Theme trends in research related to retinal vein occlusion: a quantitative and co-word analysis

Background. This study focused on plotting knowledge structure and exploring research hotspots of retinal vein occlusion (RVO). Methods. In this study, research articles, with subject of RVO, were acquired from PubMed. Bibliographic Item Co-Occurrence Matrix Builder (BICOMB) was used for MeSH terms acquisition, evaluation and high-frequency MeSH term determination. Biclustering analysis and knowledge structure were conducted based on the MeSH term-source article matrix. RVO theme trends were illustrated with social network analysis (SNA), along with strategic diagrams. Results. A total of 3179 articles on RVO were retrieved, and the annual research output increased with time. USA ranked first with the most publications, with Retina as the most prolific journal in RVO research. MeSH terms were characterized into five different genres. As shown by the strategic diagram, the complications of RVO, the etiology of macular edema, as well as the therapeutic use of anti-VEGF, steroids and anti-inflammatory agents were well developed (Quadrant I). In contrast, epidemiology, metabolism and genetics related research on RVO were relatively immature (Quadrant III). Research on surgical treatments of vitrectomy, diagnostic methods and pathology of RVO were centralized but undeveloped (Quadrant IV). The SNA results was exhibited by the centrality chart, on which the node position was represented by the centrality values. Conclusions. By providing a bibliometric research, the overall RVO research trends could be revealed based on the five categories identified by this study. The mathematical bibliometric study could shed light on new perspectives for researchers.

three major categories of RVO based on the occlusion site. Patient symptoms may generally include, depending on the classification of RVO, blurry or missing vision, floating dots or lines, and defective visual field. The main complications of CRVO include the macular edema formation, neovascularization, neovascular glaucoma, and vitreous hemorrhage, while complications of BRVO depend more on the vessel occluded [2].
Academic journals have published a vast amount of papers in RVO related research over recent decades. In order to reduce the effort and time required for a traditional systematic literature review, we applied bibliometric methods to explore the research status in an RVO study.
Bibliometry is a computational analytical method, which is based on mathematical and statistical analysis of an article's attributes, and it can be used to describe, assess and predict the current status and future development of science and technology [3]. Co-word and co-citation analyses are the common means employed in bibliometry analysis to demonstrate research trends. It is assumed that a collection of academic words from an article of interest can be used to outline this article. Based on this presumption, co-word analysis can be employed to assess the relationship of two academic words in an research literature. This study is focusing the assessment of RVO research trends with co-word analysis.

Data Collection
Medical subject headings (MeSH) is a universal terminology used in academic publications to index and categorize biomedical information. In general, approximately ten to fifteen titles and subtitles are applied to index each published paper [4]. Co-word clustering analysis can be performed based on MeSH terms. In this study we select all journal articles in English from the PubMed database with the MeSH term "retinal vein occlusion", and 3179 articles in total were identified and used in our analysis.

Data Extraction and Bibliographic Matrix Setup
Bibliographic Item Co-occurrence Matrix Builder (BICOMB), designed and built by Prof. Cui from China Medical University, was employed to examine the distribution features including publication time, nations, journals and researchers. In addition, BICOMB was also used to obtain the main MeSH terms/MeSH subheadings of these selected literatures [5].
The inventory, acquired from the PubMed database, was implemented for the generation of a term-source and co-occurrence matrix, and these matrices were further applied for subsequent bibliometric analysis.
Bi-clustering analysis of the high-frequency main MeSH terms/MeSH subheadings Threshold value, T= (1+)/2, was used to evaluated the quantity of high-reoccurrence main MeSH terms/MeSH subheadings, and in this equation, "i" represents the quantity of key MeSH terms/MeSH subheadings with single appearance. RVO hot spots was explored with bi-clustering analysis which was based on the evaluation of high-reoccurrence MeSH terms and RVO-associated research articles. A binary matrix, with its rows built with highreoccurrence key MeSH terms and its columns composed of source article, was developed.
Then, the term-source article matrix was used to conduct co-occurrence double cluster analysis with gCLUTO software which can be retrieved from http://glaros.dtc.umn.edu/gkhome/cluto/gcluto. For hill diagram visualization, peaks are in accordance with the hotspots of the theme, which can be used to roughly estimate the clustering results. The different color appearing on the hill diagram represents different standard deviation (SD), the height of the hill is proportional to the similarity in intra-class and the hill volume is correlated with the quantity of MeSH terms. For dendrogram, highreoccurrence key MeSH terms were displayed as the row labels and the PubMed Unique Identifiers (PMIDs) of the source publications were listed as the column names.

Strategic diagram analysis
The strategic diagram analysis is based on themes of centrality and density. Centrality is represented by the external cohesion index, indicating the them position in the framework, and density is illustrated by the internal cohesion index, reflecting the progression of the themes. With X-axis representing centrality and Y-axis illustrating density, four quadrants were developed. Based on the biclustering assessment, different clusters generated were distributed in different quadrants of this strategic diagram generated by Graph.
Social network analysis SNA network was developed using Ucinet 6.0 (Analytic Technologies Co., Lexington, KY, USA) software according to the high-reoccurrence key MeSH terms/MeSH subheadings cooccurrence matrix. A two-dimensional map, for visualization, was generated with the key MeSH terms/MeSH subheadings using NetDraw 2.084 software. On this map, the key MeSH terms/MeSH subheadings were shown as the nodes and the frequency of their cooccurrence was displayed as the links. Furthermore, closeness, betweenness and degree centralities were employed to examine the location of the key MeSH terms/MeSH subheadings, in order to obtain in-depth understanding of RVO network organization.

Distribution characteristics of relevant literatures
With the searching criteria described above, 3179 publications in total were included in this study. As displayed in Fig.1A, research articles published yearly in the RVO field has gradually increased from 90 in 2004, to over 200 in 2015. Among all the first authors involved in this topic, Noma H ranked first by publishing 52 articles (Fig.1B). As for the amount of RVO research output, the United States ranked first with 1,544 publications, making up almost 50% of the research in this specific area (Fig.1C). Among the top 10 most productive journals displayed on Fig.1D Table S1.  To better understand this, the betweenness centrality was used to develop the SNA. As shown in Fig. 4, the betweenness centrality is represented by the node size and cooccurrence frequency is displayed by line width. anti-VEGF therapy [9]. VEGF is an inflammatory cytokine that promotes vascular permeability and is upregulated in eyes with vein occlusion [10]. Cluster 0 is associated with research on epidemiology and the metabolism of RVO. The prevalence of RVO has been reported to range between 0.4% and 4.6%. Of the two main types of RVO, BRVO is four to six times more prevalent than CRVO [12]. The balance between inflammatory cytokines and angiogenesis in eye fluid is disturbed in patients with RVO. Exposure of endothelial cells to proinflammatory cytokines can cause oxidative stress and apoptosis, aggravating leukocyte efflux and thrombosis. Significantly increased concentrations of IL-1α, -6, and -8; IP-10; and PDGF-AA were observed in RVO patients when compared to control patients [10]. Macular edema secondary to RVO is associated with increased levels of VEGF in the aqueous humor. Therefore, the management of macular edema secondary to RVO, especially in the presence of capillary con-perfusion areas, should aim at reducing ocular VEGF concentration [13]. Cluster 3 relates to genetic studies on RVO. Thrombophilic diseases like factor V Leiden mutation, hyperhomocysteinemia and anticardiolipin antibodies increase the risk of RVO [14].

Discussion
Proteomic studies suggest that RVO is associated with the remodeling of the extracellular matrix and adhesion processes. However, many areas of proteome changes in RVO remain unstudied. Future studies may address long-lasting retinal changes following intervention with anti-VEGF agents, such as dexamethasone intravitreal implants [15]. These two clusters, assigned to Quadrant III, represents research hotspots which are marginal and immature, and future research on these topics is suggested. representing the key components with the highest influence in the determination of other components' co-occurrence. "Retinal vein/pathology" and "Retina/pathology" are among the top ten MeSH terms with betweenness centrality; however, these two components are located in the IV quadrant and are not included in the MeSH terms listed with the top ten high-reoccurrence. This demonstrates that although these two components are important in the network stability, but the research on this topic is not well developed.

Conclusions
In summary, the structure and maturity of an identified field can be evaluated and

Availability of data and materials
The datasets included in this study are available in the PubMed database.    Table 1.  Table 2. The size of a signal node represents the number of major MeSH terms/MeSH subheadings involved in each cluster.