Trends in research on corneal cross linking from 2001 to 2020: a bibliometric analysis

ABSTRACT Clinical relevance The research status and hotspots in the field of corneal cross linking (CXL) can benefit clinicians, researchers and the general public. Background The purpose of this study was to map the publishing trend on CXL research and explore the research hotspots. Methods A bibliometric analysis was performed using the Web of Science Core Collection to investigate the publishing trend on CXL research. VOSviewer was used to build the knowledge map to visualise the number of annual publications, distribution of countries and institutions, international cooperation, author productivity, source journals and research hotspots in the field of CXL. Results A total of 2061 peer-reviewed articles on CXL research were collected from 2001 to 2020, and the annual research production increased over time. The United States was the country with the largest number of published articles, and the University of Zurich was the most active institution. Hafezi F published the largest number of articles on CXL, while Cornea was the journal with the largest number of studies on CXL. The most frequently cited references mainly focus on CXL in the treatment of keratoconus. The keywords were divided in 5 categories: 1) CXL mechanism, 2) ectasia diseases and refractive surgery, 3) corneal biomechanics, 4) efficacy evaluation, 5) treatment of infectious keratitis. Conclusion The quantity and quality of articles on CXL were evaluated using bibliometric techniques by extracting the data from the Web of Science Core Collection. The research hotspots could provide insights on CXL research, providing valuable information for clinicians to perform research in this field and find potential partners.


Introduction
Corneal cross linking (CXL) is a procedure used to strengthen and stabilise the cornea to prevent or delay the progression of specific diseases through a photochemical reaction using ultraviolet A (UVA) and riboflavin as photosensitisers. 1 The induced photopolymerisation forms additional covalent bonds in the collagen fibres to enhance the biomechanical hardness and biomechanical resistance of the cornea. 2 A large number of articles on CXL research have been published in academic journals over the past few decades. Therefore, it is of utmost importance to explore the hotspots and trends in CXL research.
Bibliometric analysis is a process of extracting measurable data through the statistical analysis of published studies. 3 Bibliometric analysis provides a qualitative and quantitative evaluation of the research trend in a specific field. Previous studies reported that bibliometric analysis explores the research hotspots and development trends in a certain field. [4][5][6] CiteSpace and VOSviewer software present the structure, rules and distribution of a specific scientific knowledge through visualisation. 7 Knowledge mapping can be obtained by the co-occurrence analysis and co-citation analysis. 8 To the knowledge of the authors, a bibliometric analysis on CXL research has not been conducted yet. Therefore, this study used bibliometric and visualised analysis to describe the overall situation in the field of CXL. This study objectively presents the number of articles, international cooperation, author productivity and source journals in the CXL field to perform cocitation analysis and keyword co-occurrence analysis related to CXL research.

Data source and research process
The search of articles included in this study was performed using the database Science Citation Index Expanded in the Web of Science Core Collection. The comprehensive search was performed on January 12th, 2021. The considered topic was 'corneal cross linking', and the time span was 'from 2001 to 2020'. No language restrictions were set. Journal articles and reviews were used for the bibliometric analysis because they represented the majority of the documents that contained all the ideas and results of the research. 9 The content of the record derived from the articles was 'full record and cited references' and the derived format was 'tab-delimited'. The initially download was performed on the raw data from the Web of Science Core Collection. The authors Wang S and Yang K performed the extraction of the data and verified the data loss or the duplicate data input of the collected results. The basic information collected from each considered article was the following: title, author, journal, institution, abstract, keywords, and references.

Analytical Tool and Method
The collected data were imported into VOSviewer v.1.6.14 to perform the system analysis. VOSviewer is a literature visualisation software developed by Eck and Waltman, 10 which can be used in almost all common bibliometric analysis, such as co-authorship, co-occurrence, cocitations, and bibliographic coupling; the results can be displayed and the scientific knowledge maps can be drawn. The nodes in the knowledge maps represent an element, such as the country, organisation, journal, author, co-citation article and keyword, while the size of the nodes reflects the frequency of the elements. The links between two nodes reflect the relationship between these elements. The higher the frequency of cooperation, the thicker the connection line. 11 CiteSpace was run to find the most cited keywords to help the prediction of new frontier topics or research trends in the future. 12 In this study, the map on CXL research was constructed by co-citation cited reference and keyword co-occurrence network, which are types of analysis in VOSviewer. The clustering of similar references produces clusters of cocitation cited reference that can be used to explore the main topics regarding a certain matter. The keywords reflect the topic of the academic articles. 13 The knowledge and research hotspots in this field were revealed through the co-occurrence network along with the cluster analysis of the keywords. However, it is necessary to use a unified expression for the same keyword to avoid bias because different expressions of the same keyword may lead to errors in the results. 14 Thus, the keyword data were standardised by the authors before the analysis to unify the different expression of the same keyword.

Annual quantitative distribution of publications
According to the selection criteria and reprocessing, 2061 publications on CXL research that were indexed in the Web of Science Core Collection from 2001 to 2020 were identified and included. The annual research output generally increased with time, especially since 2010 ( Figure 1). The keyword burst detection analysis detected 30 keywords that represented the citation bursts (Figure 2), and among 'collagen' 'UVA' and 'protocol' were the keywords with the highest citation strength. Research hotspots in the field of CXL have changed over the past 20 years compared to the initial CXL mechanism such as 'collagen' and 'keratocyte' to the evaluation of the clinical application of CXL such as 'term follow up', 'protocol' and 'pack-CXL'.

Distribution of productive countries
The 2061 studies on CXL were performed in 74 different countries. The top 10 active countries are listed in Table 1, with 1856 articles accounting for 90.05% of the whole number of publications. The top 3 countries with the most publications were the United States (551 articles, 26.73%), China (210 articles, 10.19%) and Germany (199 articles, 9.66%), which constituted the 46.58% of the total publications. The citation analysis according to countries revealed that the United States had 15,248 citations, ranked first among all countries, followed by Germany (7598 citations) and Switzerland (7455 citations). The cooperation relationship is shown in Supplemental Figure 1.

Quantitative analysis of the research organisations
The 2061 publications were performed by 2061 organisations. The top 10 most productive organisations in this research field published a total of 347 articles, accounting for 16.84% of the total publications ( Table 2). The most productive organisations on CXL research were the University of Zurich (42 publications), followed by the University of Crete (41 publications), and Tehran University of Medical Sciences (41 publications). Supplemental Figure 2 shows the co-authorship analysis network, which reflected the cooperation relationship among these institutions.

Distribution of authors and co-authorship analysis of the research groups
A total of 6725 authors participated in the study of CXL. Among the top 10 authors, Hafezi F (62 publications) was the first, followed by Kymionis GD (43 publications) and Kling S (30 publications), indicating their outstanding contribution to CXL research. The information regarding the author cocitations was analysed. Wollensak G (2963 co-citations) was the first, followed by Spoerl E (1191 co-citations) and Kymionis GD (1087 co-citations) among the top 10 co-cited authors, according to VOSviewer, indicating their important role and powerful influence in CXL research (Table 3). Supplemental Figure 3 shows the cooperative network among authors, and some reference information regarding the cooperation is provided. The first group (red colour) has Prof. Spoerl E as its core; the second group (green colour) has Prof. Kymionis GD as its core; the third group (blue colour) has Prof. Hashemi H as its core; the fourth group (yellow colour) has Prof. Chan E as its core; the fifth group (purple colour) has Prof. Nuijts RMMA as its core; the sixth group (grey colour) has Prof. Elsheikh A as its core.

Quantitative analysis of the source journals
The 2061 publications belonged to 335 journals. Table 4 lists the top 10 active journals involved in CXL research. Cornea published the largest number of articles (218, 10.58%), followed by the Journal of Refractive Surgery (208, 10.09%) and the Investigative Ophthalmology and Visual Science (119, 5.77%). Articles published in these 3 journals represented more than one-fifth of all the publications included in this study.

Analysis of the co-cited references
Supplemental Table 1 lists the top 10 co-cited references. The minimum citations of a cited reference were set as 25 of the 34,627 cited references due to the large number of cited references. A total of 372 co-cited references met the threshold and were selected for analysis. According to this network, the references with high similarity were clustered. The 5 main clusters are indicated using the colours red, green, blue, yellow, and purple (Supplemental Figure 4).

Analysis of the keyword co-occurrence
The minimum co-occurrence of a keyword was set as 20.
Among the 5158 extracted keywords used in CXL research, 153 met the threshold and were selected for analysis. On the basis of this network, the keywords with high similarity were clustered. The 5 main clusters are indicated using the colours red, yellow, green, blue, and purple (Supplemental Figure 5).

Discussion
In this study, the visualisation of the bibliometric information was used to analyse the literature on CXL research from 2001 to 2020, to understand the current research status, trends and hotspots. A total of 2061 articles on CXL were collected. The number of publications gradually increased over time, especially since 2010. It might be explained by the fact that the study published by Caporossi A in 2010 reported that the keratoconus is stable for a long time after CXL and no relevant side effects are observed. 15 After reviewing the results of the bibliometrics analysis of the countries and institutions, it was not difficult to conclude that the United States, China and Germany are the 3 countries that occupied the pivotal position in the research on CXL. The co-authorship network map of countries revealed the existence of some collaborating relationships among the top 10 active countries, with the United States and Switzerland that collaborated most often. The United States accounted for 26.73% of publications and ranked first in the number of citations, revealing that the United States is the international scientific centre in CXL research. The co-authorship network of institutions showed that Harvard University was the most productive institution in the United States. The institutions in China that have published the most were Wenzhou Medical University and The Chinese University of Hong Kong, but no cooperative relationship was found between these two institutions. The University of Zurich, Université de Genève and ELZA Institute were the most productive institutions in Switzerland; they often cooperated and occupied an important position in co-authorship network. As regards the number of links among two nodes, in this case representing two organisations, the Federal University of São Paulo has the highest number of links (25 links), indicating that this institution is the key node in the collaborative network.
Cornea, Journal of Refractive Surgery, and Investigative Ophthalmology and Visual Science were the top 3 productive journals on CXL. Therefore, authors interested in CXL research should pay more attention to these journals. In addition, there were other journals published less than 60 articles on CXL, which could also greatly help the dissemination of CXL related studies. According to the fact that high-quality studies are widely cited, the relevant topics in the selected articles can be described by the use of citation parameters. 16 A large number of cited references can effectively illustrate the research background. 13 Therefore, the cluster analysis was carried out on the main research topics of CXL through a co-citation analysis. The cited references were divided into 5 clusters, and the centralised connections among the clusters indicate that the topics in the background knowledge were centralised. Cluster #1 (red colour) mainly represented topics associated to keratoconus. Cluster #2 (green colour) and Cluster #4 (yellow colour) focused on CXL treatment applied to ectasia diseases, mainly represented by progressive keratoconus. The top 10 co-cited high-frequency references in CXL research mainly focused on this field. The publication with the title 'Riboflavin/Ultraviolet-A-induced collagen crosslinking for the treatment of keratoconus' ranked first in the count of the frequency. In this study, 23 eyes with moderate or advanced progressive keratoconus were subjected to a follow-up from 3 months to 4 years and the results of this follow-up revealed that CXL can be considered as a new method to prevent the progression of keratectasia in patients with keratoconus. 17 Cluster #3 (blue colour) mainly focused on the mechanism of CXL and the biomechanical changes of the cornea. Cluster#5 (purple colour) mainly represents the use of the CXL in the treatment of infectious keratitis.
The specified keywords represent the search topic. The topic of CXL research mainly formed 5 clusters, and the keywords in the same cluster were more similar to specific research topics. The following 5 groups were analysed combined with the characteristics and current situation of CXL research.
Cluster #1 (red) represents keywords related to CXL mechanism, and the extracted co-occurrence keywords included 'cornea', 'collagen', 'tissue', 'stroma', 'epithelium', and 'mechanical-properties'. The cross linking creates chemical bonds among macromolecules such as proteins that polymerise. The biomechanical strength of the cornea is directly associated to the tissue of collagen fibres and lamellae. 18 Wollensak et al. 2 in 2003 proposed the use of photosensitiser riboflavin and UVA in the treatment of progressive keratoconus. This photooxidation induces the formation of covalent cross-linking bonds between collagen fibres in the corneal stroma, thus improving the mechanical hardness of the cornea.
Cluster #2 (green) represents keywords related to ectasia diseases and refractive surgery, such as 'keratoconus', 'refractive surgery', 'ectasia', 'LASIK', 'PRK', 'photorefractive keratectomy', and 'keratectasia'. Corneal ectasia after refractive surgery is a very serious postoperative complication. 19,20 CXL has been used to treat patients with post-refractive ectasis and has shown promising results. 21,22 On the other hand, CXL alone is not sufficient to restore the functional vision in patients with keratoconus; thus, the combination of CXL with adjuvant refractive surgery has been introduced to provide both disease stability and enhanced functional visual correction. 23 Several refractive procedures such as conductive keratoplasty, photorefractive keratectomy, implantation of intrastromal corneal ring segments, transepithelial phototherapeutic keratectomy, and implantation of phakic intraocular lens are combined with CXL to enhance its outcome. 24 Cluster #3 (blue) represents keywords related to corneal biomechanics, such as 'biomechanical properties', 'ocular response analyser', 'intraocular-pressure', 'thickness', 'hysteresis', and 'elasticity'. The biomechanical properties of the cornea are corneal hysteresis and corneal resistance. The main purpose of CXL is to improve the biomechanical hardness of the corneal collagen to prevent keratocele progression. 25 The analysis of the biomechanical changes after CXL can provide information for more efficient protocols and treatment options in the use of CXL on patients. 26 Further studies on the biomechanical changes after CXL might offer the opportunity to evaluate the efficacy of CXL.
Cluster # 4 (yellow) represents the evaluation on the efficacy of CXL in the treatment of keratoconus, such as 'riboflavin', 'progressive keratoconus', 'ultraviolet', 'efficacy', 'safety', and 'demarcation line'. CXL is the only method that can stop or slow down the progression of keratoconus. 27 In addition to the standard epithelium-off CXL, other surgical methods are available including accelerated CXL, transepithelial CXL and iontophoresis CXL. At present, a large number of studies are available reporting the observation and evaluation of CXL efficacy in the treatment of progressive keratoconus in adults or paediatric patients. [28][29][30][31][32] The safety and efficacy of the standard epithelium-off CXL in halting the progression of keratoconus have been confirmed by multiple short-term and long-term studies. 28,30 However the results of other surgical methods are inconsistent and the efficacy is not sure 33 ; thus, further studies, as well as long-term efficacy evaluation, should be performed.
Cluster # 5 (purple) represents keywords related to CXL in the treatment of infectious keratitis, such as 'ultraviolet-a', 'infectious keratitis', 'therapy', 'bacterial keratitis', and 'fungal keratitis'. Infectious keratitis is a serious vision-threatening disease that can be caused by bacterial, viral, fungal or protozoal infections. 34,35 CXL hardens the stroma of the cornea, increasing its resistance to enzymatic degradation caused by microorganisms and thus arresting the progression of corneal melting. 36,37 Furthermore, the Reactive Oxygen Species produced by the cross-linked photooxidation react with the intracellular components of the microorganism, resulting in cell inactivation and death. 38 Clinical studies revealed that CXL combined with an antimicrobial treatment is effective in the treatment of bacterial, fungal and acanthamoeba keratitis compared with the effect of the standard antimicrobial therapy alone. However, it may aggravate or activate the infection by herpes simplex virus and should not be used to treat viral keratitis. 39,40 This bibliometric analysis of CXL research could describe the research results and knowledge structure of the past two decades. However, this work contains some limitations. First, few publications on CXL before 2001 are available, thus, only publications from 2001 to 2020 were extracted from the Web of Science Core Collection, being not sufficient to represent all the topics on CXL research. Therefore, since the database is constantly updated, new topics should be analysed in the future. Second, this analysis was carried out using data extracted from the Web of Science Core Collection, and the articles are not included in this database were not analysed in this study, limiting the completeness of the study. Third, the bibliometric software cannot distinguish the real contribution of authors in complex cooperative relationships; thus, the researchers need to read the original literature themselves.

Conclusion
In conclusion, the first bibliometric analysis of CXL research showed the hotspots and development trends over the past years, providing a basis and ideas for further research in this field.

Disclosure statement
No potential conflict of interest was reported by the author(s).