The Web of Science Core Collection (WoSCC) is one big and comprehensive database for scientific research (18), with more than 12 000 high-quality journals. Also, previous studies have indicated that WoSCC is the most appropriate database for bibliometric analysis (19). According to the literature related to glioma, we determined the common keywords about glioma, and then we preliminarily formulated the search terms related to “glioma”. At the same time, by reading the original articles and reviews of glioma immunotherapy, we have a preliminary understanding of the mainstream immunotherapy methods for glioma, which have been included in the relevant key words of “immunotherapy”. Taking the intersection of the two, we get our search formula. Publications pertaining to glioma immunotherapy were searched from WoSCC with the following queries: #1: TI=(“glioma”) OR AK=(“glioma”) OR TI=(“malignant glioma”) OR AK=(“malignant glioma”) OR TI=(“high grade glioma”) OR AK=(“high grade glioma”) OR TI=(“glioblastoma”) OR AK=(“glioblastoma”); #2: TI=(“immunotherapy*”) OR AK=(“immunotherapy*”) OR TI=(“immune NEAR/2 therap*”) OR AK=(“immune NEAR/2 therap*”) OR TI=(checkpoint inhibitor) OR AK=(checkpoint inhibitor) OR TI=(PD-1) OR AK=(PD-1) OR TI=(PD-L1) OR AK=(PD-L1) OR TI=(nivolumab) OR AK=(nivolumab) OR TI=(pembrolizumab) OR AK=(pembrolizumab) OR TI=(CAR-T) OR AK=(CAR-T) OR TI=(chimeric antigen receptor T-cell therapy) OR AK=(chimeric antigen receptor T-cell therapy) OR TI=(dendritic cell vaccines) OR AK=(dendritic cell vaccines) OR TI=(viral therapy) OR AK=(viral therapy); the ultimate dataset: #1 AND #2. Only English-written articles or reviews published during the period from 1st January 1990 to 27th March 2023 were chosen. A truncation symbol “*” was used and the use of truncation searches improved recall and prevented missing inspection. The articles with research content related to the theme of “glioma immunotherapy” were included by reading the titles, abstracts and keywords of the detected articles. Articles with incomplete research information, book chapters, proceedings papers, early accesses, retrieved publications and duplicate articles were excluded. Eventually, after excluding 22 book chapters, 20 proceedings papers, 4 early accesses, 2 retrieved publications, a total of 1,929 eligible publications were included ( Figure 1 ). All data used in this work were downloaded from a public database; therefore, no ethics approval or informed consent was required. The above processes were carried out under the guidance of glioma experts and bibliometrics experts.

The publications were downloaded and their data were put into Microsoft Office Excel 2019, analyzed and visualized using CiteSpace 6.2.R2, VOSviewer 1.6.18, Online Analysis Platform of Literature Metrology (https://bibliometric.com/) and R 4.2.2 (20–22).

Tables were depicted using Microsoft Office Excel. The Online Analysis Platform of Literature Metrology was used to visualize networks of collaborations between countries/regions and trends in the volume of publications over time. R software was used to visualize the collaborative relationships between countries, authors, and institutions, and to analyze references.

VOSviewer is a widely-used software in bibliometric analysis developed by van Eck and Waltman. It is powerful in building and visualizing bibliometric networks, such as the co-authorship, co-occurrence and co-citation networks of counties, organizations, journals, authors and keywords (23, 24). It was used in the present study to visualize co-authorship between authors, institutions and countries, and the co-occurrence of keywords. Co-authorship analysis is a measure to establish similar relationships among items through the number of coauthored documents, and co-occurrence analysis is to build a network that represents the relationships between items according to the quantity of publications occurring together. Co-citation refers to that two documents were cited simultaneously in the bibliography of the third document (19, 23).

CiteSpace is another information visualization software for bibliometric analysis founded by Professor Chaomei Chen (25). This software can be used to find the key points, especially intellectual pivotal points, or turning points in the research history of a certain field, and predict the research trend through data mining. Additionally, this software is unique in that it can be used to build a dual map of journals, which is generated in the context of 10 000 journals indexed in WoSCC (24). In this study, it was used to (1) analyze the countries, institutions and authors involved in the study; (2) construct dual-map of journals involved; (3) demonstrate reference clustering and references/keywords burst. In CiteSpace, the results are visualized as nodes and lines, and for these nodes, the centrality score is calculated to measure the relative importance of a node in a network. When a node has a centrality score greater than or equal to 0.1, it is usually considered a relatively important hub node (19, 25). In co-cited reference clustering, CiteSpace adopts several metrics to measure the outcome clusters, which include modularity and silhouette. The modularity measures the significance of the divided modules. While the modularity value is greater than 0.3, the clustering structure is usually considered as significant. The silhouette is an indicator used to measure the consistency of a cluster. A cluster is considered reasonable or convincing when its values are greater than 0.5 or 0.7, respectively (19, 26).

From 1st January 1990 to 27th March 2023, a total of 1,929 publications pertaining to glioma immunotherapy were identified, including 1,285 articles and 644 reviews ( Figure 1 ). All of them had been cited for 61,875 times, and 47,878 times after removing self-citations. The mean citation and H-index of a publication were 32.08 times and 109, respectively. Until 2005, just less than 20 articles were published in this field each year ( Figure 2 ). However, in the following decade (except for a few short periods showing a decrease in publication volume), articles related to glioma immunotherapy flourished gradually. Starting from 2006, the annual number of articles published kept increasing, with over 30 articles in each of the 8 years. Starting from 2016, the number of articles in this field grew rapidly. Except for the number of publications in 2019 remained the same as that in the previous year, the number of articles published in all other years had continuously increased, so did the number of citations. The number of articles published in 2023 (as of March) had reached 30, and is predicted to explode in the future.

A total of 62 countries participated in the research of glioma immunotherapy. Table 1 presents the top 10 countries/regions with the largest numbers of articles published, and Figure 3A presents a map of the contributions from different countries. The top country was the United States (n=918, 47.589%), followed by China (n=429, 22.240%), Germany (n=165, 8.554%), Japan (n=104, 5.391%), and Italy (n=101, 5.236%), while the rest countries/regions had released less than 60 publications. Among them, the number of citations of published articles from the United States was 39,991, far exceeding the 7,460 from China, in the second place, and the 7,296 from Germany, in the third place. H-index indicated that each of at least H publications was cited for at least H times (27). A high H-index is parallel to a high academic influence of a country. The United States ranked first with a score of 97, far above other countries. Centrality score of a single country indicates the position of a country in this scientific field. The United States ranked first with a score of 0.64, followed by Germany (centrality=0.23), China (centrality=0.11), France (centrality=0.11); the other countries in the top ten achieved a centrality less than 0.1. As shown in Figure 3B , the US had a relatively stable contribution, which was manifested by its stable number of articles published in this field every year; China, the second contributor, published a large number of articles in the last few years, especially in 2021 and 2022, which could be confirmed by the trend chart of China’s publications over the years ( Supplementary Figure 1 ). It is also worth mentioning that China ranked second in the numbers of publications (429). China published articles in a number almost one-half of that in the US ( Table 1 ), but its total number of citations was just one-fifth of that in the US (7,460), and its average number of citations per article ranked the bottom in the top ten (17.43). Figures 3C , D show the cooperation patterns between different countries/regions. International cooperation was still relatively frequent, mostly occurring between European countries, East Asian countries, and American countries. Among them, cooperation with the US or cooperation between the US and China was the most frequent. In Figure 3D , the colors corresponding to the nodes and timeline represent the time of cooperation between the two. Around 2015, the US was the most active country in this research field, cooperating with many countries. However, around 2020, China became the most active country, and in recent years, most countries have enhanced their cooperation with China.

Table 2 shows the top 10 institutions with the highest total numbers of publications, with Harvard University (n=118) leading in the top five, followed by Duke University (n=82), Harvard Medical School (n=75), Johns Hopkins University (n=69) and University of California San Francisco (n=65). In terms of citations, however, the top five changed to Harvard University (7261), Duke University (4985), UTMD Anderson Cancer Center (4414), University of California San Francisco (4327) and Johns Hopkins University (4001). Harvard University, Duke University, Johns Hopkins University and the University of California San Francisco ranked the top five in terms of either total publications or total citations. Notably, UTMD Anderson Cancer Center, which ranked only 7th in total publications with 55, rose to the third place in the total citations with 4414. In addition, among the top ten institutions, all but University of California Los Angeles (centrality 0.1) had centrality values less than 0.1. The different colors in Figure 4 represent different clusters, and the major institutions are divided into four clusters. The research content in each cluster was closer and more collaborative, which could also be confirmed by the graph drawn by R Software ( Supplementary Figure 2 ). Cooperation among major institutions was relatively regional. For example, US institutions were more likely to cooperate within the US, while major Chinese institutions tended to work together. But this did not mean the lack of international cooperation ( Figure 4 ; Supplementary Figure 2 ).

Of the 1929 articles published in 502 journals, the largest proportion (100, 5.184% of the total) was accepted by JOURNAL OF NEURO-ONCOLOGY (4.506, Q2) ( Table 3 ), followed by FRONTIERS IN IMMUNOLOGY (8.787, Q1), FRONTIERS IN ONCOLOGY (5.738, Q2), NEURO ONCOLOGY (13.029, Q1), and CANCERS (6.575, Q1). Among the top 10 articles, nine journals had an impact factor (IF) of more than 5.0, and eight were in the Q1 JCR division, indicating that most of the articles had been published in journals of high quality and influence.

While in the analysis of co-cited journals ( Table 3 ), CLINICAL CANCER RESEARCH (13.801, Q1) had the highest number of co-citations, followed by CANCER RESEARCH (13.312, Q1), NEURO ONCOLOGY (13.029, Q1), NEW ENGLAND JOURNAL OF MEDICINE (176.082, Q1) and JOURNAL OF NEURO-ONCOLOGY (4.506, Q2). Among the top 10 journals cited, nine had an IF exceeding 10 and were in the Q1 JCR division. Among them were also famous journals, such as NEW ENGLAND JOURNAL OF MEDICINE (176.082, Q1), NATURE (69.504, Q1), JOURNAL OF CLIMATIC ONCOLOGY (50.739, Q1), NATURE MEDICINE (87.244, Q1), and SCIENCE (63.832, Q1).

A dual-map overlay displays the distribution of journals in different disciplines, and the trajectories between the cited and citing publications (28, 29). The distributions of citing journals (left) and cited journals (right) in each field are shown in Figure 5 , in which the connection lines represent the trajectories through which the citing publications cited the cited publications. The horizontal and vertical axes of the oval, respectively, represent the number of authors and publications, the length of which is positively correlated with the number (28). Dual map contained three main pathways, showing that the articles in glioma immunotherapy mainly cited the research outcomes in Molecular Medicine, Biology, and Genetics, and were published in two different fields: Molecular Medicine, Biology, Immunology; and Medicine, Medical, Surgery.

The top 10 productive authors in the research on glioma immunotherapy are listed in Figure 6A . Lim M was the most productive, with 54 publications and 3449 citations. From the perspective of H-index, the gaps between the top 10 authors were not significant; but in terms of citation, Weller M left a significant gap in front of others. His articles had been cited for the largest times (3756 times), with each of his articles cited by 110.47 times averagely. In addition, as shown in Figure 6B , the collaborative relationships between authors indicated a lack of collaboration at the global level. The authors were more likely to collaborate in their own cluster, and the collaboration between different clusters remained to be expanded. This phenomenon is also visualized in Supplementary Figure 3 drawn by VOSviewer. The nodes represent the clusters of authors, and there are fewer connections between clusters.

Table 4 shows the details of the top 10 most cited articles. The most cited article was produced by BROWN CE et al., published in NEW ENGLAND JOURNAL OF MEDICINE in 2016, and titled “Regression of Glioblastoma after Chimeric Antigen Receptor T-Cell Therapy” (30). In a clinical study conducted by their team, they administered CAR-T therapy targeting IL-13Ra2 to a patient with recurrent GBM, and observed a reduction in the size of all lesions and an increase in cytokine and immune cell levels in the cerebrospinal fluid, as well as a significant improvement in the patient’s quality of life.

After normalizing the total number of citations by R software, the highest score (22.42) was achieved by a review from TAN AC et al. published in 2020 in CA-A CANCER JOURNAL FOR CLINICIANS ( 31 ). This article reviewed the current mainstream treatment of glioblastoma. For newly diagnosed tumors, standard treatments included concurrent radiotherapy with temozolomide and further adjuvant temozolomide after surgery. For recurrent tumors, surgery, radiotherapy, chemotherapy or systemic therapy with bevacizumab were all alternative treatments. Finally, the author suggested that in order to improve the prognosis of patients with glioblastoma, more biomarker-based treatment methods should be carried out in the future.

The cluster analysis of co-cited references can reveal the knowledge structure in a research field. Clusters were generated by log-likelihood ratio (LLR) algorithm based on the keywords extracted from the co-cited references (32). The top 11 clusters are depicted in Supplementary Figure 4 . The number of references (size), silhouette value, mean publication year of cited references (i.e., the main year of this cluster), and label of each cluster obtained by LLR algorithm are listed in Table 5 . The overall modularity and weighed mean silhouette of all the clusters were 0.7676 and 0.9185, respectively, suggesting that the clustering results were convincing. In particular, the silhouette value of each cluster was higher than 0.8, indicating that the co-cited references in one cluster were well matched, with a high heterogeneity. Cluster#0 (tumor microenvironment) was the largest one consisting of 187 co-cited references, followed by Cluster#1(pd-1), Cluster#2 (microglia), Cluster#3 (oncolytic), Cluster#4 (dendritic cells), Cluster#5 (chimeric antigen receptor), Cluster#6 (peptide vaccination), Cluster#7 (epidermal growth factor receptor), Cluster#8 (interleukin-2) and Cluster#9 (bevacizumab). A timeline was used to visualize the references in each cluster and the association between clusters ( Figure 7A ). Cluster#1 had the largest number of periods, while Cluster#8 contained studies that had started earlier. From the perspective of time, Cluster#0, #4, #6-8 all contained early studies with closer connections. From the perspective of label, most of them were early treatment strategies for glioma. However, Cluster#1-3, 5, #9, which demonstrated more and newer references, contained four relatively new therapeutic strategies, namely PD-1, oncolytic therapy (virus therapy), CAR-T therapy and bevacizumab.

The analysis of co-cited references can outline the trends in a certain research field, and predict the research hotspots in the future (14). Its results become more pronounced if further processed with co-cited reference bursts. Selection criteria of co-cited references using CiteSpace were as follows: the number of states=2; γ [0,1] =1.0; minimum duration=2. The top 25 co-citation bursts are shown in Figure 7B . Among them, the first co-citation burst emerged in 2001, when an article was published by Cancer Research, titled “Vacation of malignant glioma patients with peptide-pulsed dendritic cells elicits systemic cytotoxicity and internal T-cell infiltration” Yu JS (33). It reported the feasibility, safety, and biological activity of a DCs-based vaccine in the treatment of malignant glioma. It can be seen that when immunotherapy for glioma had just slipped into the research hotspot in the early 21^st century, DCs were then exploited to design a breakthrough treatment.

It is worth noting that as of 2023, two articles are still in a prominent state, including “Neoadjuvant anti-PD-1 immunotherapy promotes a survival benefit with intratumoral and systemic immune responses in recurrent glioblastoma” authored by Cloughey TFADDIN and published in Nature Medicine in 2019 (34) and “Effect of Nivolumab vs Bevacizumab in Patients With Recurrent Glioblastoma: The CheckMate 143 Phase 3 Randomized Clinical Trial” authored by Reardon DA and published in JAMA Oncology in 2020 (35). Cloughey TF et al. found that neoadjuvant therapy of PD-1 blockers was stronger anti-tumor effects on GBM (34), while Reardon DA et al. confirmed through clinical studies that the safety of PD-1 inhibitor Nivolumab in the treatment of GBM was consistent with those in other tumor types (35). With two reports on PD-1 inhibitors in glioma therapy highlighted during this period, it can be tentatively assumed that PD-1 inhibitors remain in the mainstream of glioma immunotherapy research.

Keywords can also be analyzed to display research hotspots and directions. Keywords that had been cited for more than 30 times were visualized and presented in Figure 8A through VOSviewer. The center node was immunotherapy (total link strength 7724), followed by glioblastoma (total link strength 5596). Figure 8B shows the distribution of keywords over time, and the most recent included “tumor microenvironment”, “prognosis”, “blockade”, “classification”, “pd-l1”, “pembrolizumab” and so on. Figure 8C shows the density distribution of keywords. The three most densely cited keywords were immunology (1273 times), glioblastoma (905 times) and glioma (697 times).

CiteSpace was used to detect keyword bursts ( Figure 8D ), aiming to discover changes in hotspots over time. Three keywords “brain tumor”, “adoptive immunotherapy” and “activated killer cells” made the earliest burst in 1991. From 1990 to 2023, the keywords with the highest burst intensity were “malignant glioma” (28.86) and “dendritic cells” (28.73). The keywords that had burst out after 2016 included “safety”, “nivolumab”, “mismatch repair”, “double blind” and “tumor microenvironment”, with the latter two remaining in the current hotspot.

The combination of reference clustering, timeline, and bursts better pictured the evolution of glioma immunotherapy research.

The analysis of keywords can tease out the hotspots in the research field, while keyword bursts can demonstrate how research hotspots have changed over time. In the early 1990s, “malignant glioma” and “dendritic cells” rushed into the hotspot, and dendritic cells were then used to design vaccines for glioma treatment, because of their antigen-presenting cell characteristics (46). According to VOSviewer’s analysis, popular keywords that are closer to the present time include “tumor microenvironment”, “prognosis”, “blockade”, “classification”, “pd-l1” and “pembrolizumab”. CiteSpace showed that keywords that had burst after 2016 included “safety”, “nivolumab”, “mismatch repair”, “double blind”, and “tumor microenvironment”, with “double blind” and “tumor microenvironment” still in the swim. Based on the outcome of the two software, it is not difficult to find many keywords that are related to the treatment of PD-1, including “PD-L1”, “pembrolizumab” and “nivolumab”, which again suggests that ICIs such as PD-1 is the mainstream immunotherapy for glioma. “Safety” can suggest that, with multiple immunotherapies showing clinical responses for glioma, the survival and ensure the safety of patients should be well concerned, which is also in line with the fundamental principles of drug development. Meanwhile, “tumor microenvironment” was mentioned by both software, reflecting the importance of tumor microenvironment for glioma treatment. Therefore, current immunotherapy research still aims to suppress the immunosuppressive cells by various methods, and to “blockade” the immune escape of the tumor by various small molecules, such as ICI or antibodies. In short, ICIs such as PD-1 are in the current hotspot of glioma immunotherapy. Through ICIs, the immune escape of glioma may be hindered by modulating the tumor microenvironment, thus allowing various therapies to exert their anti-tumor effects. Therefore, the hotspot of future research on glioma immunotherapy will still be ICIs represented by PD-1. Based on the limited number of ICI targets currently available, it is necessary to explore new and more ICI targets and apply them for clinical treatments. In addition, due to the poor efficacy of a single ICI treatment, it is urgent to conduct research on the combination therapy of multiple ICIs and the combination therapy of ICIs and other therapies to increase the anti-tumor effect. In the future, the research of glioma immunotherapy will no longer be limited to the exploration of certain therapeutic methods. It is advisable for researchers to focus on the comprehensive therapy based on ICIs, with the aim of achieving anti-tumor effect by comprehensively inhibiting the tumor microenvironment, so as to improve the prognosis of patients.

There are still several limitations to this study. First of all, all publications pertaining to glioma immunotherapy were searched out of the WoSCC. Although it is one of the most reliable online databases (75), some publications may have been missed. Secondly, only publications in English were collected, which means several potential studies in other languages could be missed. Thirdly, since our study is based on published literature, which might introduce publication bias. Accordingly, we expanded the search scope as much as possible to prevent missing literature in order to reduce publication bias. Fourthly, there may have differences in keywords bursts, clustering analysis of co-citations and institutions contributing to glioma immunotherapy due to the limitations of CiteSpace, VOSviewer and other software.