A comprehensive literature search was performed in the WOS Core database (Clarivate Analytics, United States), which is considered the most appropriate database for bibliometric analysis. We used WOS to identify the most frequently cited papers. The reason for using this database is that it is the most frequently used and accepted database in scientific or bibliometric research, includes almost all influential and high-quality journals, and contains a comprehensive citation index record. We searched relevant kinds of literature involving “endoscopic mucosal resection” or “endoscopic mucosal resections” or “endoscopic mucous membrane resection” or “mucosal resection and endoscopic” or “resection and endoscopic mucosal” in their titles or keywords.

Two authors (YYH and XX) independently confirmed all relevant publications, including the title, keywords, publication year, country/region, institution, author, and citation count. We applied filters to limit the search to original articles and reviews. The period of the literature search was from 1991 to 2021, and non-English literature was excluded from our study. We completed our literature retrieval and data downloads in the course of 1 day, June 23, 2022, to reduce bias arising from frequent updates of the database. Given that data were directly downloaded from the database, ethical approval was not needed.

For bibliometric analysis, we first converted the WOSCC data to TXT format. Next, we imported these data into CiteSpace V5.7.R3 SE, 64-bit (Drexel University, Philadelphia, USA) and VOSviewer1.6.15 (Leiden University, Netherlands). Then, we performed quantitative and qualitative analyses of these data in CiteSpace and VOSviewer.

In CiteSpace, we used the following options: the time slice is set to “1991–2021”, the number of years for each slice is set to “1”, the selection criteria are set to “g-index”, and the scale factor k is set to “25”. Moreover, to preserve the most significant structure and reduce the number of links, we selected the options “pathfinder” and “trim the merged network”. For node types, we could only select one option at a time from “author”, “institution”, “country”, “reference”, “citation author” and “keywords” (18).

The role of CiteSpace is important for bibliometric analysis, which can explore the knowledge base and frontier of a certain research field by performing cocitation analysis and burst detection. When two articles are simultaneously cited by a third article, the two articles form a cocitation relationship. The strength of the cocitation relationship between two cited papers is directly proportional to the similarity of their research content. The more times they are cited at the same time, the stronger the cocitation relationship. In addition, references with a strong cocitation relationship form a certain category reflecting the same research topic. Centrality is a major indicator to determine the importance of nodes in the network and a higher centrality means that the node is more important in this network. Generally, nodes with a Centrality value of more than 0.1 occupy pivotal positions connecting a large number of nodes and are usually identified as hubs of nodes displayed in purple.

We selected the top 25 keywords with the largest citation volume to explore the research hotspots of EMR. In the burst detection results, “Begin” refers to the year when a burst of citation begins with a reference or keyword, and “End” refers to the year when a burst of citation ends with a reference or keyword. The red line refers to the duration of the burst of citation, and “Strength” refers to the intensity of the burst of citation.

In VOSviewer, we created keyword maps by using the following options: “create a map based on bibliographic data”, “read data from bibliographic database files”, “analysis type: co-occurrence”, “analysis unit: all keywords”, “count method: full count”, and “minimum occurrences of keywords: 20”. The circle size of a project is proportional to its number of publications. The width of the lines between the two projects is proportional to the level of cooperation. Objects of the same color belong to the same cluster, which indicates that they cooperated closely in this field. Meanwhile, the journal impact factors and quartiles of the journal category were identified according to the journal citation report 2021 criteria to evaluate the scientific impact of the country/region and journal (https://jcr.clarivate.com/).

We obtained 4,223 papers from WOSCC by searching for relevant keywords concerning EMR. Excluding nonoriginal articles and reviews and non-English language papers, 2,766 papers were obtained. Then, excluding papers not published between 1991 and 2021, we finally obtained 2,695 papers to analyze in this study. The flow chart for this retrieval is shown in Figure 1.

After screening, a total of 2,695 publications were identified in the present study, including 2,266 (84.1%) original articles and 429 (15.9%) reviews. Figure 2 shows the chronological distribution of the publications from 1991 to 2021. The first article we retrieved was published in 1991. INOUE, H reported a case of carcinoma in situ of the esophagus accompanied by esophageal varices that were treated by endoscopic mucosal resection using a transparent tube following eradication of the varices via injection sclerotherapy. As depicted in the diagram, the number of articles and reviews grew steadily in the past two decades and peaked in 2021.

At least 2,349 institutions from 65 different countries/regions have published papers on EMR. The papers are mainly on Japan (958 papers, 35.5%) and the United States (606 papers, 22.5%), and they have together published more than half of the relevant studies on EMR. South Korea (261 papers, 9.7%), China (205 papers, 7.6%), and the United Kingdom (176 papers, 6.5%) also contributed significantly to EMR research. Germany, the United States, Japan, the United Kingdom, and Italy scored highest for the centrality of research cooperation. Table 1 shows the detailed distribution of these countries/institutions, and Figure 3A shows their co-occurrence networks. According to Table 2, five of the top 10 research institutions with the most publications are based in South Korea, two are in Japan, two in Australia, and one in the United States. Table 3 shows the top 10 most cited institutions, none of which are from South Korea. The organization with the most publications is the NATIONAL CANCER CENTER JAPAN. The symbiotic network between research institutions shows a low-density graph (density = 0.0091), with most of the centrality below 0.1 in Figure 3B. This means that research groups are relatively dispersed across institutions, but most institutions have a restrictive influence in the field.

A total of 12,052 authors participated in these papers, 106 of whom published more than 10 studies. Table 4 shows the 10 most prolific authors in the study. Bourke, Michael J. from the Department of Gastroenterology of WESTMEAD HOSPITAL in Sydney, Australia ranked first in the number of papers published (63 papers). This was followed by YUTAKA SAITO (50 papers) and TAKAHISA MATSUDA (33 papers) of the Department of Endoscopy, National Cancer Center, Japan. It is worth noting that the number of cited papers of MICHAEL J BOURKE and YUTAKA SAITO is significantly higher than all researchers, indicating that the two authors have made great achievements and have become authorities in the field of EMR research. We visualized the authors using CiteSpace software in Figure 4. The network among authors showed a low density (density = 0.0054). These authors are groups, but there is a lack of cooperation between the groups.

A total of 365 journals published papers in the EMR field. Among 2,695 papers concerning EMR in our study, 1,086 papers (40.3%) were published in the top 11 journals, as shown in Table 5. Judging by the number of posts, the top 3 were ENDOSCOPY (IF = 9.776), GASTROINTESTINAL ENDOSCOPY (IF = 10.396), and SURGICAL ENDOSCOPY AND OTHER INTERVENTIONAL TECHNIQUES (IF = 3.453). It is worth noting that GASTROENTEROLOGY (IF = 33.883) far outperformed the other journals in terms of the number of average citations. The top 3 most published journals in the field of EMR are endoscopy journals. These journals have shown a strong interest in EMR and are more likely to accept articles in this area. The coauthorship analysis of journals was performed, and a network map was constructed, as shown in Figure 5. The weight of the node is represented by the number of posts, and the color of the node is represented by the average number of citations.

Keyword co-occurrence analysis provides a detailed description of hot topics involved in EMR research. Each paper has corresponding keywords. By analyzing the titles and abstracts of the included papers, VOSviewer identified 183 keywords that appeared at least 20 times and visualized citation data with a bubble graph. In the VOSviewer keyword co-occurrence visualization map, all keywords are grouped into clusters, and different clusters are marked by different colors.

The three categories are “early gastric cancer”, “colorectal cancer”, and “Barrett's esophagus” in Figure 6A. There is a color bar in the lower right corner of the map with keywords in different colors based on the average year of publications in Figure 6B. For example, “early gastric cancer” and “lymph node metastasis” mainly appeared before 2010. In subsequent years, the keywords “Barrett's esophagus”, “endoscopic submucosal dissection”, “colorectal cancer” were more common. Keywords marked in yellow, such as “outcomes,” “safety,” “colonoscopy,” “adverse events,” and “cold snare polypectomy” are the latest. This shows that these areas have become increasingly popular in recent years and could become hot spots in the future.

Burst detection reveals a sudden increase in emerging concepts over time. The timeline is drawn as a blue line divided into years. The periods in which a topic's outbreak was observed to be marked in red indicate the start and end years and duration of the referenced outbreak. The burst pattern of keywords reveals new content and the relevant research focus in EMR. We combined keywords of the same meaning, such as carcinoma, tumors, and cancer. Figure 6C shows that, over the past three decades, the stomach ranked first in terms of outbreak intensity (21.47), followed by photodynamic therapy (21.25), cancer (21), and early gastric cancer (19.83). Since 1993, endoscopic ultrasonography, cancer, and early gastric cancer have become the focus of research, followed by lining and photodynamic therapy. It is worth noting that adverse events, management, European society, cold snare polypectomy, and recurrence were the strongest bursts since 2017, indicating that they have become new research topics of EMR.

A cocited network is a network of references cocited by some group of publications. A concept cluster is an edge generated when a group of references is repeatedly cited. We generated cluster network diagrams (Figure 7) and literature cocitation diagrams (Figure 8) through CiteSpace. We chose the “Pathfinder” and “Pruning Networks” options to preserve the best network structure. The visualization with references shows 1,455 nodes and 3,844 links. In this network, each node represents a cited article, and the size of each node is proportional to the total cited frequency of relevant articles. In Figure 7, the cocited literature is grouped into 16 major classification labels, indicating that the field of endoscopic mucosal resection has focused on the following 16 topics in the past 30 years: Barrett's esophagus, polypectomy, photodynamic therapy, early gastric cancer, esophagectomy, gastric cancer, endoscopic submucosal dissection, polyps, colonic lesions invading the submucosa, neuroendocrine tumors, duodenal neoplasms, tumor marker, small cell carcinoma, endoscopic ultrasonography, sessile serrated adenoma and eleview. The smaller the label number, the more literature it contains. We list the information for each cluster in Table 6. The silhouette value of a cluster reflected its homogeneity. The closer the value was to 1, the more homogenous it was. When the silhouette value of a cluster was >0.7, this cluster could be considered highly reliable. The silhouette of the 16 clusters ranged from 0.867 to 1, reflecting their relatively high homogeneity. A timeline view of distinct cocitations is presented in Figure 8. According to Figure 8, the research hotspot and topics have shifted from gastric cancer, colonic lesions invading the submucosa, tumor marker and endoscopic ultrasonography to Barrett's esophagus, polypectomy, endoscopic submucosal dissection, neuroendocrine tumors, duodenal neoplasms and esophagectomy.