Automated Disease Classification of Veterinary Clinical Narratives for Antimicrobial Stewardship Guideline Monitoring

Abstract

Antimicrobial resistance (AMR) is a global challenge for public health with increasing impacts for animal welfare. Companion animals are increasingly being recognised as critical agents in the proliferation and transmission of AMR bacteria. Existing approaches to monitoring antimicrobial utilisation, such as annual aggregate sales data, provide very limited insights into the specific disease contexts and prescribing practices driving antimicrobial use. In this study, we employ a hierarchical large language model to analyse 195,012 highest priority critically important antimicrobials (HPCIA) prescriptions in companion animals and over 1.2 million antimicrobial non-HPCIA prescriptions, revealing the nuances of disease patterns and prescribing motivations that have previously remained obscure. Our model achieves an accuracy exceeding 88\% in identifying 39 distinct diseases associated with antimicrobial prescriptions. We demonstrate that the tool provides a robust foundation for assessing the influence of antimicrobial stewardship guidelines (ASGs) on prescribing practices. Our novel methodology advances our understanding of AMR dynamics within veterinary care by bridging the gap between what is actual practiced and guideline-aligned prescribing. We outline an actionable path towards preserving the efficacy of essential antimicrobial agents, safeguarding the health of future generations of humans and animals.