Background: Feline idiopathic cystitis (FIC) is the leading cause of cats with lower urinary tract signs (LUTS). Although bacterial urinary tract infections (UTI) are less common, antimicrobial therapy is often prescribed. Methods to identify cats with a high likelihood of UTI may improve antimicrobial stewardship. Hypothesis: Machine learning models (MLM) trained with clinical data and urinalysis results can predict which cats with LUTS have a high risk of UTI. Animals: 425 cats.
Methods: Records from cats with LUTS and FIC or UTI from 2009-2020 were reviewed; urinalysis and contemporaneous urinary culture were recorded. Cats with urolithiasis, masses, or other causes for LUTS were excluded. These data were split into 80% training and 20% test sets. Body weight, age, sex, and urinalysis results were used to train a gradient-boosted tree MLM, and feature importance was determined. The performance statistics from the test set are presented.
Results: 230 cats with FIC and 195 with UTI were identified. The MLM had a sensitivity of 84.8% (95% confidence interval [CI];74.4-94.0%) and specificity of 81.8% (95% CI;67.6-93.9%) for predicting UTI. The area under the receiver operating characteristic curve was 0.885 (95% CI;0.802-0.955). The positive predictive value was 82.3% (95% CI; 66.5-93.9%), and the positive likelihood ratio was 4.6. The top contributing features to the MLM were urine white blood cell concentration, specific gravity, and age. Conclusion and Clinical Importance: MLMs can help identify cats at high risk of UTI when exhibiting LUTS and may aid in clinical decision-making.
