Natural language processing to enhance rheumatoid arthritis care in clinical studies: a scoping review of...

Structured claims or EMR datasets have limitations, such as lacking important clinical variables, upcoding, or potential coding errors. Unstructured data powered with natural language processing (NLP) might bridge these gaps. We assessed the integration of NLP with unstructured data in advancing care for rheumatoid arthritis (RA). We conducted a scoping literature review search in PubMed, Embase, Web of Science and Directory of Open Access Journal to identify full-text studies published through February 27 2026. We used the search terms and relevant MeSH terms involving "rheumatoid arthritis" and "natural language processing" to identify English publishing studies that used the NLP methods while also considering any real world clincial application. Abstracts, reviews, reports, or commentaries were excluded. We extracted clinical problems, data, use of NLP and main findings across each study. Through the search terms, there identified a total of unique 345 citations after removing 54 duplicates. After assessing eligibility criteria, 27 studies met the inclusion criteria. We thoroughly reviewed all 27 qualified publications, and found that they were conducted in US, Japan, Germany, UK, and others. The tasks involved in these eligible studies were divided into several categories to show how NLP was explored to advance RA care: (1) extract a variety of clinical features to support predictive tasks; (2) identify cohorts of RA patients with specific phenotypes; (3) aid in ICD diagnosis codes to improve the precise diagnosis of RA or RA related comorbidities through extraction of clinical notes; (4) social media analysis; (5) extract information about RA-related medication use, indications, reasons for adjustment, and detecting medication induced safety signals. The unstructured datasets leveraged using NLP were divided into categories: medical health records, physician chart notes, purely social media texts, medical text notes and others. NLP, which is able to extract information from unstructured text data, has been increasingly used in clinical studies for RA. Recent work have shown that wide use of NLP to capture concepts from the unstructured clinical notes from EHR data, to improve the identification of RA or RA related comorbidities or RA disease phenotyping. Integrating NLP into structured data to detect observed confounders, in powering pharmacoepidemiologic studies of comparative effectiveness research in RA could be an important future area.