Code to accompany the publication "A machine learning approach to diagnosing feline infectious peritonitis, a fatal coronavirus infection of cats" in R markdown format.
Title: Assessing the feasibility of applying machine learning to diagnosing non-effusive feline infectious peritonitis
Authors and Affiliations: Dawn Dunbar a, Simon A. Babayan b, Sarah Krumrie a, Hayley Haining a, Margaret Hosie c and William Weir a a School of Veterinary Medicine, University of Glasgow, Glasgow, United Kingdom b Institute of Biodiversity, Animal Health and Comparative Medicine, University of Glasgow, Glasgow, United Kingdom c MRC-University of Glasgow Centre for Virus Research, Glasgow, United Kingdom
Abstract Until recently Coronaviruses have existed in relative obscurity in terms of their impact on human health. In contrast, in the field of veterinary medicine, feline coronavirus (FCoV) is a well-recognised pathogen and a major contributor to mortality in felids, both wild and domestic. Infection with FCoV in cats bears similarity to SARS-CoV2 infection in humans in that a proportion of patients are asymptomatic, a proportion are ill experiencing mild symptoms while a minority of cases develop into severe, life-threatening disease. Feline infectious peritonitis (FIP) is a severe FCoV-associated syndrome in cats which is invariably fatal without anti-viral treatment. Due to the varied nature of the clinical presentation and similarity to other feline diseases, accurate and efficient diagnosis of FIP remains a challenge even to the experienced clinician. A battery of tests is required to support diagnosis of FIP, however, interpretation guidelines are complex, involving the subjective evaluation of a wide range of clinical and laboratory markers. We hypothesised that machine learning could be used to construct highly accurate predictive models to aid diagnosis using a dataset representing thousands of suspected cases over a period of two decades. Ensemble models were created which had prediction accuracies of 97.5%, confirming the hypothesis that machine learning is a valuable addition to the FIP diagnostic toolkit. More broadly, these results illustrate the benefit of applying machine learning to extensive clinical datasets to standardise, accelerate and increase accuracy of result interpretation, thereby improving the diagnostic capacity of existing laboratory tests.