Article Text

Network analysis of stroke systems of care in Korea
  1. Jihoon Kang1,
  2. Hyunjoo Song2,
  3. Seong Eun Kim1,
  4. Jun Yup Kim3,
  5. Hong-Kyun Park4,
  6. Yong-Jin Cho5,
  7. Kyung Bok Lee6,
  8. Juneyoung Lee7,
  9. Ji Sung Lee8,
  10. Ah Rum Choi9,
  11. Mi Yeon Kang9,
  12. Philip B Gorelick10 and
  13. Hee-Joon Bae3
  1. 1Neurology, Seoul National University Bundang Hospital, Seongnam, Korea (the Republic of)
  2. 2School of Computer Science and Engineering, Soongsil University, Seoul, Korea (the Republic of)
  3. 3Neurology, Seoul National University Bundang Hospital, Seongnam, Gyeonggi-do, Korea (the Republic of)
  4. 4Neurology, Inje University Ilsan Paik Hospital, Goyang, Korea (the Republic of), Korea (the Republic of)
  5. 5Neurology, Inje University Ilsan Paik Hospital, Goyang, Korea (the Republic of)
  6. 6Neurology, Soonchunhyang University Hospital, Yongsan-gu, Seoul, Korea (the Republic of)
  7. 7Biostatistics, Korea University School of Medicine, Seoul, Korea (the Republic of)
  8. 8Clinical Research Center, Asan Institute for Life Sciences, Asan Medical Center, Seoul, Korea (the Republic of)
  9. 9Health Insurance Review & Assessment Service, Wonju, Korea (the Republic of)
  10. 10Neurology, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA
  1. Correspondence to Dr Hee-Joon Bae; braindoc{at}


Background The landscape of stroke care has shifted from stand-alone hospitals to cooperative networks among hospitals. Despite the importance of these networks, limited information exists on their characteristics and functional attributes.

Methods We extracted patient-level data on acute stroke care and hospital connectivity by integrating national stroke audit data with reimbursement claims data. We then used this information to transform interhospital transfers into a network framework, where hospitals were designated as nodes and transfers as edges. Using the Louvain algorithm, we grouped densely connected hospitals into distinct stroke care communities. The quality and characteristics in given stroke communities were analysed, and their distinct types were derived using network parameters. The clinical implications of this network model were also explored.

Results Over 6 months, 19 113 patients with acute ischaemic stroke initially presented to 1009 hospitals, with 3114 (16.3%) transferred to 246 stroke care hospitals. These connected hospitals formed 93 communities, with a median of 9 hospitals treating a median of 201 patients. Derived communities demonstrated a modularity of 0.904, indicating a strong community structure, highly centralised around one or two hubs. Three distinct types of structures were identified: single-hub (n=60), double-hub (n=22) and hubless systems (n=11). The endovascular treatment rate was highest in double-hub systems, followed by single-hub systems, and was almost zero in hubless systems. The hubless communities were characterised by lower patient volumes, fewer hospitals, no hub hospital and no stroke unit.

Conclusions This network analysis could quantify the national stroke care system and point out areas where the organisation and functionality of acute stroke care could be improved.


Data availability statement

Data is available with permission from the Korean Health Insurance Review and Assessment Service (HIRA). As this study used a government-owned database, it requires special permission to use it.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • Stroke care has recently been evolving towards forming collaborative networks among hospitals, but efficiencies have not been evaluated.


  • A total of 1009 nationwide hospitals involved in acute stroke care were clustered into 93 functional network units, most of which were centralised linkages towards single or double hubs.

  • While single-hub and double-hub systems presented similar efficacies in patient transfers, endovascular treatment and mortalities, it could identify the hubless system that was vulnerable to the configuration of hospital networks and near-zero endovascular treatment performance.


  • This could be used to take an assessment and renovate local stroke systems in terms of the regional hospital organisations.


Advancements in stroke treatment, such as endovascular treatment (EVT), have triggered changes in stroke care systems, expanding access to advanced treatments.1 This shift has transformed stroke care from a single hospital model to networks of hospitals that work together to provide comprehensive acute stroke treatments.2 3 These networks have been established through prehospital triage restructuring, interhospital transfer enhancement and the formation of regional stroke networks.4 5

While stroke networks have evolved in various forms over the past decade,6–8 the diversity among them may result in uneven functioning.9 10 Understanding the characteristics and functionality of stroke networks is essential to optimise those imbalanced situations by restructuring the delivery of acute stroke care.11 In Korea, there are over a thousand hospitals where patients seek treatment for acute stroke. However, only a small fraction of these hospitals provide acute stroke care; even fewer are accredited stroke centres or offer EVT.12 13 The Korean hospital system, historically, has not implemented an official tier system among hospitals, leading to significant regional disparities in accessing acute stroke treatments like intravenous thrombolysis (IVT), EVT, decompressive surgery and stroke unit (SU) care. This lack of a structured system often necessitates interhospital transfers, typically arranged through individual contacts rather than a formal network.14 Recognising the need for more coordinated stroke care, the Korean government has recently initiated a pilot project. This project aims to establish predefined stroke networks, incentivising their use through preferential reimbursement policies.

To address this issue, our study applies contemporary network science techniques, offering a more comprehensive perspective than traditional analyses in epidemiology or clinical research.15–17 These techniques acknowledge the interdependencies within the hospital network, allowing us to capture and understand the structure of interactions among hospitals. In our analysis, we have transformed these interactions into nodes (representing hospitals) and edges (representing interhospital transfers), consistent with the standard format in network science. Furthermore, we have grouped densely connected hospitals as a ‘stroke community’ using a popular community detection method. The quality and characteristics of these derived stroke communities were assessed using a variety of measures from network science. To comprehensively evaluate the impact of these stroke communities, we chose to focus on two critical response variables: EVT rates and 1 year mortality. This dual focus is particularly relevant in network analysis as it encapsulates both the effectiveness of clinical care (1 year mortality) and the diffusion of innovation within the network (EVT rates).


Study subjects and data collection

This study used data from the Acute Stroke Quality Assessment Program (ASQAP), a national stroke audit programme administered by the Korea Health Insurance Review and Assessment Service (HIRA).13 The ASQAP aims to evaluate the quality of acute stroke care provided by stroke care hospitals (SCHs) in Korea. All individuals admitted to SCHs through the emergency room within 7 days of stroke onset were included in the ASQAP. For this study, we analysed data from patients with acute ischaemic stroke (AIS) obtained from the seventh ASQAP assessment. To be eligible for the seventh assessment, SCHs were required to have treated 10 or more patients with acute stroke between 1 July and 31 December 2016.

Demographics, time intervals from onset to arrival, interhospital transfers, baseline stroke severity measured by National Institutes of Health Stroke Scale (NIHSS), administration of IVT, and functional outcome at discharge were obtained from the ASQAP database.12 In addition, hospital size, number of neurologists and provision of SU care at SCHs were extracted from the ASQAP database. To gather information on hospital connections during the acute stage of stroke, administration of EVT and mortality, the ASQAP database was linked with reimbursement claims and death certificate data. Since South Korea has a single-payer system where the National Health Insurance Service and complementary medical aid programme cover the whole national population and reimbursement claims from any type of hospital,18–20 this linkage provided us with comprehensive data. During the data collection process, interhospital transfers were defined as sequential hospitalisations of the same patient at two different facilities, with the discharge day of the sending hospital being the same or 1 day earlier than the admission date of the receiving hospital.21 This enabled the identification of small-sized or medium-sized hospitals or clinics other than SCHs, which were operationally defined as adjacent facilities.

Study protocol approval and ethical statements

In compliance with the Act on the Protection of Personal Information, the HIRA provided patient and hospital data without any individual identifiers under the Joint Project on Quality Assessment Research, thus ensuring the study’s compliance with all applicable privacy regulations.


Establishing the interhospital connection database

The process involved compiling initial hospital visits and subsequent transfers, thereby classifying patient transitions into direct visits/stays, as well as sending and receiving actions. Patient transitions were then transformed into nodes (representing hospitals) and edges (representing interhospital transfers), which detailed both the direction and volume (weight) of these transfers.

Applying network analysis to identify stroke communities

Interhospital transfers, symbolised by the direction and weight of edges, allowed us to cluster densely linked hospitals heuristically. Our goal was to maximise the volume of interhospital transfers within a given cluster while minimising transfers between hospitals in different clusters. This was achieved using the Louvain algorithm,22 which maximises the modularity of a partition. Modularity quantifies the difference in the density of internal links between a clustered network and a randomly connected network with the same clusters (ie, a modularity of 1 implies an upper limit, whereas 0 indicates no significant difference from a random network). The algorithm initially assigns each node (representing a hospital) to a separate cluster. In the first step, the algorithm finds the maximum modularity gain from reassigning each node to one of its neighbouring clusters and relocates it only if the gain is positive. It then merges all nodes in the same cluster into a single node and repeats the first step until there are no more changes in modularity.23 A more detailed explanation of the methods employed is provided in online supplemental box 1.

Supplemental material

Characterising the derived stroke communities

We used the modularity value Q to evaluate the overall quality of the derived stroke communities. In addition, to assess the quality and capacity of each community, we used three key density metrics24 (global density (K), mean intracommunity density (Embedded Image), and mean intercommunity density (Embedded Image) (online supplemental table 1). K measures the overall connectivity in the network, calculated as the ratio of the actual number of connections (edges) to the maximum possible number. Embedded Image and Embedded Image assess the connectivity among nodes within individual communities and between different communities, respectively, both using a similar ratio-based approach. According to empirical studies, a higher Embedded Imagecompared with a K, and a lower Embedded Image compared with a K, indicate well-defined, high-quality communities. This means that the average connections within each community are more numerous than those between different communities, signifying distinct and well-integrated stroke care communities within the network.25

Stroke communities were characterized based on patient-level and hospital-level attributes and network metrics.26 We evaluated the capacity of each stroke community using the number of direct visits/stays, interhospital transfers, recanalisation therapies, bed size and provision of SU care. Degree centrality, group degree centralization (GDC), and the abovementioned three density metrics were employed as network metrics.

Degree centrality is a measure used in network science to assess the importance of each node within a network. As the network has directed edges, we summed up the number of incoming links for each node. Then, we divided it by the number of possible connections within the belonging community. GDC measures the level of centralisation in each community. It first calculates the difference between each node’s degree centrality and the theoretical maximum in the belonging community. It then divides these values by the theoretical maximum and sums them up. A value close to 1 indicates high centralisation and a value close to 0 indicates low centralisation. A more detailed explanation for calculating GDC is provided in online supplemental table 2.

Clinical interpretation of the network model

We classified the communities by visually inspecting a diagram that schematically represents the network topology. We visualised each community as a node-link diagram that places nodes with a higher degree of centrality in the centre. A node with a degree centrality higher than 0.5, 0.3 and the rest were heuristically designated as a hub, an auxiliary hub and a spoke, respectively. Subsequently, based on their topology and the number of hubs in the community, we categorised them as single-hub, double-hub and hubless communities.

We made comparisons among the abovementioned three community categories based on patient-level and hospital-level attributes, including transfer rates within and between communities. EVT rates and 1 year mortality were also compared as outcome variables. We obtained complete 1 year mortality data by linking with the death certificate data from the National Statistics Office of Korea and used this measure as a surrogate for functional outcomes.27 28 Network analysis was conducted using SQL-based in-house programmes, while the statistical analysis was performed using SAS Enterprise Guide V.7.1 (SAS Institute, Cary, North Carolina, USA).


Over the 6-month study period, a total of 19 113 patients suffering from AIS were admitted to SCHs. The ASQAP evaluated the quality of acute stroke care delivered to these patients, which is analysed in this study. In the immediate days following stroke onset, patients primarily sought care at 1009 hospitals, which included 246 SCHs and 763 adjacent facilities. Among these, 1831 patients (9.6%) were transferred from adjacent facilities to SCHs, while 1283 (6.7%) were transferred from SCHs to other SCHs. Ultimately, 2900 patients (15.2%) were transferred to the EVT-capable SCHs, SCHs that performed at least one EVT procedure for patients with AIS during the 6-month study period.

The median (IQR) of patient admission volume per SCH was 33.5 (14–124) patients for the 6-month study period. The distribution was skewed to the right, with the highest 30 SCHs (12.2%) treating 30 or more patients per month, whereas the lowest 109 SCHs (44.3%) treated less than 4 patients per month (figure 1 and table 1).

Figure 1

The distribution of AIS patient visit volume per hospital (y-axis on the left side) and the sending and receiving proportions (y-axis on the right side) in 246 SCHs. The stack bar graph represents the number of patients who were classified into direct-visit-and-stay (blue sky), sending (orange) and receiving (blue) in each SCH. The sending and receiving proportions in each SCH are overlaid on the red and blue lines, respectively. See the footnote of table 2 for the definition of patient visit volume. AIS, acute ischaemic stroke; SCH, stroke care hospital.

Table 1

Baseline characteristics of study subjects (n=19 113) and SCHs (n=246)

In examining the sending and receiving tendencies of each SCH, we found that the median values of the receiving and sending proportions regarding AIS were 8.3% and 9.7%, respectively. This differentiation is vividly represented in the encoded stacked bar graph (figure 1), which illustrates the divergent roles of SCHs according to their patient volumes. High-volume centres demonstrated a greater likelihood of receiving patients from other hospitals, whereas low-volume centres were more likely to transfer patients to other hospitals. When we examined the relationships between patient volume and the receiving and sending proportions, we observed a Spearman correlation coefficient of 0.61 (p<0.001) for the receiving proportion and −0.70 (p<0.001) for the sending proportion.

The Louvain algorithm obtained 93 densely linked stroke communities, displaying modularity (Embedded Image), indicative of a strong community structure. The use of weighted and directional edge information elevated the modularity from 0.35 (a value derived when the communities were clustered only considering the existence of hospital links, or in other words, using unweighted and unidirectional edge information) to 0.90. Moreover, the derived stroke communities satisfied the condition for high-quality clustering; Embedded Image (0.312) exceeded Embedded Image, and Embedded Image was less than K.

The 93 communities consisted of a median of nine hospitals, including two SCHs, and a median of 201 patients (mean±SD, 218±152) visited each community over a span of 6 months. Of these patients, 82.7% remained at their initial hospitals, 5.7% were transferred to hospitals within their community, 5.1% were transferred outside their own community and 5.2% were transferred from other communities (table 2). The median EVT rate per community was 5.9%, and the median 1 year mortality per community was 12.5%. There was a strong correlation between patient admission volume and other volumetric parameters, such as transfer volume, IVT volume and EVT volume (r>0.8, p<0.001, (online supplemental figure 1).

Table 2

Characteristics of derived stroke communities (n=93)

The use of geographical mapping afforded us unique insights into the size of the communities. It was observed that the longest diameter of the majority of these communities was typically less than 130 km in length. However, in approximately 14.0% of these communities, it extended beyond 200 km, with some reaching as far as 354.3 km (figure 2 and online supplemental figure 2).

Figure 2

Geographical visualisation of stroke communities (A) with presumed boundaries (B) and 11 communities with a hubless system (C). The Louvain algorithm using weighted-and-directional connecting information partitioned 93 stroke communities represented by different coloured nodes (hospitals) with the thickness of edges (interhospital transfers and numbers) on the map (A). The boundaries of the communities visualised the extent of individual communities and overlapped territories among adjacent communities (B). In addition, 11 communities with hubless systems were depicted (C), which had no hub-and-spoke system (n=8), or only one hospital with no connection to other hospitals (n=3).

The structure of most of these communities suggests the formation of highly centralised networks, often resembling a spindle shape. Such networks typically revolve around one or two major hospitals (hubs), which are interconnected with adjacent hospitals (spokes). Hubs received a median of 21 (IQR, 9–36) transfers, which corresponded to a median of 97.4% (IQR, 84.3%–100.0%) of total community transfers.

We measured the GDC to quantify the degree of centralisation within 93 individual communities. Results indicate a high degree of centralisation, as evidenced by a median GDC value of 0.99. This finding implies that patient transfers within most communities were heavily concentrated in a singular hub hospital.

Our study used schematic illustrations to elaborate on communities, revealing nodes and edges within each. The communities were arranged according to their GDC values, with an emphasis on their hub hospitals. This arrangement allows us to clearly depict the inherent network structures within these communities (figure 3). This diagram notably displayed close connections centralised around one or two hub hospitals integral to the transportation system. The hubs and auxiliary hubs were defined as nodes with degree centrality >0.5 and >0.3, respectively (online supplemental figure 3).

Figure 3

The network structure of stroke communities arrayed by group degree centralisation (A) and grouped by communities’ characteristics (B). According to the number of connected nodes and degree centrality, individual nodes (hospital) were coloured: red >0.5 of degree centrality and yellow 0.3–0.5 with two or more edges, while others (blue, ≤0.3 of degree centrality or single edge). The configuration of communities was classified into single-hub (green), double-hub (orange) and hubless (grey) systems (B). GDC, group degree centralisation.

According to the GDC values, the number of hubs and the discernible patterns, we classified the stroke communities into three distinct types: single-hub, double-hub and hubless systems (figure 3B and online supplemental figure 4). Age, baseline NIHSS Score, onset-to-arrival time and usage of emergency medical services (EMS) were found to be consistent across all three community types (table 3). The double-hub communities numerically had larger patient volumes, a greater number of hospitals and neurologists, and higher intracommunity transfer rates than the single-hub system. The EVT rate was highest in the double-hub system, followed by the single-hub system, and was almost zero in hubless systems. The 1 year mortality did not show any significant differences across these three types. The hubless systems, characterised by lower patient volume and fewer hospitals, did not feature a hub hospital. This absence of a hub hospital might result in no identifiable SUs, infrequent interhospital transfers and near-zero rates of EVT.

Table 3

Comparisons between three categories of stroke communities according to their network structures


This study used contemporary network science techniques to illustrate the network structure of stroke care systems across Korea. We analysed 93 stroke communities, which were derived from 1009 hospitals involved in acute stroke care. These hospitals serve a nation of approximately 50 million inhabitants residing within an area of around 10 000 km2.

Our analysis revealed that most of these stroke communities were highly centralised networks, primarily consisting of one-hub or two-hub hospitals. However, there were also a few communities that were poorly structured. Such inadequately formed communities might not be suitably equipped to provide advanced stroke treatment.

This study identified that 16.3% of patients were transferred from their first-visit hospitals to other medical facilities for acute stroke care. Over half of these interhospital transfers were from non-SCHs to SCHs, while the remaining transfers occurred between SCHs. These transfers were mostly necessitated by disparities in the transferring hospital’s capacity to deliver higher levels of stroke care, deemed crucial in certain cases. Hospitals with a lower patient volume were more likely to refer patients to other facilities (figure 1).29 30

Interhospital transfer between SCHs may be unavoidable and sometimes essential. This is because it is not feasible for all SCHs to provide the highest levels of stroke care around the clock and throughout the year. However, transfers from non-SCHs to SCHs could be avoided. There is no reason to spend time in non-SCHs if transfers to SCHs are required. This situation could be improved through education to inform the importance of direct access to SCHs and the reorganisation of regional stroke systems to facilitate the direct dispatch of patients to SCHs via EMS. Enhancing such an intracommunity patient transportation process, with the support of EMS, could reduce intercommunity transfers, leading to a more efficient system.

Our study revealed that the median annual patient volume per SCH in Korea was 67 for AIS. This figure is lower than those reported in Japan (185 for AIS in 2015) and Denmark (330 for all types of strokes between 2003 and 2009). However, it is higher than that in Ontario, Canada, where the median annual patient volume per SCH was 28 for AIS between 2005 and 2012.31–33 The distribution of volume of patients who had a stroke per SCH shows a right-skewness in both Korea and Canada, where over half of the hospitals provide acute stroke care to one or fewer patients per week.33 Yet, it is essential to consider the gross geographical differences: Canada’s land area is roughly 100 times larger than South Korea’s. This vast geographical spread in Canada can present significant challenges in rapidly transferring patients with AIS to SCHs, contributing to the right-skewness of patient volume distribution. In contrast, South Korea’s smaller geographical size, combined with adequate identification and planning, could facilitate more effective patient transfers. As a result, avoiding the skewness observed in the distribution of volume of patients who had a stroke might be possible.

The Louvain algorithm employed in our study led to the identification of 93 communities, with a median of nine hospitals, including one or two hubs. These networks served as functional units providing acute stroke care in South Korea. The median GDC was 0.99, indicating that most derived communities were highly centralised. Moreover, the quality of the clusters estimated by the modularity (Q=0.90) would represent as notably high, indicating an overall high quality of clustering. The comparisons of the three density measures affirmed this high-quality clustering.

In our analysis, we observed a median intercommunity transfer rate of 5.7% (ranging from 3.9% to 9.1%), which might indicate either the incompleteness of stroke networks or potential overlaps between communities. This observation is particularly relevant given the constraints of the Louvain algorithm, which is designed to identify non-overlapping communities. This inherent limitation of the Louvain method highlights the need for more advanced community detection algorithms, such as Network Decomposition Overlapping Community Detection, to more accurately reflect the complex realities of healthcare networks.34

The hub hospitals were mostly equipped with SUs and capacitated to provide EVT. Among the three types of community structures, the single-hub and double-hub systems were largely comparable in many aspects. However, the hubless system drew our attention regarding lower patient volumes, no hub, no SU, lower transfer rates and a near-zero EVT rate. Accordingly, we could characterise the relatively low quality of stroke care in these hubless communities. Also, the urban location of these hubless communities underscores the importance of interhospital connectivity, surpassing the consideration of geographical proximity alone (figure 2C).

Each community in our study attended to approximately 200 patients with AIS over the 6-month study period. Comparatively, this patient volume per community is lower than that observed in a Comprehensive Stroke Center in the USA (~ 431 AIS).35 Furthermore, nearly 40% of the communities in our study lacked SUs, and 48% reported an annual EVT case volume of fewer than 15. This figure is below the threshold case volume required to qualify as a Thrombectomy-Capable Stroke Center in the USA,36 and inconsistent with the acknowledgement that a higher case volume correlates with improved quality of stroke care.37 38 In addition, the variability in community size within South Korea is noteworthy. The longest diameter was less than 140 km in 86% of communities, extending over 200 km in 14% of communities. Smaller communities in close proximity to larger ones present an opportunity for reorganising the regional stroke systems in South Korea, thereby potentially enhancing the quality of acute stroke care.

In South Korea, the healthcare market is predominantly private and highly competitive, with minimal restrictions on accessing higher-level hospital services, even in emergency stroke care. This situation is unique compared with other countries in the Organization for Economic Co-operation and Development,39 40 as evidenced by the lower usage of EMS in our study. This unrestricted access has led to over 1000 hospitals providing acute stroke care.7 41 In response to this fragmented landscape, HIRA has recently introduced accreditation for stroke centres as part of the ASQAP. Additionally, the Ministry of Health and Welfare has begun a pilot project for stroke networks. Our study’s findings could be instrumental in supporting and shaping these new initiatives.

The study has several limitations. First, the retrospective analysis of national audit data, an administrative data set, limits the comprehensiveness of the study data set. For example, the data set lacks detailed information regarding the rationale for interhospital transfers, time metrics relating to interhospital transfers and EVT, as well as the 3-month functional outcomes.

Second, we acknowledge a significant discrepancy in the number of ischaemic stroke cases in the ASQAP data set compared with the expected figures based on a recent statistical report from South Korea.42 This variance could be partially attributed to the exclusion criteria of the ASQAP, which omits data from smaller hospitals (those with fewer than 10 patients during the 6-month study period) and cases not admitted through emergency departments. This exclusion potentially under-represents the total number of stroke cases. While leveraging the national claims data using International Classification of Diseases (ICD) codes could provide a broader scope of case identification, this approach has its drawbacks. Specifically, the accuracy of stroke case identification using ICD codes is not fully validated, and differentiating acute from chronic stroke events remains a challenge. Recent studies have indicated that only about 30% of cases are accurately classified as stroke when using ICD codes alone.43

Our third point of consideration is that this retrospective analysis limited our ability to detail patient dispositions among hospitals under specific clinical or regional conditions. Additionally, we didn't account for emerging connectivity aspects like telehealth, which are increasingly crucial in healthcare. Recognising these gaps, we suggest future research should employ advanced methods like the Exponential Random Graph Model to explore these dispositions and connectivity facets in-depth.44 This approach would provide a more comprehensive understanding of healthcare networks and align with the evolving digital landscape in healthcare delivery.

Fourth, we acknowledge that the choice of 2016 as a study period, selected due to its proximity to the pivotal EVT trials of 2015,45 and the ensuing reorganisation of regional stroke networks, may not fully capture the subsequent advancements in stroke care. Since then, it is likely that the infrastructure and implementation strategies for EVT in Korea, as in many other countries, have undergone significant evolution. This context should be considered when interpreting our findings. Future research may benefit from examining more recent data to provide insights into the current state of EVT utilisation and its outcomes. Lastly, the study subjects were limited to those admitted with a final diagnosis of ischaemic stroke. For more reliable network analysis, the next studies would warrant enrolling patients who were initially suspected of having a stroke.

In conclusion, our study results have significant practical implications for the organisation and delivery of acute stroke care. Our network analysis points out areas where the current geographical organisation and functionality of AIS care in Korea could be improved. Specific areas for improvement include addressing the right-skewness of patient volume per community, consolidating the smaller and more numerous communities, and rectifying the issues of hubless, small adjacent and excessively lengthy communities. These findings suggest that the regional stroke systems in Korea could be reorganised to optimise patient transportation to appropriate hospitals providing higher-level stroke care.

Data availability statement

Data is available with permission from the Korean Health Insurance Review and Assessment Service (HIRA). As this study used a government-owned database, it requires special permission to use it.

Ethics statements

Patient consent for publication

Ethics approval

The Institutional Review Board of HIRA approved the study protocol, which waived the informed consent due to the study’s retrospective nature and the use of anonymised data (registered number, X-2202-736-902).


The authors thank the National Health Insurance Service of Korea, for the opportunity to analyse the databases (NHIS-2015-4-022-1).


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Contributors Study design and conceptualisation: JK, H-JB. First draft: JK. Investigation: HJS, SK JK. Data acquisition: AC and MYK. Data acquisition and data analysis: JK, H-KP, Y-JC, KBL, JSL, JK. Statistics: SK, JL. Draft edit and review: H-JB, PG, JK. Study Guarantor: H-JB.

  • Funding This research was supported by a fund (2023-ER0903-00) by the Korea Centers for Disease Control and Prevention.

  • Map disclaimer The depiction of boundaries on this map does not imply the expression of any opinion whatsoever on the part of BMJ (or any member of its group) concerning the legal status of any country, territory, jurisdiction or area or of its authorities. This map is provided without any warranty of any kind, either express or implied.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.