Introduction:
This study aimed to identify research trends and hidden thematic structures in case reports related to rheumatoid arthritis (RA) and complications published between 1995 and September 2025. By applying unsupervised machine learning (ML), the study sought to uncover longitudinal patterns and cross-domain relationships that are not fully captured in traditional reviews or large-scale epidemiological studies.

Material and methods:
Bibliographic data were collected from the Web of Science (WoS) Core Collection using the search term “rheumatoid arthritis and complications,” limited to case reports published between 1995 and 2025. Titles, keywords, and abstracts were combined into unified text documents and processed with natural language processing (NLP) techniques. Term Frequency–Inverse Document Frequency (TF-IDF) was applied for text vectorization. Topic extraction employed Non-negative Matrix Factorization (NMF), and clustering was performed using KMeans. The optimal number of clusters was determined based on the highest Silhouette score (0.636). Analyses were conducted in Python (Version 3.10.5) within the PyCharm environment (Version 2022.1.3).

Results:
The unsupervised machine learning (ML) framework identified two major and stable clusters among 1,200 case reports: a pharmacological and immunological management cluster, and a surgical and postoperative complications cluster. Cluster 1 included 890 reports characterized by pharmacological and immunological terms such as “TNF,” “therapy,” and “anti,” while Cluster 2 contained 310 reports dominated by orthopedic terms such as “arthroplasty,” “knee,” and “revision.” Reproducibility tests across five runs showed perfect topic consistency, confirming methodological reliability.

Conclusions:
This NLP and ML–based analysis revealed a dual structure in RA complication literature—one related to immunologic drug complications and the other to surgical outcomes. Beyond confirming known domains, the study provides a data-driven characterization of their evolution and overlap, offering insights into the shifting landscape of RA management. This approach demonstrates the potential of unsupervised ML for longitudinal mapping of clinical research domains.