Applying artificial intelligence to automate clinical data extraction from veterinary electronic health records: A practical guidance and comparative scoping review.

Abstract

Veterinary electronic health records are often stored as unstructured free text, and structuring this information into analyzable formats is essential for downstream research. Natural language processing methods, including rule-based systems, machine learning algorithms, and, more recently, large language models (LLMs), provide tools to achieve this goal. This scoping review follows the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) Extension for Scoping Reviews (PRISMA-ScR) guideline to examine how information extraction (IE) has been applied in veterinary medicine and extended to existing LLM approaches in human medicine. Literature research was conducted on 4 databases: PubMed, CAB Abstracts, Web of Science, and ACL Anthology, with stricter criteria limiting human medicine to prompt-based studies. After screening 5796 original research papers, a total of 23 veterinary and 31 human studies were selected for inclusion. In the veterinary literature, larger data sets were more commonly used to train supervised models, whereas human studies increasingly employed prompt-based LLMs, such as LLaMA and GPT, enabling IE with smaller annotated data sets. We developed a practical framework covering data preparation, platform and privacy considerations, and prompt engineering, with a corresponding workflow and prompt example to demonstrate the application of artificial intelligence (AI) in clinical data extraction. This review offers a practical tool to help veterinary researchers effectively integrate AI tools and LLMs into clinical research workflows.