Applications of natural language processing and large language models in sports injury assessment and rehabilitation decision-making: a scoping review

Objective: This study aims to provide a scoping review of the current applications of natural language processing (NLP) and large language models (LLMs) in the assessment of sports injuries and rehabilitation decision-making, with the goal of identifying the technical methods, data sources, target populations, key findings, and knowledge gaps in existing research, and to provide evidence-based guidance for clinical practice. Methods: We followed the Joanna Briggs Institute (JBI) scoping review framework and the PRISMA-ScR guidelines to search the PubMed, Scopus, and Web of Science Core Collection databases (up to March 31, 2026). We included studies on sports injury and rehabilitation that utilized NLP or LLMs to process unstructured text. Two researchers independently screened the studies and extracted data; disagreements were resolved through discussion or third-party arbitration. Study quality was assessed using MINimum Information for Medical AI Reporting (MINIMAR). Results: A total of 27 studies were included. The majority of the studies originated from the United States (37.0%). The primary research types were algorithm evaluation and benchmarking. Healthcare professionals were the main target population. The text data sources consisted primarily of simulated/synthetic question-answering scenarios. The included studies faced challenges such as readability issues and AI hallucinations. The overall compliance rate with the MINIMAR report quality assessment was 61.70%. Conclusion: LLMs have shown great potential as tools for sports injury assessment and rehabilitation decision-making. However, they still have significant shortcomings in terms of readability, hallucination control, patient perspective, methodological reporting, and geographic coverage. Future efforts should be grounded in the sports field, with a focus on athletes and team physicians, to improve model reliability, bridge geographic gaps, elevate research standards, and facilitate the transition from baseline assessment to clinical application.