Developing a Feline Infectious Disease Triage Model: Insights from Logistic Regression Models in Data from a Veterinary Isolation Unit

Abstract

Abstract Background: The Biological Isolation and Containment Unit (BICU) of the Faculty of Veterinary Medicine, University of Lisbon, is dedicated to treating animals with suspected or confirmed infectious diseases. Feline Immunodeficiency Virus (FIV) and Feline Leukemia Virus (FeLV) are two of the most common infections reported in this unit. This study explored the use of logistic regression to predict FIV and FeLV infections in the triage stage. Results: Of 1211 cats treated at the BICU since its opening, 134 cats were FIV-positive and 126 FeLV-positive. Significant triage-related factors for FIV-related hospitalization included being an adult or senior cat, intact males, having access to the outdoors, and presenting concomitant disorders. In contrast, mixed-breed cats with concomitant disorders and a low hematocrit count were significant risk factors for FeLV-related hospitalization. The estimated logistic regression models without cross-validation showed areas under the Receiver Operating Characteristic curve (AUC) of 0.71 for FIV and 0.67 for FeLV, with 95% CI of [0.66-0.76] and [0.62-0.73], respectively. Cross-validation highlighted high sensitivity but low specificity for both infections, indicating a higher propensity for false positives. When cross-validation was performed for FIV infections, the resulting AUC was 0.66, and the specificity was 0.33 using 10- and 5-fold cross validations. The models for FeLV exhibited similar predictive performance with an AUC of 0.63 and specificity of 0.29, which decreased further with 10- and 5-fold cross validation. Conclusions: This study highlights significant triage-related factors for FIV and FeLV infections, in agreement with existing literature. These findings indicate a need for better clinical vigilance and owner education, mainly on neutering and the risk of outdoor access. Future research should expand to other predictive models and include other variables important to predict FIV and FeLV at the triage stage.