC) and Ministry of Human Resources and Social Security (MOHRSS) are clustered in the center of the network, and the distance between the agencies is relatively small. The MOF and National Development and Reform Commission (NDRC) have high betweenness centrality (274.129) and high closeness centrality (29.927 and 29.496, respectively) and have a high eigenvector centrality (0.356), which indicates that these two agencies play a "bridge" role, they have high degrees of independence, and their collaborative partners are relatively important. In the fourth phase, the scale of the involved agencies was further expanded; the network density reached the highest; the involved agencies were diversified; and the collaboration between agencies was in a balanced state. The collaborative decision-making network formed a connected structure. NDRC, MOF, MOHRSS, and MITT occupy the important nodes of the network. Their values of closeness centrality, betweenness centrality, and eigenvector centrality are relatively close.

Overall, as COVID-19 pandemic emergency management developed, the number of policy-issuing agencies expanded from 44 in the first stage to 59 in the last stage, and the links between the agencies became closer and more stable.

On the basis of the above research results, it is found that NHC, MOF, NDRC, MOT and MOHRSS are core participants in the collaborative decision-making network. To further explore the changing role of each agency, this study drew two-dimensional "breadth-depth" matrices based on the degree of nodes and the frequency of links. The degree of the node indicates the number of partners being connected to one single policy-issuing agency, which can be used to measure the "breadth" of collaboration. The higher the degree, the wider the range of collaboration, and the bigger the role the agency played. The frequency of links indicates the frequency of an agency collaborating with other agencies in decision-making. Considering that the frequency of links is affected by the degree of nodes, this study uses the ratio of the frequency of links to the degree of nodes to measure the "depth" of collaboration. Then, four two-dimensional "breadth-depth" matrices in four phases are constructed. The results are shown in Figure 3 . In a two-dimensional matrix, the subjects in the first quadrant and far away from the mean lines are regarded as leading subjects who play the leading role in a collaborative network. As defined, the leading subjects are the agencies who not only collaborated with many other agencies in policy-making, but also participated in collaborations many times. The subjects located in other quadrants but close to the mean lines are regarded as key subjects. As defined, the key subjects are the agencies who either collaborated with many other agencies or participated in collaborations many times. The subjects located in other quadrants and close to the horizontal or vertical axis are regarded as auxiliary subjects. As defined, the auxiliary subjects are agencies who have collaborated with fewer agencies in the decision-making although are rarely also involved in the collaborations. Table 2 shows the evolution of the roles of policy-issuing agencies in the collaborative decision-making network at four stages. Although the structural differences of each agency in the collaborative networks have been identified, we still cannot judge their level of influence. Borrowing the idea from the study made by Wu et al. (2021) [1] , we further analyzed the administrative influence agencies. The leading nodes in the collaborative networks have the strongest influence on the other agencies. The key nodes in the networks have a strong influence on the other agencies. The auxiliary nodes have a weak influence on other agencies. Thus, according to the administrative influence of agencies involved in policy-making in response to COVID-19, the agencies can be categorized into three types-in descending order: leading agencies, key agencies, and auxiliary agencies. Evidently, in the early phase, the breadth and depth of collaborative decision-making networks were relatively low; the scale and scope of cooperative co-governance were relatively small; the strength of collaboration among the agencies was weak; the extensibility was insufficient; and the structure of the collaborative network was relatively unstable. With the strengthening of COVID-19 prevention and control efforts, the number of policy-issuing agencies gradually increased, as well as the breadth and depth of the collaborative decision-making networks. Some agencies evolved into key agencies in the networks. It is worth noting that the leading agencies in the second and third phases were single, and their networks featured multiple agencies led by only one leading agency. In addition, in the early phase, the number of key structural agencies was higher than the number of key functional agencies. In the later phase, however, the number of key functional agencies was on the rise. The form of collaboration gradually evolved from a symbolic style to a complementary style. Collaborative emergency governance had emerged. In the last phase, the number of leading agencies increased significantly, and a clear distinction existed among leading agencies, key agencies, and auxiliary agencies.

The aforementioned single-mode collaborative decision-making network revealed the structural characteristics. However, to obtain knowledge about how agencies participate in different policy topics during different phases of the COVID-19 pandemic emergency management, an "agency-topic" two-mode network needed to be constructed. In detail, this study first drew the document-term matrices by calculating TF-IDF based on 584 jointly issued policy documents as source data and linked agencies to high-frequency feature words, which could be generalized into specific topics. If a co-occurrence existed between policy-issuing agency O and term T, it was recorded as "1"; otherwise, it was recorded as "0". Then, we constructed the agency-term co-occurrence matrices and draw the two-mode "agency-topic" network using Gephi, which is complex network analysis software. In the networks, green nodes represented agencies and red nodes represented topics. Figure 4 shows the topic relationship of the COVID-19 policies jointly issued by the MOF, NHC, and other agencies in the first phase. In the networks, green nodes represented agencies and red nodes represented topics. From the connected keywords, we could determine the work themes. From Figure 4 , we found that the work themes at this stage are mainly related to the emergency prevention and control of the epidemic, and the work themes of each institution vary according to their functions, as shown in Table 3 . Figure 5 shows the topic relationship of the policies combating COVID-19 jointly is-sued by the MOF, NHC and other core agencies in the second phase. The policy topics are mainly related to the resumption of work and production. Table 4 displays the results of Figure 5 in table format. In the networks, green nodes represented agencies and red nodes represented topics. Figure 6 shows the topic relationship of the policies combating COVID-19 jointly is-sued by the MOF, MOA, and other core agencies in the third phase. This stage shows three major work themes, namely, industrial poverty alleviation and poverty alleviation, digital construction and informatization development, and local epidemic prevention and control management. In the networks, green nodes represented agencies and red nodes represented topics. The themes of work of each agency are shown in Table 5 . Figure 7 shows the topic relationship of the policies combating COVID-19 jointly issued by the NHC, MST, and other core agencies in the third phase. Table 6 displays the result of Figure 7 in table format. In the fourth phase, COVID-19 pandemic prevention and control in China had become regular, and government attention had changed to protecting the achievements and recovering economic and social development. It is a remarkable fact that information technology has caught the policy attention of government agencies in this phase. In the networks, green nodes represented agencies and red nodes represented topics. The agencies and their work foci are detailed in Table 6 . 

Based on the topic-term probability distribution in the LDA, we continued to calculate the similarity between topic terms to measure the topic correlation among adjacent phases and reveal the evolution of topic foci across different phases.

The topics are summarized according to topic similarity and high-frequency words. On this basis, a Sankey diagram is drawn to achieve visualization because it is a much more readable way to show the topic evolution in each phase (see Figure 8 ). The Sankey diagram is mainly composed of edges, flows and fulcrums, where the edges represent the flowing data, the flows represent the specific values of the flowing data, and the nodes represent different classifications. The column in Figure 8 indicates the various phases of combating COVID-19; the element blocks in the same column represent the concerning topics in the phase; and the size of the element blocks indicates the degree of similarity between the topic in the current phase and the topics in the next phase, and the rectangular blocks with different colors represent the flow paths of the government agencies' work topics. From observing the whole period of combating COVID-19, NHC and MOF could both clearly be seen in every phase of collaborative emergency decision-making. The policy topic foci of NHCs evolved along the lines of "medical treatment and prevention and control management → pandemic prevention and control during resumption of production and work → regular pandemic prevention and control → digitization of pandemic prevention and control", and their cooperation agencies adjusted with the change in policy focus. In the early phase, the NHC collaborated with the MPS to arrange medical treatment and prevention and control management. Subsequently, the NHC worked with the MOE and MOHRSS to clarify the requirements for pandemic prevention and control in the resumption process of work and production. As combating COVID-19 achieved staged success, NHCs joined forces with SAMR and MPS to regulate market disorder. When COVID-19 prevention and response became routine, the NHC worked with MOIIT and MST to promote the digitalization of pandemic prevention and control.

The policy attention of the MOF, NDRC, and PBC was relatively fixed and mainly focuses on financial supports and guarantees. With the improvements in epidemic prevention and control situations, they have gradually realized the transition from emergency prevention and control to promoting the economic recovery of enterprises and then to the normalization of maintaining stability. In the end, two evolutionary paths were formed. The first path was "Funds guarantee for epidemic prevention and control → Financial preferential policies, enterprise fund guarantee → Poverty alleviation during the epidemic period → Financial support for poverty alleviation", and the second path was "Funding guarantee for pandemic prevention and control → Financial preferential policies, corporate fund guarantee → Digitization of resumption of production and resumption, digitization of industrial chain and supply chain → Financial service digitization, information technology assisting the resumption of work and production".

It is worth noting that although MIIT and MOST did not appear in the first stage, since the second stage, they have set out to promote the integration of information technology into various tasks. Policy foci, such as the digitalization of pandemic prevention and control and the intelligentization of the resumption of work and production, began to appear and then evolved into other policy foci, such as the digitalization of production and sale, the digitalization of the resumption of work and production, and the digital development of industries and the supply chain. In the final phase, the policy focuses converged on the digitization of epidemic prevention and control. Specifically, the digitization of financial services and integration information technology into the resumption of work and production became two major tasks in the later phase.

Through social network analysis and topic evolution path based on the emergency policies against COVID-19 jointly issued by Chinese government agencies, we not only classified the structure of collaborative interagency decision-making networks in combating COVID-19, but also found that an evolution path existed in the dynamic network. The Chinese experience of combating COVID-19 can be summarized into many features, such as building an adaptive public health emergency system [8] . In contrast to existing studies, this study attempted to summarize the Chinese experience of combating COVID-19 from a new perspective, namely, collaborative emergency management. For China itself, the experience of drawing from the government's efforts against COVID-19 is beneficial to emergency public health management in the future. For other countries suffering as a result of COVID-19, the Chinese experience could also provide several valuable and feasible tactics to contain pandemics, and help with more rapid recovery.

First, the government should pay different attention to different issues according to changing environments. As we know, government attention is a limited resource when coping with public health crises [8] . This means that only some important issues will be addressed in the policy-making agenda. Moreover, the dynamic evolution of public health crises determines that governments would suffer from different problems in different phases. It requires the government to pool limited resources to solve urgent problems within a specific time period. Take combating COVID-19 in China as an example: medical treatment was the top priority in the early phases, which required health-related agencies to play the leading role. Instead, economic recovery became the primary task in the later phases, and economic-related agencies took over leading responsibilities. In summary, emergency public health management needs multiple government agencies to serve in different roles to come up with specific policy implementation plans.

Second, the government should balance the attention given to COVID-19 control and economic recovery. In practice, some countries attached too much importance to the prevention and control of COVID-19, resulting in a serious impact on economic development. In contrast, some countries put too much emphasis on economic recovery while ignoring the prevention and control of COVID-19, which resulted in substantial increases in the COVID-19 infection rate and death rate. The experience in China showed that we should balance the attention given to COVID-19 control and economic recovery. Once new local cases are found to spread, the government should respond quickly and implement stricter prevention and control measures, such as quarantines and lockdowns. As long as the spread is under control, the government should set out to return people's lives to normal and promote the resumption of work and production.

Third, a connected collaborative decision-making network is needed in public health emergencies. Collaboration is the basic requirement for emergency decision-making. Government agencies should cooperate in a close network. As we know, different agencies have different responsibilities and professional abilities. Although collaborative interagency decision-making networks exist in many countries suffering from the COVID-19 pandemic, their structures are always somewhat loose. As for China, the emergency collaborative decision-making network was also discrete in structure in the early phases; however, the network structure evolved into a connected form in the later phases, which meant there were many government agencies acting as key moderators to connect all the policy-issuing agencies together. In other words, it not only requires multiple actors to participate in the decision-making process in public health emergencies, but also needs some government agencies to shoulder coordination responsibilities.

Finally, shared leadership at the national level needs to be gradually switched and expanded during crisis management. From China's experience of combating COVID-19, the collaborative interagency decision-making network was switched to adapt to the changing situation, which makes us reflect on the leadership in crisis management. Evidently, adaptive leadership is needed in combating public health crises. Take the COVID-19 pandemic as an example: it has brought about many wicked challenges. Some of them have never happened before, and require a rapid and comprehensive government response. Nevertheless, the emergency leadership in most countries would invariable so that they would suffer from a lack of flexibility. In contrast, the emergency decision-making network in China exhibits adaptability, which contributes to building leadership competency. There is an argument that leadership competency has a positive relationship with the effectiveness of crisis management [75] . China's experience shows that adaptive leadership in crisis management plays a crucial role in combating COVID-19, and significantly affects the effectiveness of emergency plans.

The present study has several limitations. First, the research sample of this study was limited to the Chinese central government level. The response to major public health emergencies not only depends on the horizontal collaborative decision-making of central agencies, but also depends on the vertical collaborative work of central and local agencies at all levels. This study ignored the collaborative relationship between local governments and the central government in coordinating the COVID-19 response.

Second, although policy document analysis is an effective method to identify agency collaboration, the sudden characteristics of COVID-19 have determined the real-time and dynamic nature of collaborative decision-making across agencies. It only takes the joint documents published by the central government regularly as objects of analysis; thus, it will ignore the real-time characteristics of multi-sectoral collaborative decision-making in the process of responding to COVID-19. At the same time, other important factors in the collaborative response of various agencies will also be ignored.

Third, in this study, the LDA model was used to extract the topics of policy texts jointly issued by governments, whereas more powerful semantic mining models, such as Word2Vec and LDA2evc, have not been applied. On the one hand, the effect of topic mining needs to be improved; on the other hand, due to the limitations of the LDA clustering algorithm, the topic extraction of policy text has strong subjectivity.

In follow-up studies, collaborations across different levels and different factors are also notable for inquiry. As for research methods, subsequent studies could consider employing other text mining techniques, such as Word2Vec and LDA2evc.

Professionalism and adaptive governance capacity are embodied in the Chinese governmental effort to combat COVID-19. Chinese emergency public health management features unified commands, adaptive responses, and collaborative governance. China's experience helps to shed light on the necessity of interagency collaborative governance in public health emergencies. It is crucial to introduce the collaborative governance concept to the policy-making process, as well as in the policy implementation process. Collaborative governance in emergency management requires government agencies to clarify their own responsibilities and labor division to formulate an effective joint prevention and control mechanism. In responding to the COVID-19 pandemic, Chinese government agencies quickly integrated resources and established collaborative interagency decision-making networks that varied throughout different phases of emergency pandemic management.

This study explored the dynamic interagency collaborative decision-making network, characterized structural variation and agency role change, and identified the evolution of policy topics for each agency through policy document analysis. The results show that there are three forms of network structure in the emergency process: discrete structure in the early phase, subgroup structure in the middle phase, and connected structure in the later phase. Agencies embedded in the network can be categorized into three types: leading agencies, key agencies, and auxiliary agencies. Furthermore, agencies each have their own primary policy focus, but share some common foci across all four phases and shift their attention in the emergency management process. These research findings can provide implications for understanding collaborative decision-making in public health emergencies by summarizing the Chinese experience of combating COVID-19. Understanding the dynamic evolution of collaborative decision-making networks can also help to obtain knowledge about how to improve emergency management ability when facing public health emergencies. Institutional Review Board Statement: Ethical review and approval were waived for this study, because all data are available in the public domain.

The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy restrictions.

The authors would like to thank all the anonymous reviewers for their constructive comments regarding this article.

The authors declare no conflict of interest. 

Characterizing the patterns of China's policies against COVID-19: A bibliometric study

Crisis decision-making on a global scale: Transition from cognition to collective action under threat of COVID-19

Preparedness for emergency response: Guidelines for the emergency planning process

Disaster management: Findings from a systematic review

Collaborative decision-making in emergency and disaster management

What have we learned since September 11, 2001? A network study of the Boston marathon bombings response

Policy design and state capacity in the COVID-19 emergency in Italy: If you are not prepared for the (un)expected, you can be only what you already are

Understanding the evolution of government attention in response to COVID-19 in China: A topic modeling approach

Crisis governance, Chinese style: Distinctive features of China's response to the Covid-19 pandemic

Multilevel responses to risks, shocks and pandemics: Lessons from the evolving Chinese governance model

Collaborative emergency management: Effectiveness of emergency management networks

Decision making in emergencies

A case study of decision making in emergencies

Understanding decision making during emergencies: A key contributor to resilience

Organizational communication and decision making for hurricane emergencies

Policy change analysis based on "Policy Target-Policy Instrument" patterns: A case study of China's nuclear energy policy

Quantitative research on policy literature

Lessons learned from a policy decision to coordinate a multijurisdiction H1N1 response with a single incident management team

Pandemic H1N1 in Canada and the use of evidence in developing public health policies-A policy analysis

Mobilizing policy (in)capacity to fight COVID-19: Understanding variations in state responses

Policy capacity and Singapore's response to the COVID-19 pandemic

Policymaking in a low-trust state: Legitimacy, state capacity, and responses to COVID-19 in Hong Kong

Collaborative emergency management and national emergency management network

Collaboration and leadership for effective emergency management

Codifying a crisis: Progressing from information sharing to distributed decision-making. J. Contingencies Crisis Manag

Inter-organizational coordination in extreme events: The world trade center attacks

Managing transboundary crises: Identifying the building blocks of an effective response system

Wicked problems, knowledge challenges, and collaborative capacity builders in network settings

Collaborative distributed decision making for large scale disaster relief operations: Drawing analogies from robust natural systems

Interagency communication networks during emergencies: Boundary spanners in multiagency coordination

Collective decision-making for developing emergency management capabilities

Understanding the collaborative process and its effects on perceived outcomes during emergency response in China: From perspectives of local government sectors

Decision making application in collaborative emergency response: A new PROMETHEE preference function

Luke) supporting collaborative sense-making in emergency management through geo-visualization

Improving emergency response training and decision making using a collaborative virtual reality environment for building evacuation

Construction of collaborative mapping engine for dynamic disaster and emergency response

Knowledge Management Support for Collaborative Emergency Response

Towards the development of a decision support system for multi-agency decision-making during cross-border emergencies

Collaborative decision making in sustainable flood risk management: A socio-technical approach and tools for participatory governance

Collaborative Decision Making: Perspectives and Challenges

Exploring agent-based modeling for emergency logistics collaborative decision making

Study on emergency decision-making of natural disaster based on collaboration of multi-agent

Modeling of emergency response decision-making process using stochastic petri net: An e-service perspective

Evaluating network arrangements: Toward revised performance measures. Public Perform

Information communication technology utilization for effective emergency management networks

Sustaining networks in emergency management. Public Perform

Task-adaptive information distribution for dynamic collaborative response

Power and cognitive accuracy in local emergency management networks

When practice matters more than government plans: A network analysis of local emergency management

Measuring performance for collaborative public management using network analysis methods and tools. Public Perform

Examining intergovernmental and interorganizational response to catastrophic disasters: Toward a network-centered approach

Designing, managing, and sustaining functionally collaborative emergency management networks

Structure and network performance: Horizontal and vertical networks in emergency management

Structural dynamics of organizations during the evolution of interorganizational networks in disaster response

The core and periphery of emergency management networks

A bibliometric study of China's science and technology policies

Measurement on the policy diffusion: Evidence from China's policy documents

Quantitative evaluation of China's cultivated land protection policies based on the PMC-Index model

Document analysis as a qualitative research method

Closing the policy-practice gaps in Nigeria's desertification interventions: A qualitative document analysis of sustainable practice

Health equity lens embedded in the public health policies of Saudi Arabia: A qualitative document analysis

Equity in public health standards: A qualitative document analysis of policies from two Canadian provinces

The evolution of smoke-free spaces policy literature: A bibliometric analysis

Integrating planetary health into healthcare: A document analysis

US state-level policy responses to the Ebola outbreak

Understanding the pre-medical emergency team tier of a mature rapid response system: A content analysis of guidance documents

Comparative analysis of COVID-19 guidelines from six countries: A qualitative study on the US

Responses to COVID-19 in five Latin American countries

Protection of children in difficulty in China during the COVID-19 pandemic

Migrant carers in europe in times of COVID-19: A call to action for European health workforce governance and a public health approach

Latent dirichlet allocation

Federalism, intergovernmental relationships, and emergency response: A comparison of Australia and the United States

Understanding multiplexity of collaborative emergency management networks

Collaborative crisis management and leadership in the public sector