Researchers at the University of Liverpool have developed an innovative artificial intelligence (AI) approach that can rapidly analyse large scale veterinary clinical records, offering a powerful new way to monitor the health of the UK’s dog population.
Published in the Journal of Big Data, the study shows how machine learning can uncover emerging diseases, breed-related risks and seasonal health patterns in pets at a national scale.
Veterinary research has traditionally relied on relatively small datasets, limiting the ability to detect wider trends in animal health. In contrast, this study analysed one million electronic clinical notes collected from one million individual dogs from veterinary practices across the UK. However, the sheer volume and complexity of these records make them difficult for humans to interpret alone.
To address this challenge, the team evaluated an unsupervised machine learning method known as BERTopic, which can automatically group and cluster clinical notes without any prior human annotation. The results show that the model can identify a wide range of important diseases, from diabetes and dermatitis to heart disease and Cushing’s disease, as well as behavioural issues and injuries such as grass seed complications.
In a single study, this AI approach highlighted how disease risks vary across dog breeds, sexes and ages. It detected seasonal patterns including tick infestations and firework-related anxiety and even identified the signal of a nationwide outbreak of acute vomiting and diarrhoea in 2020 in dogs.
PJ Noble, Professor of Small Animal Internal Medicine in the Institute of Infection, Veterinary and Ecological Sciences, added: “With the advent of these huge clinical datasets, our team is leveraging advanced AI tools to ensure data from every record informs the observations we can make.”
The ability to process new data continuously means the AI model could support real time monitoring of pet health across the UK. With approximately 10,000 new veterinary records added daily, it offers a way to detect emerging diseases sooner, monitor shifts in disease patterns and deliver timely insights for veterinary public health.
The team emphasises that the model is not designed to diagnose individual animals. Instead, it offers a powerful research tool for understanding population level trends. As more data are collected, and as the model is further refined using veterinary specific language, its accuracy and usefulness are expected to grow.
The researchers expressed their gratitude to the veterinary practices and dog owners who contribute anonymised data to the Small Animal Veterinary Surveillance Network (SAVSNET), as well as to Dogs Trust for supporting the work. A key partner here is the CVS group of veterinary practices who commented: “CVS is delighted to partner with SAVSNET by contributing anonymised health data to the project. This paper illustrates how we can leverage the power of AI to make crucial observations about animal health across populations of dogs at previously unimaginable scales.”
Click here to read the full study published in the Journal of Big Data.