Research trends in cytokine regulation of immunotherapy in non-small cell lung cancer: a bibliometric and BERTopic analysis from 2015 to 2025

Introduction

Non-small cell lung cancer (NSCLC) is one of the most prevalent malignancies worldwide and remains a leading cause of cancer-related mortality, imposing a substantial burden on global health systems (1-3). In recent years, immune checkpoint inhibitors (ICIs) have reshaped the therapeutic landscape of NSCLC and produced durable clinical benefit in a subset of patients (4,5). However, overall response rates remain suboptimal (6), and both primary and acquired resistance are frequently observed. These limitations suggest that immunotherapeutic responsiveness in NSCLC cannot be explained by checkpoint pathways alone, but rather reflects a complex tumor-immune regulatory network (7).

Within this context, cytokines have emerged as pivotal mediators of NSCLC immunotherapy by coordinating intercellular communication within the tumor immune microenvironment (TIME) (8). Through the regulation of immune-cell recruitment, activation, differentiation, and functional polarization, cytokines dynamically shape the balance between immune activation and immunosuppression (9-11). Accumulating evidence indicates that cytokines contribute not only to antitumor immune responses but also to treatment sensitivity, therapeutic resistance, immune-related adverse events, and the efficacy of combination regimens (12-15).

Since the clinical implementation of ICIs in NSCLC, research in this area has progressively moved beyond pivotal efficacy-oriented trials toward mechanistic and translational investigations of cytokine-mediated TIME regulation. Rather than focusing solely on individual cytokines, recent studies increasingly emphasize cytokine-driven crosstalk among immune-cell populations, including T cells, macrophages, and dendritic cells, and its collective influence on therapeutic responsiveness (16-18). Notably, cytokine-related biomarkers, including interleukin-6 (IL-6) and chemokine signatures, have shown potential for predicting immunotherapy efficacy and prognosis (12,13,19-22). These developments indicate a transition from clinical efficacy-centered immunotherapy research toward a more integrated framework encompassing cytokine-network biology, TIME regulation, resistance mechanisms, biomarker development, and precision-oriented therapeutic strategies.

Despite these advances, research on cytokines in NSCLC immunotherapy has not been systematically synthesized. Existing studies have largely focused on individual cytokines, specific molecular mechanisms, or discrete therapeutic contexts, leaving the broader research architecture—including international collaboration networks, intellectual foundations, and thematic evolution—insufficiently defined. Conventional bibliometric approaches can characterize research output, collaboration patterns, and hotspot distribution; however, they are less effective in capturing latent semantic structures and their temporal dynamics. An integrative analytical framework combining structural mapping with semantic modeling is therefore needed to clarify the developmental trajectory and emerging priorities of this field. The present study integrates bibliometric analysis with Bidirectional Encoder Representations from Transformers-based topic modeling (BERTopic) to systematically examine research on cytokines in NSCLC immunotherapy published between 2015 and 2025. We characterize the global research landscape, collaborative networks, intellectual base, and hotspot evolution through analyses of countries, institutions, authors, journals, co-cited references, and keywords. We further apply BERTopic to identify latent thematic structures and temporal dynamics. By combining structural and semantic perspectives, this study aims to provide a comprehensive overview of the field, identify emerging research priorities, and support the development of more precise immunotherapeutic strategies. We present this article in accordance with the BIBLIO reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1058/rc).

Methods

Data source and search strategy

The data analyzed in this study were retrieved from the Web of Science Core Collection (WoSCC), a bibliographic database widely used in bibliometric research because of its broad coverage and standardized indexing system (23). A topic search (TS) strategy was developed to capture the intersection of NSCLC, immunotherapy, and cytokine-related research. The search syntax comprised three conceptual domains: disease-related terms; immune checkpoint inhibitor-associated terms, including programmed cell death protein 1 (PD-1), programmed cell death ligand 1 (PD-L1), and cytotoxic t-lymphocyte-associated protein 4 (CTLA-4); and cytokine- and chemokine-related terms, including interleukins, interferons, tumor necrosis factor, and chemokines. The complete search query and study selection workflow are shown in Figure 1.

Figure 1 Flowchart of the search strategy and screening process. For country-, institution-, and author-level analyses, a full-counting approach was used. For publications involving multiple countries or institutions, each distinct country or institution listed in the affiliation fields was credited once. Author-level productivity was calculated by crediting each listed author once per publication. Collaboration links were defined by the co-occurrence of countries or institutions in the affiliation fields, or by co-authorship within the same publication. Accordingly, aggregate country-, institution-, or author-level counts may exceed the total number of included publications. BERTopic, Bidirectional Encoder Representations from Transformers-based topic modeling; TS, topic search.

The literature search was conducted on 2 March 2026 as a single-day retrieval to minimize bias introduced by database updates. The time span was restricted to publications from 1 January 2015 to 31 December 2025. This interval was selected to balance field relevance and data completeness. From a field-development perspective, 2015 represented a key inflection point because the first PD-1 ICI for NSCLC was approved in that year (24). Publications before this period were relatively limited and showed insufficient thematic coherence for bibliometric synthesis. The endpoint of 31 December 2025 was chosen to include the most recent complete publication year while avoiding bias associated with incomplete indexing and citation accumulation in 2026. This strategy improved the comparability and stability of longitudinal analyses.

The analysis was restricted to publications indexed as Articles or Reviews, and only English-language studies were included. All records were exported in the “Full Record and Cited References” format. After export, data cleaning was performed, including duplicate removal and standardization of author names, institutional affiliations, and keywords where necessary. Non-research materials, including meeting abstracts, proceedings papers, editorial materials, letters, corrections, and retracted publications, were excluded. The remaining records were independently screened by two investigators according to titles and abstracts, with full texts reviewed when necessary. Discrepancies were resolved through discussion or consultation with a third investigator. Ultimately, 1,186 publications were included in the final analysis.

The inclusion criteria were as follows: (I) studies focused on NSCLC; (II) studies involved ICI-related therapy; and (III) studies included cytokine-related analysis. Records were excluded if they (I) were not directly relevant to NSCLC immunotherapy; (II) did not involve cytokine-related analysis; (III) represented non-research materials, such as abstracts, editorials, or letters; or (IV) were duplicate or incomplete records. Because this study was designed as a bibliometric and topic-modeling analysis based on bibliographic metadata rather than a systematic review or meta-analysis of clinical efficacy, no formal methodological quality assessment or risk-of-bias evaluation was performed for individual publications. This issue is further addressed as a limitation in section “Discussion”.

Bibliometric analysis and visualization

Bibliometric analyses and visualizations were performed using Microsoft Excel 2021, VOSviewer version 1.6.20, CiteSpace version 6.4.R1, SciMago Graphica, and Charticulator. Microsoft Excel 2021 was used to summarize annual and cumulative publication trends from 2015 to 2025. VOSviewer was used to construct collaboration networks among countries, institutions, and authors, as well as keyword co-occurrence networks (25). In these visualizations, node size represents publication output or keyword frequency, links indicate collaborative or co-occurrence relationships, and link strength reflects the degree of association.

To reduce citation-age bias, citation-based indicators were reported both as total citation counts and as annualized citation rates. The annualized citation rate was calculated as total citations divided by the citation window from publication year to the retrieval date, with a minimum window of 1 year. For grouped analyses, average citation indicators were calculated at the publication level and interpreted in conjunction with annualized citation metrics to improve comparability across different publication years.

Country-level analyses were conducted using a full-counting approach based on all author affiliations recorded in WoSCC. For publications involving authors from multiple countries, each distinct contributing country was counted once for that publication, rather than assigning the publication solely to the first author’s or corresponding author’s country. International collaboration was defined as the co-occurrence of two or more countries in the affiliation fields of the same publication. Accordingly, the sum of country-level publication counts may exceed the total number of included publications.

Institution-level analyses were also performed using a full-counting approach. Each distinct institution appearing in the author affiliation fields was credited once for a given publication. For multi-institutional publications, all contributing institutions were included in the institutional productivity and collaboration analyses. Institutional collaboration links were established when two or more institutions appeared in the same publication. Institution names were standardized where necessary by harmonizing spelling variants, abbreviations, and hierarchical forms referring to the same institution.

For author-level analyses, author names were standardized before network construction by unifying capitalization, removing redundant punctuation and spacing, and merging clearly identical name variants. To reduce potential author name ambiguity, particularly among prolific authors and major collaboration nodes, full names, institutional affiliations, research topics, and co-author patterns were further checked where available. Nevertheless, because bibliometric databases do not provide stable unique identifiers for all authors, residual author disambiguation errors could not be completely excluded and are acknowledged as a limitation. CiteSpace was used for co-citation analysis, clustering, and burst detection (26). Time slicing was set from 2015 to 2025, with a slice length of one year. Selection criteria were defined as the top N most cited or most frequent items within each time slice. The Pathfinder algorithm was applied to prune the network and improve structural interpretability. Cluster quality was evaluated using Modularity Q and Silhouette S to assess clustering significance and internal consistency. CiteSpace was also used for dual-map overlay analysis, keyword burst detection, and temporal evolution analysis, thereby enabling the identification of research frontiers and knowledge development pathways. SciMago Graphica was used to visualize the geographic distribution of research output and international collaboration, whereas Charticulator was used to generate chord diagrams illustrating collaborative relationships.

BERTopic-based topic modeling

To identify latent research themes, BERTopic-based topic modeling was performed using the titles and abstracts of the included publications (27). Textual data were preprocessed through case normalization and the removal of stop words and irrelevant symbols. The titles and abstracts used for topic modeling were not manually edited, thereby minimizing subjective intervention in the semantic modeling process. Semantic vector representations were generated using a pretrained Transformer-based embedding model. Uniform Manifold Approximation and Projection (UMAP) was then applied to reduce the dimensionality of high-dimensional embeddings (28,29).

Representative keywords for each topic were extracted using class-based term frequency-inverse document frequency (c-TF-IDF) to characterize semantic features. The identified topics were integrated with publication-year information to evaluate temporal trends and reveal changes in thematic emphasis over time. All analyses were conducted in Python using relevant packages to ensure reproducibility of the modeling process. By integrating bibliometric analysis with semantic modeling, this approach provides a comprehensive understanding of the research structure and developmental trajectory of cytokine-related immunotherapy in NSCLC.

Results

Publication overview and trends

A total of 1,186 publications on cytokine-regulated immunotherapy in NSCLC were included, encompassing contributions from 60 countries, 2,120 institutions, and 10,281 authors across 324 academic journals. These publications cited 39,964 references from 3,358 journals, reflecting a broad and multidisciplinary intellectual foundation. Country-, institution-, and author-level productivity analyses were interpreted according to the full-counting strategy described in section “Methods”; accordingly, publications involving multiple countries, institutions, or authors could contribute to more than one analytical category.

As shown in Figure 2, annual publication output increased substantially between 2015 and 2025, rising from 24 publications in 2015 to 222 in 2025, with a cumulative total of 1,186 publications. The publication trajectory could be broadly divided into three developmental phases. During 2015–2018, the field underwent an initial post-approval expansion stage, with annual output increasing from 24 to 62 publications. During 2019–2021, research activity accelerated, and annual output rose from 85 to 144 publications. During 2022–2025, the field entered a high-output stage: publication output remained stable in 2022 and 2023, with 151 publications in each year, before increasing to 180 in 2024 and 222 in 2025. Polynomial fitting yielded the equation y = 10.12x² − 5.9951x + 21.903, with an R² value of 0.9997, suggesting a strong descriptive fit and the continued expansion of scholarly activity in this research area.

Figure 2 Annual and cumulative publication trends in research on cytokine regulation in NSCLC immunotherapy from 2015 to 2025. Bars indicate annual publication output, and the cumulative curve indicates the overall growth trajectory of the field. NSCLC, non-small cell lung cancer.

Analysis of contributions by countries

At the country level, publication output in cytokine-regulated NSCLC immunotherapy showed a markedly uneven distribution. Country contributions were calculated using all author-affiliated countries recorded in the WoSCC, rather than being assigned only to the first author’s or corresponding author’s country. For publications involving multiple countries, each distinct contributing country was counted once, and international collaboration was defined as the co-occurrence of two or more countries in the affiliation fields of the same publication. Citation-based indicators were interpreted under the same full-counting framework, whereby citations from internationally collaborative publications could be attributed to each contributing country. As shown in the annual publication bubble plot (Figure 3A), research productivity increased across most major contributing countries from 2015 to 2025, with China and the United States consistently ranking as the leading contributors. China showed a particularly pronounced increase in publication output after 2019, followed by sustained growth from 2021 onward. By contrast, the United States maintained a relatively stable but consistently high level of output throughout the study period. Japan, Italy, Germany, Spain, France, South Korea, the United Kingdom, and Switzerland also contributed continuously, although at a smaller scale than China and the United States.

Figure 3 Global distribution and international collaboration of countries in research on cytokine regulation in NSCLC immunotherapy. Country-level analyses were based on a full-counting strategy using all available author affiliations. (A) Annual publication trends of the top contributing countries. Bubble size and color intensity represent publication output. (B) Country collaboration network, in which node size represents publication output and link thickness represents collaboration strength. (C) Geographic distribution of international collaborations. (D) Chord diagram of country collaborations, with links representing country-level co-authorship within the same publication records. NSCLC, non-small cell lung cancer.

According to Table S1, China ranked first in publication output, with 584 publications, followed by the United States with 270 publications. Because raw citation counts are influenced by publication year and citation-window length, the citation impact of countries was interpreted using both absolute and year-normalized indicators. The United States ranked first in total citations [19,250], year-normalized citations [2,734.16], and H-index [70], indicating stronger citation-based impact. Although countries such as the United Kingdom, France, and Spain produced fewer publications, they showed relatively high average citations per publication and year-normalized citation impact. For example, the United Kingdom reached 115.81 average citations per publication and 16.44 average year-normalized citations, suggesting that its representative studies received substantial international attention even after accounting for citation-window effects. By comparison, China had 25.91 average citations per publication and 4.56 average year-normalized citations, indicating that its rapidly expanding publication output was accompanied by a comparatively lower average citation impact.

The national collaboration network further revealed the international structure of research cooperation in this field. In the VOSviewer network (Figure 3B), China and the United States occupied central positions and were represented by large nodes, indicating their important roles in both publication output and international collaboration. Several European countries, including the United Kingdom, France, Germany, Italy, Spain, and Switzerland, were also located near the center of the network and maintained collaborative links with multiple partners. Overall, the country-level collaboration network showed a clear clustering pattern, with several collaborative groups organized around core countries and extensive interconnections across clusters.

The global geographic distribution of collaboration (Figure 3C) showed that research activity was concentrated mainly in Asia, Europe, and North America. China, the United States, and several European countries constituted the principal contributors to the international collaborative landscape. This map was generated using all contributing countries identified from author affiliations; therefore, internationally collaborative publications contributed to each relevant country and to the corresponding country-level collaboration links. China maintained extensive cooperative links with the United States, European countries, and nations in the Asia-Pacific region. Similarly, the United States showed close collaboration with Canada, Europe, and the Asia-Pacific region. Collaboration within Europe was also frequent, indicating a high degree of regional connectivity.

This pattern was further illustrated by the country collaboration chord diagram (Figure 3D), in which China and the United States emerged as the two dominant nodes in the global network. In this diagram, links represent country co-occurrence within the same publication based on all author affiliations. These countries participated in the broadest range of collaborative relationships and maintained stable partnerships with numerous other nations. In addition, substantial collaboration was observed both among European countries and between Europe and the two major contributors. Overall, China ranked first in publication volume, the United States demonstrated greater citation-based impact, and several European countries served as important bridging nodes in the international collaboration network.

Analysis of contributions by institutions

Institution-level analysis showed that this field has developed a broadly interconnected collaborative network with a clear multicenter clustering pattern (Figure 4A). Institutional productivity and collaboration were analyzed using a full-counting approach based on all author affiliations. Thus, when a publication involved multiple institutions, each contributing institution was credited once, and collaboration links were established between institutions appearing in the same publication. Citation-based indicators were interpreted under the same full-counting framework, whereby citations from multi-institutional publications could be attributed to each contributing institution. In the VOSviewer map, institutions such as Sun Yat-sen University, Shanghai Jiao Tong University, Tongji University, Nanjing Medical University, Zhejiang University, and Guangzhou Medical University were represented by relatively large nodes located near the center of the network, indicating both high research activity and central positions in inter-institutional collaboration. Among these institutions, Sun Yat-sen University and Shanghai Jiao Tong University were particularly prominent.

Figure 4 Institutional collaboration and temporal evolution in research on cytokine regulation in NSCLC immunotherapy. Institutional analyses were performed using a full-counting strategy based on all available author affiliations. (A) Institutional collaboration network. Node size represents publication output, link thickness represents collaboration strength, and colors indicate collaboration clusters. (B) Overlay visualization of institutions according to average publication year, with warmer colors indicating more recent research activity. NSCLC, non-small cell lung cancer.

As shown in Table S2, research productivity and citation-based impact varied across institutions. Sun Yat-sen University ranked first in publication output, with 39 publications, whereas The University of Texas MD Anderson Cancer Center had the highest average citations per publication (55.19) and the highest average year-normalized citations (4.60), indicating substantial citation-based impact. When institutional citation performance was interpreted in conjunction with year-normalized indicators, this pattern suggested that citation influence was not determined solely by publication volume. These findings suggest that Chinese universities dominated in publication volume, whereas leading international medical research centers showed higher citation-based performance.

The clustering pattern further highlighted the central role of Chinese universities and research institutes in driving this field. Multiple collaborative groups were formed around core institutions, with frequent links among Sun Yat-sen University, Guangzhou Medical University, Shanghai Jiao Tong University, Zhejiang University, Nanjing Medical University, Soochow University, and Capital Medical University. Meanwhile, internationally recognized institutions, including Harvard Medical School, Dana-Farber Cancer Institute, the University of Pennsylvania, and The University of Texas MD Anderson Cancer Center, also occupied important positions in the network and maintained extensive collaboration with leading Chinese institutions.

Temporal overlay analysis (Figure 4B) suggested that institutions entered this field at different stages. Nodes corresponding to Harvard Medical School, The University of Texas MD Anderson Cancer Center, and the University of Pennsylvania appeared in relatively cooler colors, indicating earlier involvement in this research area. In contrast, institutions such as Nanjing Medical University, Huazhong University of Science and Technology, and Capital Medical University appeared in warmer tones, reflecting more recent research activity. In addition, Tongji University, the Chinese Academy of Sciences, Wuhan University, and the Dana-Farber Cancer Institute showed numerous cross-cluster connections, suggesting that they served as important bridges between different collaborative groups.

Overall, Chinese institutions occupied central positions in both publication output and institutional collaboration, whereas leading international medical research centers played important roles in citation-based impact and cross-regional knowledge exchange.

Analysis of prolific authors and author collaboration

The author collaboration network (Figure 5A) showed that this field has developed into a constellation of relatively stable collaborative teams with a multinodal clustering pattern. Author-level productivity was analyzed using a full-counting approach, whereby each author listed in a publication was credited once, and collaboration links were established on the basis of co-authorship within the same publication. Before author-level analysis, author names were standardized to reduce inconsistencies caused by capitalization, punctuation, spacing, and obvious name variants. For prolific authors and major network nodes, author identity was further checked using available information on full names, affiliations, research topics, and co-author patterns. VOSviewer analysis indicated that authors such as Wang Jing, Zhou Caicun, Ren Xiubao, and Tokito Takaaki were represented by relatively large nodes near the center of the network, suggesting central positions within the collaborative structure. According to Table S3, Zhang Li ranked first in publication output with 15 publications, followed by Wang Jing with 14 publications, and Zhou Caicun and Azuma Koichi with 11 publications each.

Figure 5 Author collaboration patterns and temporal evolution in research on cytokine regulation in NSCLC immunotherapy. Author-level analyses were conducted using all authors listed in each publication, regardless of country affiliation. For publications involving authors from multiple countries, all listed authors were included, and collaboration links were established on the basis of co-authorship within the same publication. Author names were standardized before analysis, and major author nodes were checked using affiliation and co-authorship information where possible. (A) Author collaboration network. Node size represents publication output, link thickness represents collaboration strength, and colors indicate collaboration clusters. (B) Overlay visualization of author activity according to average publication year, with warmer colors indicating more recent activity. NSCLC, non-small cell lung cancer.

Citation-based indicators revealed a partially different pattern from publication productivity. Hellmann Matthew D. had the highest total citations [1,626], average citations per publication [203.25], year-normalized citations [234.06], and average year-normalized citations [29.26], despite contributing fewer publications than the most prolific authors. Lee Se-Hoon and Heymach John V. also showed relatively high citation-based impact, with average year-normalized citations of 16.25 and 13.96, respectively. These findings indicate that author-level productivity and citation-based impact were not fully concordant, underscoring the need to interpret publication counts alongside citation-window-adjusted indicators.

Overall, the field was not dominated by a single large authorship bloc, but rather by multiple medium- and small-scale research teams developing in parallel. Several clearly defined collaborative clusters were observed. One cluster centered on Wang Jing and showed close ties with Wang Qi, Diao Lixia, Gibbons Don L., and Heymach John V. Another cluster, organized around Zhou Caicun, included Li Xuefei, Jiang Tao, Zhao Chao, and Zhou Qing. A further cluster, represented by Tokito Takaaki, reflected a relatively stable Japanese collaborative group comprising Azuma Koichi, Sasada Tetsuro, and Matsuo Norikazu. In addition, Kim Tae Min, Lee Se-Hoon, and Jeon Yoon Kyung formed another comparatively independent collaborative team. These patterns indicate that the authorship structure of this field is largely team-based and led by several core investigators rather than a single dominant center.

Temporal overlay analysis (Figure 5B) provided additional insight into the evolution of the author network. Nodes corresponding to Heymach John V. and Hellmann Matthew D. appeared in cooler colors, indicating earlier involvement in the field. By contrast, Murata Daiki, Murotani Kenta, Song Yong, and Lv Tangfeng appeared in warmer colors, suggesting more recent activity. Notably, many principal authors fell within the blue-to-green range, indicating contributions spanning multiple developmental phases of the field.

Overall, the author collaboration network was characterized by the parallel development of multiple research teams organized around a small number of central investigators. The temporal overlay further indicated that while several core teams remained active over time, new author groups continued to enter the field. Given the inherent possibility of residual author name ambiguity in bibliometric databases, author-level findings should be interpreted as reflecting major collaboration patterns rather than definitive individual-level productivity rankings.

Analysis of contributions by journals

The journal landscape showed a dual structure in this field, comprising both the main publication venues for newly generated findings and the key sources that form the intellectual basis of the field. Research in this area was published mainly in journals related to tumor immunology, general oncology, and thoracic oncology. In the temporal overlay map (Figure 6A), journals such as Journal for ImmunoTherapy of Cancer, Frontiers in Immunology, Cancer Immunology, Immunotherapy, Cancers, Lung Cancer, and Frontiers in Oncology were represented by relatively large nodes near the center of the network, indicating both high publication activity and strong network connectedness. As shown in Table S4, Frontiers in Immunology (77 publications), Journal for ImmunoTherapy of Cancer (74 publications), and Cancers (51 publications) were the most prolific journals in this field.

Figure 6 Journal landscape and knowledge flow in research on cytokine regulation in NSCLC immunotherapy. (A) Overlay visualization of journals according to average publication year. (B) Journal bibliographic coupling network. (C) Dual-map overlay of journals, showing citation pathways between citing and cited journal domains. NSCLC, non-small cell lung cancer.

When considered together with the journal co-citation results presented in Table S4, these findings indicate that journals differed in their roles as publication platforms and sources of intellectual influence. Journals such as Frontiers in Immunology and Journal for ImmunoTherapy of Cancer published a large proportion of the literature and served as major outlets for ongoing research dissemination. By contrast, highly co-cited journals such as the New England Journal of Medicine, Clinical Cancer Research, Nature, and the Journal of Clinical Oncology appeared more prominently in the co-citation network, suggesting that they formed major components of the field’s knowledge base. This divergence suggests that the main publication venues do not entirely overlap with the journals providing the core intellectual foundation of the field.

The journal bibliographic coupling network (Figure 6B) further showed that journals were connected through shared reference profiles and formed several interrelated clusters. These clusters were organized mainly around cancer immunotherapy, lung cancer, molecular mechanisms, and translational medicine, reflecting the interdisciplinary nature of this research domain.

The dual-map overlay (Figure 6C) further supported this cross-disciplinary structure. The literature in this field was published mainly in journals associated with clinical medicine and molecular biology/immunology, whereas its cited knowledge base was derived primarily from journals in molecular biology, genetics, and medicine. The dominant citation trajectories indicated that studies published in the domains of “Medicine, Medical, Clinical” and “Molecular Biology, Immunology” most frequently cited journals in “Molecular Biology, Genetics” and “Health, Nursing, Medicine”, highlighting the close interaction between basic mechanistic research and clinical investigation.

Analysis of references and co-cited references

Co-citation analysis (Figure 7A), together with Table 1, indicated that the intellectual foundation of this field was shaped predominantly by pivotal clinical trials that established the therapeutic role of ICIs in NSCLC. To mitigate bias arising from unequal citation accumulation periods, Table 1 reports both total co-citation counts and year-normalized co-citation indicators, expressed as co-citations per year. The most frequently co-cited references were published mainly in the New England Journal of Medicine and The Lancet, including landmark studies by Borghaei et al. (2015) (30), Brahmer et al. (2015) (31), Garon et al. (2015) (32), Herbst et al. (2016) (33), Reck et al. (2016) (34), Rittmeyer et al. (2017) (35), and Gandhi et al. (2018) (36). Collectively, these studies provided the principal clinical evidence supporting the implementation of PD-1/PD-L1 blockade and formed the central cluster of the co-citation network.

Figure 7 Co-citation structure and frontier evolution in research on cytokine regulation in NSCLC immunotherapy. (A) Co-citation network of references. Node size represents co-citation frequency, and colors indicate temporal distribution. (B) Clustered network of co-cited references. Cluster labels were generated using CiteSpace. (C) Top 25 references with the strongest citation bursts. Red bars indicate burst periods, and cited references may have been published before 2015 because the analysis was based on references cited by publications retrieved from 2015 to 2025. NSCLC, non-small cell lung cancer.

As shown in Table 1, clinical trials accounted for most of the highly co-cited references, underscoring the extent to which this research field has been structured by practice-changing clinical evidence. Reck et al. (34) ranked first in both total co-citations and year-normalized co-citation score, with 203 co-citations over an 11-year citation window and a year-normalized score of 18.45, indicating both cumulative prominence and sustained annual influence. Borghaei et al. (30) and Brahmer et al. (31), both published in 2015, also showed high total co-citation counts; however, their lower year-normalized scores relative to Reck et al. (34) highlight the effect of longer citation windows on cumulative co-citation metrics. By contrast, Gandhi et al. (2018) (36) and Sung et al. (2021) (37) achieved relatively high year-normalized scores despite shorter citation windows, suggesting rapid assimilation into subsequent research. Notably, Sung et al. (37) ranked eighth by total co-citations but second by year-normalized co-citation score, indicating the increasing relevance of epidemiological context within the recent knowledge base of this field.

The combined interpretation of total co-citations and year-normalized co-citation scores provides a more discriminating assessment of reference influence. Total co-citation counts primarily identify long-established landmark studies, whereas year-normalized scores reveal references with rapid or continuing annual impact. Pardoll’s conceptual review on immune checkpoint blockade, for example, remained among the top co-cited references, yet its relatively modest year-normalized score reflects a longer citation window and suggests a foundational rather than rapidly expanding influence. Taken together, these findings indicate that the intellectual base of cytokine-related NSCLC immunotherapy remains anchored in landmark ICI trials, while recent scholarly attention has increasingly incorporated global cancer burden, combination treatment strategies, and broader translational interpretation.

Beyond clinical trials, several mechanistic studies and review articles also occupied important positions in the intellectual structure of the field. For example, the study by Rizvi NA (2015) (38), published in Science, provided a mechanistic interpretation of immunotherapy responsiveness from the perspective of tumor mutational burden, whereas the review by Pardoll DM (2012) (39) in Nature Reviews Cancer presented a conceptual framework for immune checkpoint blockade. Although not directly focused on immunotherapy mechanisms, the epidemiological analysis by Sung H (2021) (37) in CA: A Cancer Journal for Clinicians provided important disease-burden context for subsequent studies in this area. These distinct categories of literature formed a layered knowledge structure that integrated clinical evidence, molecular mechanisms, conceptual frameworks, and epidemiological context.

Reference clustering analysis (Figure 7B) further revealed the internal organization of this knowledge base. The clustering network yielded a Modularity Q of 0.7178 and a weighted mean silhouette score of 0.861, indicating a well-defined clustering structure with strong internal consistency. Major cluster labels included prospect, EGFR-mutated NSCLC, combination therapy, PD-1 resistance, PD-L2 expression, T cell, dendritic cell, B cell, signaling pathways and metabolism, and cGAS-STING pathway strategies. These themes indicate that the field has been organized mainly around optimization of immunotherapeutic strategies, resistance mechanisms, oncogenic-driver-defined subtypes, immune-cell regulation, and signaling- and metabolism-related pathways.

Burst-reference analysis (Figure 7C) revealed a stage-specific evolution of research frontiers. Early citation bursts were concentrated in seminal studies such as Brahmer J (2015) (31), Garon EB (2015) (32), Borghaei H (2015) (30), and Rizvi NA (2015) (38), highlighting the foundational role of key ICI clinical trials. This was followed by burst references such as Reck M (2016) (34), Rittmeyer A (2017) (35), Gainor JF (2016) (40), and Ayers M (2017) (41), reflecting increased attention to first-line treatment strategies, predictive biomarkers, and resistance-related mechanisms. More recent burst references, including Sung H (2021) (37), Reck M (2022) (34), Thai AA (2021) (42), and Forde PM (2022) (43), indicate that the research frontier has expanded toward perioperative immunotherapy, combination treatment paradigms, and more refined clinical strategies.

Taken together, the evolution of the field’s intellectual base followed a discernible trajectory: it was initially anchored in landmark clinical trial literature, subsequently expanded to mechanistic interpretation and precision-oriented therapeutic research, and more recently extended to perioperative treatment and combination strategy development.

Keyword clustering and co-occurrence analysis

The keyword timezone map (Figure 8A) illustrated the temporal evolution of research priorities in this field. During 2015–2017, the earliest and most prominent keywords included NSCLC, immunotherapy, PD-L1 expression, T cells, dendritic cells, interferon (IFN)-γ, and chemotherapy, indicating that early studies focused mainly on the basic concepts of immunotherapy, immune-cell function, and related mechanisms during the early post-approval expansion of ICIs in NSCLC. Between 2018 and 2021, keywords such as ICIs, TIME, pembrolizumab, resistance, CD8⁺ T cells, tumor-infiltrating lymphocytes, and IL-6 became increasingly prominent, suggesting a shift toward checkpoint-based therapeutic strategies, TIME regulation, and resistance-related mechanisms. Since 2022, keywords including cytokines, chemokines, macrophages, transforming growth factor (TGF)-β, mutations, CTLA-4, immune escape, STING, combination therapy, and durvalumab have become more concentrated on the right side of the timezone map, indicating that recent research has increasingly focused on cytokine networks, immune microenvironment remodeling, immune evasion, molecular stratification, and combination strategies.

Figure 8 Keyword structure and evolution of research hotspots in cytokine regulation of NSCLC immunotherapy. (A) Time-zone view of keywords, showing the temporal distribution of major research terms. Node size represents keyword frequency, and links represent keyword co-occurrence. (B) Top 25 keywords with the strongest bursts. Red bars indicate burst periods. (C) Keyword clustering network. Node size represents keyword co-occurrence frequency, and colors indicate thematic clusters. NSCLC, non-small cell lung cancer.

Keyword burst analysis (Figure 8B) was consistent with these temporal patterns. Early burst keywords included dendritic cells, IFN-γ, lymphocytes, chemotherapy, and cytokine-induced killer cells, corresponding to an initial post-approval period characterized by immune-cell functional studies and early clinical translation of immunotherapeutic approaches. Subsequent bursts involving metastasis, CD8⁺ T cells, docetaxel, apoptosis, ipilimumab, and lung neoplasms suggested a gradual expansion toward clinical treatment settings and associated mechanistic questions. More recent burst keywords included TGF-β, macrophages, mutations, CTLA-4, IL-6, cytokines, identification, and landscape. Notably, the bursts for cytokines, identification, and landscape persisted through 2025, indicating that characterization of cytokine-network features and the broader immune landscape remain an active research frontier.

Keyword clustering analysis (Figure 8C) further showed that the thematic structure of the field was both clearly defined and highly coherent, with a Modularity Q of 0.7861 and a weighted mean silhouette score of 0.9543. Based on the cluster labels, the principal research themes centered on ICIs and related therapeutic strategies, while extending to inflammatory responses, the tumor microenvironment, biomarker discovery, and prognostic evaluation. The appearance of themes such as PD-1, pembrolizumab, nivolumab, tumor microenvironment, and biomarker further indicates that the field is characterized by a dual focus on clinical application and mechanistic investigation.

Viewed across the time-zone map, burst analysis, and clustering results, the field has evolved from a post-approval stage anchored in ICI efficacy and immune-cell function toward a broader framework encompassing TIME regulation, resistance mechanisms, cytokine-network biology, immune evasion, biomarker development, and combination or precision-oriented therapeutic strategies.

BERTopic-based topic modeling analysis

To further characterize the latent semantic structure of research on cytokine involvement in NSCLC immunotherapy, BERTopic-based topic modeling was performed on the textual corpus. This analysis identified eight principal topics (topics 0–7). As shown in the topic distribution map (Figure 9A), these topics occupied relatively discrete but closely adjacent regions within the two-dimensional semantic space, suggesting both thematic differentiation and semantic continuity across the field.

Figure 9 BERTopic-based thematic structure and evolution in research on cytokine regulation in NSCLC immunotherapy. (A) Intertopic distance map illustrating the spatial distribution of the identified topics in semantic space. Each circle represents one topic, and the distance between circles reflects the degree of semantic similarity. (B) Word clouds of representative terms for each topic generated from c-TF-IDF. (C) Cosine similarity heatmap showing semantic relationships among topics, with darker colors representing higher similarity. (D) Distribution of c-TF-IDF weights of representative keywords across topics. (E) Temporal trends in topic prevalence from 2015 to 2025. BERTopic, Bidirectional Encoder Representations from Transformers-based topic modeling; c-TF-IDF, class-based term frequency-inverse document frequency; NSCLC, non-small cell lung cancer.

The topic word clouds in Figure 9B highlighted several recurrent semantic domains, including PD-1/PD-L1 immune checkpoint blockade, patient-centered clinical outcomes, tumor and immune-cell interactions, response prediction, survival prognosis, immune-related adverse events, tumor microenvironment regulation, and combination treatment strategies. High-weight terms such as “cell”, “patient”, “cancer”, “tumor”, and “immune” constituted the core semantic framework of the corpus, whereas terms related to biomarkers, single-cell analysis, resistance, and combination therapy reflected the increasing integration of translational and mechanistic research. These findings indicate that BERTopic captured a research landscape centered on clinical immunotherapy application, refined by TIME analysis, and increasingly oriented toward biomarker-guided therapeutic optimization.

The topic similarity heatmap (Figure 9C) provided further insight into the relationships among topics. Overall, cosine similarity values were moderate, indicating that the topics were not isolated but retained varying degrees of conceptual overlap. Several topic pairs showed darker colors, indicating stronger semantic similarity, which mainly reflected intersections between clinically oriented treatment research and studies of the immune microenvironment, cytokine regulation, and resistance biology. In contrast, some topics showed relatively lower similarity to the others, suggesting that they represented more specialized research directions. The close spatial proximity of certain topics in Figure 9A further supports the presence of semantic continuity across parts of the thematic landscape. The topic structure identified by BERTopic was therefore consistent with the developmental trajectory suggested by reference clustering, namely a progression from clinical trial evidence toward mechanistic interpretation and integrated therapeutic strategies.

The c-TF-IDF weight distribution (Figure 9D) provided additional insight into the internal composition of each topic by showing differences in the relative importance of representative keywords. In some topics, the highest-ranked terms had markedly greater weights, followed by a steep decline, suggesting a relatively concentrated semantic core. In others, keyword weights declined more gradually, indicating a broader thematic scope and more complex semantic architecture. This pattern suggests that the field contains both focused topics centered on specific therapeutic modalities or biological mechanisms and broader integrative topics spanning multiple dimensions of immune regulation and clinical application.

The annual topic trend plot (Figure 9E) further captured the temporal evolution of thematic emphasis. Early-stage topics were dominated by the clinical application of ICIs and treatment-response evaluation in the post-approval development period of NSCLC immunotherapy. This was followed by a gradual rise in themes related to TIME, immune resistance, and response prediction. In more recent years, topics associated with cytokine networks, chemokines, macrophages, immune evasion, and combination therapeutic strategies have become increasingly prominent. This temporal pattern indicates that the field has gradually shifted from an early stage driven primarily by clinical trial evidence toward one increasingly centered on mechanistic investigation and precision immunotherapeutic optimization, in agreement with the trajectories inferred from co-cited references, burst references, and keyword burst analyses.

At the semantic level, Figure 9A-9E indicates that this field has developed into a multi-thematic research architecture grounded in the clinical application of ICIs, expanded through studies of immune microenvironment remodeling and cytokine-mediated regulation, and increasingly oriented toward combination treatment and precision immunotherapy. These findings reinforce the broader pattern of knowledge evolution revealed by the preceding analyses of countries, institutions, authors, journals, references, and keywords.

Discussion

Main findings

Over the past decade, research on cytokine involvement in NSCLC immunotherapy has not emerged as a field with fixed boundaries from the outset; instead, it has progressively taken shape in response to clinical questions generated by the expanding use of ICIs. The sustained increase in publication output, particularly the marked acceleration after 2019, reflects more than a simple accumulation of studies. It indicates that cytokine-related questions have shifted from a peripheral explanatory role to a more central position in immunotherapy research, with direct relevance to treatment heterogeneity, resistance evolution, and therapeutic optimization (10).

This transition reflects the changing clinical agenda of NSCLC immunotherapy. Following the approval and expanding clinical implementation of ICIs in NSCLC, earlier studies in this time window focused primarily on whether PD-1/PD-L1 blockade could improve objective response and survival (4), that is, whether immunotherapy could establish itself as a valid clinical treatment paradigm. As the use of ICIs expanded from later-line settings to first-line treatment, perioperative application, and multimodal combinations (44), the clinical questions became more specific and less amenable to explanation by a single pathway. Marked interpatient heterogeneity in response, the emergence of acquired resistance after initial benefit, and treatment failure among patients expected to respond all required an analytical framework extending beyond checkpoint signaling alone. In this context, cytokines are important not only because they are numerous, but because they occupy a strategic position at the interface between tumor cells, effector lymphocytes, suppressive myeloid populations, and the broader immune niche, thereby linking clinical observations with mechanistic interpretation (12,45,46).

The collaboration pattern further supports this interpretation. China ranked first in publication volume, whereas the United States demonstrated stronger citation-based impact, as reflected by total citations and H-index. When interpreted together with year-normalized citation indicators, this pattern suggests that the field is being advanced by both rapid expansion in publication output and highly cited contributions from established research systems. Notably, the institutional and author networks did not show the characteristics of a field dominated by a single center. They were instead characterized by multiple nodes, multiple clusters, and broad cross-disciplinary collaboration. Such a network structure is consistent with a field that depends simultaneously on clinical resources, translational platforms, molecular investigation, and computational analysis. The findings therefore capture not only a growing research topic but also a knowledge domain increasingly organized around complex tumor-immune regulatory systems.

Interpretation of hotspot evolution

The evolution of keywords reveals not only which terms have gained prominence, but also how the field has reoriented its explanatory priorities. Early high-frequency and burst keywords clustered around NSCLC, immunotherapy, PD-L1 expression, T cells, dendritic cells, IFN-γ, and chemotherapy (47). This pattern suggests that the central concern during the initial post-approval expansion period of ICIs in NSCLC was whether immunotherapy could achieve clinical legitimacy in NSCLC and what basic immunological rationale supported this possibility. During this phase, cytokines functioned mainly as contextual factors or auxiliary explanatory variables. They were present in the research landscape, but they had not yet become central to the field’s interpretive framework.

As the focus gradually shifted toward ICIs, TIME, resistance, CD8⁺ T cells, tumor-infiltrating lymphocytes, and IL-6, the internal structure of the research questions changed accordingly. The core issue was no longer whether ICIs worked, but why treatment responses varied so markedly across patients. IL-6 and CD8⁺ T cells became important not only because they were measurable, but because together they moved the field beyond a binary drug-tumor framework toward a more layered model involving treatment exposure, immune-cell state, and microenvironmental regulation (19,22,48-50). During this period, response and resistance increasingly came to be understood as properties of a dynamic immune ecosystem rather than as consequences of a single checkpoint molecule.

The distribution of keywords in more recent years extends this shift further. The sustained prominence of cytokines, chemokines, macrophages, TGF-β, mutations, CTLA-4, immune escape, and combination therapy suggests that the field is now less concerned with simply enhancing immune activation than with understanding how tumors establish and maintain an immunosuppressive state, and whether that state can be systematically reconfigured (51,52). The growing salience of macrophages and TGF-β is particularly noteworthy (53,54). These terms point not merely to insufficient immune activation, but to the processes through which suppressive microenvironments are shaped, maintained, and reorganized under therapeutic pressure. The emergence of STING and combination therapy marks a further transition from mechanistic description to intervention design. Researchers are increasingly seeking actionable leverage points through which innate immune activation, cytokine-axis reprogramming, or multi-pathway blockade may disrupt persistent immune suppression or immune escape within the TIME.

This keyword migration reflects a broader shift in the explanatory framework. In earlier phases, NSCLC immunotherapy could be discussed within a relatively simple “single pathway-single outcome” model, in which PD-L1, tumor mutational burden, and T-cell exhaustion carried much of the explanatory weight. Increasingly, however, this model has become insufficient. Treatment response is more accurately understood as the outcome of a complex regulatory architecture involving cytokines, chemokines, effector lymphocytes, suppressive myeloid cells, tumor-intrinsic signaling, and local metabolic conditions. Cytokines are becoming central not because they represent another fashionable molecular class, but because they lie at the intersection of this multidimensional regulatory network (55).

This shift also clarifies the most promising directions for future research. Studies with genuine developmental potential are unlikely to stop at extending ICI indications or identifying additional single predictive markers. Future work is more likely to move toward finer immune stratification and more deliberate microenvironmental intervention. IL-6, TGF-β, chemokine signatures, macrophage-associated signals, and innate immune activation pathways recur repeatedly because they are directly connected to the major unresolved questions in NSCLC immunotherapy: which patients benefit, which patients develop resistance, which patients require combination strategies, and which therapeutic routes deserve priority. In this sense, keyword analysis offers not simply a map of changing topics, but a trajectory showing how the field has moved from efficacy validation to network-level regulation.

Topic structure and translational implications

If keyword analysis captures the movement of explicit research hotspots, BERTopic reveals how these hotspots are organized and stabilized at a deeper semantic level. In a field spanning clinical oncology, tumor immunology, molecular mechanisms, and bioinformatics, keywords alone are insufficient to characterize the internal knowledge structure. High-frequency terms do not necessarily correspond to stable research problems, and co-occurrence among terms does not necessarily reveal the relationships among the underlying questions. Topic modeling is therefore valuable because it identifies the semantic units around which the literature is collectively organized, rather than merely counting recurrent terms.

The principal topics identified in this study centered on ICI application, the TIME, cytokine/chemokine regulation, immune resistance, combination therapy, and biomarkers. These findings support the results of keyword analysis, while BERTopic provides additional structural insight. The topic distribution map showed that the identified topics were spatially separated within semantic space while remaining closely adjacent. This pattern indicates that although the field has differentiated into several relatively distinct research modules, these modules do not have rigid boundaries. Research on clinical treatment, the microenvironment, molecular mechanisms, and therapeutic optimization does not proceed as a set of parallel and isolated tracks. Instead, these lines of inquiry remain in continuous semantic contact. This feature closely reflects the reality of NSCLC immunotherapy research, in which many key questions are inherently composite, including who develops resistance, how combination strategies should be designed, and which cytokines may serve as predictive or stratification tools. These questions lie at the intersection of clinical and mechanistic reasoning.

The topic similarity heatmap further clarifies these relationships. The relatively high similarity between clinically oriented topics and those focused on cytokine regulation, the immune microenvironment, and resistance biology suggests that the center of gravity of the field has not simply shifted from clinical research to basic research (46,56,57). Rather, clinical problems have driven inquiry deeper into mechanism, and mechanistic exploration has in turn reshaped the way clinical questions are formulated. This distinction is important because cytokine research does not merely add mechanistic detail to an established immunotherapy framework; it occupies the interface where clinical efficacy, resistance development, and therapeutic optimization converge, thereby functioning as a translational bridge.

Annual topic trends place this structure in a temporal sequence. Earlier topics were concentrated more heavily on the clinical application of ICIs and treatment-response evaluation during the post-approval development of NSCLC immunotherapy, whereas later phases expanded toward microenvironmental regulation, cytokine networks, resistance, and combination strategies. This pattern is better understood not as a simple replacement of one set of hotspots by another, but as a reorganization of the field’s internal mode of knowledge production: initially dominated by drug-efficacy questions, then by response heterogeneity-microenvironmental interpretation, and later by network regulation-strategy design. In this respect, BERTopic does more than rename successive phases. It integrates dispersed changes in keywords into a coherent semantic framework of evolution. This directly addresses the methodological question raised at the outset of this study, namely whether conventional bibliometric approaches are sufficient to uncover latent thematic structure. The present findings suggest that BERTopic provides a substantive complement to conventional keyword analysis in understanding the knowledge structure of the field.

The topic model also indicates that this research system is not organized around a single linear problem, but around a knowledge network characterized by multi-problem coupling and cross-module interaction. The most meaningful innovations are unlikely to arise from isolated topics. Instead, they are more likely to emerge at the intersections between topics, such as where cytokine networks intersect with resistance biology, where macrophage-related themes overlap with combination therapy, or where chemokine patterns intersect with response stratification. This observation has direct implications for future study design. Keywords can show which concepts are becoming prominent, whereas topic models are better suited to revealing which problems are already being discussed together and which latent connections remain underdeveloped. In this field, the most promising opportunities for conceptual advance may lie precisely in these zones of overlap.

Future directions

When keyword analysis and BERTopic are interpreted together, the broad contours of the current research frontier become clear. One central frontier concerns cytokine networks and microenvironmental remodeling. Whether considered from the prominence of keywords such as cytokines, chemokines, macrophages, and TGF-β, or from BERTopic themes clustered around immune regulation and the TIME, the same conclusion emerges: the efficacy of ICIs is determined not only by checkpoint blockade itself, but also by whether the TIME permits a durable and effective immune response to be sustained. The emphasis of the field is therefore shifting from how to initiate antitumor immunity to how to reconfigure an ecological niche that persistently maintains immune suppression.

A second frontier is resistance and immune escape. Keywords such as resistance, immune escape, and mutations, together with semantic clusters around response, biomarkers, and immune regulation, indicate that therapeutic heterogeneity and resistance development remain among the strongest driving forces in the field. Cytokines are important in this context because they participate in both response initiation and resistance maintenance. They influence the recruitment of effector cells while also contributing to the stabilization of suppressive myeloid populations. Accordingly, investigating resistance through the lens of cytokine communication is not merely a matter of filling mechanistic gaps but of revising the explanatory model itself.

A third frontier, perhaps closest to translational breakthrough, is combination therapy and precision intervention. The continued prominence of STING, CTLA-4, and combination therapy suggests that major future advances are unlikely to arise from simple extensions of existing ICI classes. Instead, they are more likely to emerge from combinatorial strategies directed at cytokine networks, innate immune pathways, and suppressive microenvironments. In other words, the field is moving from demonstrating that ICIs work to designing more adaptive and biologically appropriate immunotherapeutic systems. This transition underscores that the aims of the present study—tracing evolutionary pathways, identifying latent themes, and defining frontier directions—are not merely descriptive, but directly relevant to the ongoing reorganization of knowledge in this field.

Strengths and limitations

This study has several strengths. It provides a systematic overview of cytokine-related immunotherapy research in NSCLC over an 11-year period corresponding to the major post-approval development of ICIs in this disease. By combining conventional bibliometric analysis with BERTopic-based semantic modeling, the study captures both visible research structures and latent thematic organization. It also integrates multiple analytical dimensions, including countries, institutions, authors, journals, co-cited references, keywords, and topic evolution, thereby offering a comprehensive view of the field’s knowledge architecture and developmental trajectory. Several limitations should be acknowledged. The data source was restricted to the WoSCC, which ensures consistency in bibliographic quality and citation indexing but inevitably excludes potentially relevant studies indexed in other databases. The corpus was also limited to English-language articles and reviews, introducing unavoidable language and publication bias. Because this study was designed as a bibliometric and topic-modeling analysis rather than a systematic review or meta-analysis of clinical outcomes, no formal methodological quality assessment or risk-of-bias evaluation was conducted for individual publications. Therefore, the results should be interpreted as reflecting research activity, knowledge structure, and thematic evolution rather than the clinical validity or methodological rigor of each included study.

Bibliometric results are also influenced by database update cycles, citation accumulation, and parameter settings; consequently, network structure and clustering outcomes may vary slightly across time points or analytic conditions. Although annualized citation indicators were used to reduce the influence of different citation accumulation periods, citation-based metrics remain imperfect proxies for academic quality and may still favor older, larger, or more clinically visible studies. Country-level and institution-level analyses were based on a full-counting strategy using all available author affiliations. This approach better captures collaborative contributions, but it may cause the sum of country or institution counts to exceed the total number of publications. Institutional name variants and incomplete affiliation records may also introduce residual classification errors despite standardization. Author-level analyses are subject to potential name ambiguity, especially for authors with common names or inconsistent initials across records. Although author names were standardized and major nodes were checked using affiliation and co-authorship information where possible, residual disambiguation errors cannot be entirely excluded. Finally, this study focuses on knowledge structure and research ecology rather than direct biological causality. How cytokines shape treatment response and resistance in NSCLC still requires further validation through experimental studies, clinical cohorts, and multi-omics evidence.

Conclusions

By integrating bibliometric analysis with BERTopic-based topic modeling, this study delineates the knowledge structure, developmental trajectory, and shifting research priorities of cytokine-related immunotherapy in NSCLC from 2015 to 2025. The field has expanded steadily during this period, with its emphasis moving from clinical evidence anchored in post-approval ICI trials and early validation of immune checkpoint blockade toward a more integrated framework encompassing cytokine networks, TIME remodeling, resistance mechanisms, biomarker development, and combination therapeutic strategies. The findings further suggest that this domain has developed into a multicenter, interdisciplinary, and internationally collaborative research landscape, rooted in landmark clinical trials yet increasingly directed toward mechanistic investigation and precision immunotherapy. Collectively, this study provides a structured account of the field and offers a reference for future research on cytokine-mediated immune regulation and the optimization of immunotherapeutic strategies in NSCLC.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the BIBLIO reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1058/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1058/prf

Funding: This study was supported by the Cangzhou Key Research and Development Program (No. 222106036).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1058/coif). All authors report that this study was supported by the Cangzhou Key Research and Development Program (No. 222106036). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Wang H, Niu X, Jin Z, et al. Immunotherapy resistance in non-small cell lung cancer: from mechanisms to therapeutic opportunities. J Exp Clin Cancer Res 2025;44:250. [Crossref] [PubMed]
2. Zhou CY, Qi YF, Li HJ, et al. Current and future immunotherapies for NSCLC. J Hematol Oncol 2026;19:28. [Crossref] [PubMed]
3. Hou AY, Elantably D, Manochakian R, et al. Emerging Immunotherapy and Antibody-Derived Therapeutics for the Treatment of Advanced Non-Small-Cell Lung Cancer: A Review. Cancers (Basel) 2026;18:1291. [Crossref] [PubMed]
4. Shi X, Deng W, Pan Y, et al. Overcoming resistance to immune checkpoint inhibitor in non-small cell lung cancer: the promise of combination therapy. Front Immunol 2025;16:1691980. [Crossref] [PubMed]
5. Li B, Ren Y, Zhang X. The application and challenges of immune checkpoint inhibitors in lung cancer therapy. Front Immunol 2026;17:1811169. [Crossref] [PubMed]
6. Mariniello A, Borgeaud M, Weiner M, et al. Primary and Acquired Resistance to Immunotherapy with Checkpoint Inhibitors in NSCLC: From Bedside to Bench and Back. BioDrugs 2025;39:215-35. [Crossref] [PubMed]
7. Gomatou G, Charpidou A, Li P, et al. Mechanisms of primary resistance to immune checkpoint inhibitors in NSCLC. Clin Transl Oncol 2025;27:1426-37. [Crossref] [PubMed]
8. Rahal Z, El Darzi R, Moghaddam SJ, et al. Tumour and microenvironment crosstalk in NSCLC progression and response to therapy. Nat Rev Clin Oncol 2025;22:463-82. [Crossref] [PubMed]
9. Racacho KJ, Shiau YP, Villa R, et al. The tumor immune microenvironment: implications for cancer immunotherapy, treatment strategies, and monitoring approaches. Front Immunol 2025;16:1621812. [Crossref] [PubMed]
10. Wang X, Wang J, Wang Q, et al. Rewiring tumor cytokine networks to enhance immune checkpoint blockade: mechanisms, engineering, and clinical translation. J Exp Clin Cancer Res 2026;45:65. [Crossref] [PubMed]
11. Aghaei M, Bahreiny SS, Mahdizade AH, et al. Cytokine-Mediated tumor sensitization: mechanistic frameworks and therapeutic opportunities in cancer Immunotherapy Clin Exp Med 2026;26:135. [Crossref] [PubMed]
12. Yang L, Fan Z, Wang Z, et al. Soluble cytokines and chemokines in NSCLC: drivers of immune evasion and angiogenesis. Front Immunol 2026;17:1699065. [Crossref] [PubMed]
13. Jain K, Goel A, Mehra D, et al. Cytokine profiling identifies circulating IL-2, IL23 and sPD-L1 as prognostic biomarkers for treatment outcomes in non-small cell lung cancer patients undergoing anti-PD1 therapy. Front Oncol 2025;15:1628379. [Crossref] [PubMed]
14. Murakami K. Comprehensive management of immune-related adverse events: predictive strategies, mechanistic insights, and therapeutic approaches. Immunological Medicine 2026;1-15. [Crossref]
15. Han J, Yang Z, Zhao H. Lung cancer immunotherapy in 2025: where we stand and what comes next? Front Immunol 2025;16:1728163. [Crossref] [PubMed]
16. Jiang J, Zhong M, Wang J, et al. Tumor immune microenvironment in non-small cell lung cancer progression. Front Immunol 2026;17:1778251. [Crossref] [PubMed]
17. Guevara Lopez ML, Gebo A, Parodi M, et al. CD56(bright) cytokine-induced memory-like NK cells and NK-cell engagers synergize against non-small cell lung cancer cancer-stem cells. J Immunother Cancer 2025;13:e010205. [Crossref] [PubMed]
18. Zhang G, Wu Y, Qi D, et al. M2 macrophages modulate the differentiation of CD8 + CD101-TIM3 + T cells via the SPP1-CD44 pathway, influencing the immunotherapeutic response in NSCLC. J Transl Med 2026;24:134. [Crossref] [PubMed]
19. Wang Y, Guo Z, Xu A, et al. Mechanistic insights into IL-6-mediated NK cell dysfunction in NSCLC via the IRE1α-XBP1s-STAT3-UBE2S axis. NPJ Precis Oncol 2025;9:361. [Crossref] [PubMed]
20. Galera P, Iglesias-Beiroa A, Hernández-Marín B, et al. Chemokine Receptors in Peripheral Blood Mononuclear Cells as Predictive Biomarkers for Immunotherapy Efficacy in Non-Small Cell Lung Cancer. Curr Oncol 2025;32:583. [Crossref] [PubMed]
21. Grecea-Balaj AM, Soritau O, Brie I, et al. Serum TNF -α, IL-10 and IL-2 Trajectories and Outcomes in NSCLC and Melanoma Under Anti-PD-1 Therapy: Longitudinal Real-World Evidence from a Single Center. Curr Issues Mol Biol 2025;47:746. [Crossref] [PubMed]
22. Liu Q, Shaibu Z, Xu A, et al. Predictive value of serum cytokines in patients with non-small-cell lung cancer receiving anti-PD-1 blockade therapy: a meta-analysis. Clin Exp Med 2025;25:59. [Crossref] [PubMed]
23. Feng J, He X, Li R, et al. Trends and hotspots in gut microbiota and neonatal necrotizing enterocolitis: a bibliometric analysis. Front Cell Infect Microbiol 2026;16:1684491. [Crossref] [PubMed]
24. Kazandjian D, Suzman DL, Blumenthal G, et al. FDA Approval Summary: Nivolumab for the Treatment of Metastatic Non-Small Cell Lung Cancer With Progression On or After Platinum-Based Chemotherapy. Oncologist 2016;21:634-42. [Crossref] [PubMed]
25. Mai W, Chen X, Ye W, et al. Studies on allergic diseases and B cells in the past 20 years: a bibliometric analysis via CiteSpace and VOSviewer. Front Immunol 2026;17:1760388. [Crossref] [PubMed]
26. Chen Y, Chen T, Tian S, et al. Berberine-gut microbiota interactions based on CiteSpace: current research status and hotspots. Front Microbiol 2026;17:1751887. [Crossref] [PubMed]
27. Bae S, Kim Y, Park S, et al. Predicting atrial fibrillation and flutter using BEHRT and identifying multimorbidity patterns using BERTopic. Front Digit Health 2026;8:1722338. [Crossref] [PubMed]
28. Guo Q, Hernandez-Hernandez S, Ballester PJ. UMAP-based clustering split for rigorous evaluation of AI models for virtual screening on cancer cell lines. J Cheminform 2025;17:94. [Crossref] [PubMed]
29. Castellano-Escuder P, Zachman DK, Han K, et al. GAUDI: interpretable multi-omics integration with UMAP embeddings and density-based clustering. Nat Commun 2025;16:5771. [Crossref] [PubMed]
30. Borghaei H, Paz-Ares L, Horn L, et al. Nivolumab versus Docetaxel in Advanced Nonsquamous Non-Small-Cell Lung Cancer. N Engl J Med 2015;373:1627-39. [Crossref] [PubMed]
31. Brahmer J, Reckamp KL, Baas P, et al. Nivolumab versus Docetaxel in Advanced Squamous-Cell Non-Small-Cell Lung Cancer. N Engl J Med 2015;373:123-35. [Crossref] [PubMed]
32. Garon EB, Rizvi NA, Hui R, et al. Pembrolizumab for the treatment of non-small-cell lung cancer. N Engl J Med 2015;372:2018-28. [Crossref] [PubMed]
33. Herbst RS, Baas P, Kim DW, et al. Pembrolizumab versus docetaxel for previously treated, PD-L1-positive, advanced non-small-cell lung cancer (KEYNOTE-010): a randomised controlled trial. Lancet 2016;387:1540-50. [Crossref] [PubMed]
34. Reck M, Rodríguez-Abreu D, Robinson AG, et al. Pembrolizumab versus Chemotherapy for PD-L1-Positive Non-Small-Cell Lung Cancer. N Engl J Med 2016;375:1823-33. [Crossref] [PubMed]
35. Rittmeyer A, Barlesi F, Waterkamp D, et al. Atezolizumab versus docetaxel in patients with previously treated non-small-cell lung cancer (OAK): a phase 3, open-label, multicentre randomised controlled trial. Lancet 2017;389:255-65. [Crossref] [PubMed]
36. Gandhi L, Rodríguez-Abreu D, Gadgeel S, et al. Pembrolizumab plus Chemotherapy in Metastatic Non-Small-Cell Lung Cancer. N Engl J Med 2018;378:2078-92. [Crossref] [PubMed]
37. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
38. Rizvi NA, Mazières J, Planchard D, et al. Activity and safety of nivolumab, an anti-PD-1 immune checkpoint inhibitor, for patients with advanced, refractory squamous non-small-cell lung cancer (CheckMate 063): a phase 2, single-arm trial. Lancet Oncol 2015;16:257-65. [Crossref] [PubMed]
39. Pardoll DM. The blockade of immune checkpoints in cancer immunotherapy. Nat Rev Cancer 2012;12:252-64. [Crossref] [PubMed]
40. Gainor JF, Shaw AT, Sequist LV, et al. EGFR Mutations and ALK Rearrangements Are Associated with Low Response Rates to PD-1 Pathway Blockade in Non-Small Cell Lung Cancer: A Retrospective Analysis. Clin Cancer Res 2016;22:4585-93. [Crossref] [PubMed]
41. Ayers M, Lunceford J, Nebozhyn M, et al. IFN-γ-related mRNA profile predicts clinical response to PD-1 blockade. J Clin Invest 2017;127:2930-40. [Crossref] [PubMed]
42. Thai AA, Solomon BJ, Sequist LV, et al. Lung cancer. Lancet 2021;398:535-54. [Crossref] [PubMed]
43. Forde PM, Spicer J, Lu S, et al. Neoadjuvant Nivolumab plus Chemotherapy in Resectable Lung Cancer. N Engl J Med 2022;386:1973-85. [Crossref] [PubMed]
44. Zhao S, Zhao H, Yang W, et al. The next generation of immunotherapies for lung cancers. Nat Rev Clin Oncol 2025;22:592-616. [Crossref] [PubMed]
45. Wei Q, Huang C, Zhang Y, et al. IL‑6: A key player in the EGFR‑TKI‑resistant tumor microenvironment and its therapeutic implications (Review). Int J Oncol 2026;68:47. [Crossref] [PubMed]
46. Du W, Feng X, Liu Y, et al. IL-17D promotes ferroptosis resistance in lung cancer via activating PPARγ pathway. Free Radic Biol Med 2026;245:343-56. [Crossref] [PubMed]
47. Masroor Ali Beg M, Aslam M, Ayaz A, et al. Advances in Non-Small Cell Lung Cancer Cellular Immunotherapy: A Progress in Dendritic Cell, T-Cell, and NK Cell Vaccines. Cells 2025;14:1453. [Crossref] [PubMed]
48. Liu Y, Qin D, Fu J. T lymphocyte heterogeneity in NSCLC: implications for biomarker development and therapeutic innovation. Front Immunol 2025;16:1604310. [Crossref] [PubMed]
49. Zhang L, Guo X, Sun X, et al. Analysis of tumor-infiltrating exhausted T cells highlights IL-6 and PD1 blockade as a combined immunotherapy strategy for non-small cell lung cancer. Front Immunol 2025;16:1486329. [Crossref] [PubMed]
50. Akdogan O, Turkmen S, Uyar GC, et al. Impact of Serum GDF-15 and IL-6 on Immunotherapy Response in Cancer: A Prospective Study. Cancers (Basel) 2024;16:4146. [Crossref] [PubMed]
51. Mai VH, Prasanpanich M, Zungsontiporn N, et al. A comprehensive analysis of the correlation between plasma cytokines/chemokines and tumor immune microenvironment signature influences the response of checkpoint inhibitors in advanced non-small-cell lung cancer. Clin Transl Immunology 2026;15:e70076. [Crossref] [PubMed]
52. Kuribayashi T, Nishimura J, Okawa S, et al. TGF-β Inhibitor Potentiates Osimertinib-Induced Anti-Tumor Immunity in Egfr-Mutant Lung Cancer. Cancer Sci 2026;117:1260-72. [Crossref] [PubMed]
53. Zhu Z, Xu C, Kong X, et al. Oncolytic adenovirus encoding a TGF-β inhibitor synergizes with PD-1 blockade to potentiate NK cell cytotoxicity against NSCLC. Front Immunol 2026;17:1759236. [Crossref] [PubMed]
54. Wang L, Chu H, Chen D, et al. SLC2A1(+) tumour-associated macrophages spatially control CD8(+) T cell function and drive resistance to immunotherapy in non-small-cell lung cancer. Nat Cell Biol 2026;28:338-48. [Crossref] [PubMed]
55. Liu S, Ma S, Liu G, et al. CK2B Induces CD8(+) T-Cell Exhaustion through HDAC8-Mediated Epigenetic Reprogramming to Limit the Efficacy of Anti-PD-1 Therapy in Non-Small-Cell Lung Cancer. Adv Sci (Weinh) 2025;12:e2411053. [Crossref] [PubMed]
56. Castro N, Labiano I, Martínez-Aguillo M, et al. Circulating low-density neutrophils as biomarkers of resistance to first-line anti-PD-1/PD-L1 immunotherapy in non-small cell lung cancer. Transl Oncol 2026;67:102755. [Crossref] [PubMed]
57. Hao Y, Wang S, Wang J, et al. Reprogrammed MDSCs promote Th1-dominant antitumour response via CD40 induced by autocrine TNF-α after combining cryo-thermal therapy with IL6 and IL17A neutralization. Clin Transl Med 2025;15:e70493. [Crossref] [PubMed]