Development of machine learning models to predict clinical outcome and recovery time in dogs with parvovirus enteritis.

Canine parvovirus (CPV) is one of the most contagious viral diseases in dogs that usually presents with diarrhea, vomiting, and fever. Various clinical and laboratory biomarkers such as SIRS, leukopenia, neutropenia and CRP have been introduced to predict the final outcome of dogs with CPV. With the advent of machine learning methods/algorithms, various models can be developed using a combination of clinical and non-clinical variables to predict clinical outcome in different diseases with higher efficiency compared to traditional biomarkers. In this study, we sought to develop models to predict clinical outcome and recovery time in dogs with CPV infection using 10 and 4 machine learning algorithms, respectively. A model was developed using four variables (SIRS, deworming, vaccination and crying) to predict clinical outcome. The performance of this model was measured using three metrics: accuracy scores, AUC (area under the Receiver Operating Characteristic (ROC) curve) and AUC score. Another model was constructed using five variables (retching, foul smelling, housing, dehydration, and shift-to-left) to estimate recovery time. The performance of this model was evaluated using two criteria: mean square error (MSE) and root mean square error (RMSE). In the model developed for clinical outcome, the average of accuracy scores, AUC scores and AUCs in the test dataset were 0.84, 0.90 and 0.73, respectively. The second model predicted the recovery time in the test group with a mean error of 2 days (RMSE = 2.05). Our findings demonstrate that ML models can effectively integrate clinical and laboratory features to predict survival and recovery time in CPV-infected dogs, offering a valuable tool for early prognosis and treatment optimization.