Fine-tuning foundational models to code diagnoses from veterinary health records
13 November 2026
A study in PLOS Digital Health by Mayla R. Boguslav and colleagues shows how language models can assign diagnosis codes to veterinary health records automatically. The team fine-tuned 13 freely available models on 246,473 manually coded patient visits from the Colorado State University Veterinary Teaching Hospital, covering 7,739 SNOMED-CT diagnosis codes.
The results beat earlier efforts, and the authors found that comparable results are possible with more limited resources and with non-clinical models.