The convergence of immunotherapy and surgery: a comprehensive bibliographic analysis leveraging large language models

Background: The utilization of immunotherapeutic approaches in the operable settings have demonstrated promising efficacy and safety in preventing postoperative recurrence and improving patients’ survival across various cancer types. To delineate the developmental trajectory and future directions of this specialized field, a comprehensive bibliometric analysis is necessary. Materials and Methods: A total of 541 high-quality publications focusing on the convergence of immunotherapy and surgery were manually extracted from 6,973 papers published between 2015 and 2024 in the Web of Science Core Collection. Traditional bibliometric analysis was conducted utilizing VOSviewer, CiteSpace, and the Bibliometrix R package to identify contributions at national and institutional levels, journal impact, and pivotal keywords. The integration of large language models (LLMs)-driven text embeddings alongside BERTopic-based semantic modeling enabled a comprehensive synthesis of themes and illuminated prospective directions within this field. Furthermore, advanced dimensionality reduction techniques such as UMAP and t-SNE were employed to visualize clustering outcomes and interpret the semantic distances among these academic publications. Results: The distribution of publications over time indicated a growing interest in the integration of immunotherapy and surgery. China and the United States were leaders in publication volume and academic influence, while the “ Journal for Immunotherapy of Cancer ” stand out as a key resource. Clustering based on LLMs-driven embedding vectors identified five categories of papers, the thematic classification of which was presented in a hierarchical structure. Further in-depth analysis revealed the optimal utilization patterns of immune checkpoint inhibitors (ICIs) vary across different cancer types. Neoadjuvant ICIs are effective for triple negative breast cancer, non-small cell lung cancer, melanoma, and deficient mismatch repair colon cancer. Adjuvant mRNA vaccines are viable for pancreatic ductal adenocarcinoma and melanoma, while oncolytic virus demonstrate enhanced efficacy alongside manageable toxicity in perioperative settings. Conclusion: This bibliometric analysis, which integrates LLMs-driven modeling with manual thematic synthesis, has elucidated the developmental trajectory and future directions of the field concerning the combination of immunotherapy and surgery.