Construction of diagnostic prediction model for canine nasal diseases using less invasive examinations without anesthesia.

Abstract

Advanced imaging techniques under general anesthesia are frequently employed to achieve a definitive diagnosis of canine nasal diseases. However, these examinations may not be performed immediately in all cases. This study aimed to construct prediction models for canine nasal diseases using less-invasive examinations such as clinical signs and radiography. Dogs diagnosed with nasal disease between 2010 and 2020 were retrospectively investigated to construct a prediction model (Group M; GM), and dogs diagnosed between 2020 and 2021 were prospectively investigated to validate the efficacy (Group V; GV). Prediction models were created using two methods: manual (Model 1) and LASSO logistic regression analysis (Model 2). In total, 103 and 86 dogs were included in GM and GV, respectively. In Model 1, the sensitivity and specificity of neoplasia (NP) and sino-nasal aspergillosis (SNA) were 0.88 and 0.81 in GM and 0.92 and 0.78 in GV, respectively. Those of non-infectious rhinitis (NIR) and rhinitis secondary to dental disease (DD) were 0.78 and 0.88 in GM and 0.64 and 0.80 in GV, respectively. In Model 2, the sensitivity and specificity of NP and SNA were 0.93 and 1 in GM and 0.93 and 0.75 in GV, respectively. Those of NIR and DD were 0.96 and 0.89 in GM and 0.80 and 0.79 in GV, respectively. This study suggest that it is possible to create a prediction model using less-invasive examinations. Utilizing these predictive models may lead to appropriate general anesthesia examinations and treatment referrals.