Prediction of Urinary Tract Infection for Hospital-admitted Patients based on Demographic and Historical Data, as well as Machine Learning Approaches

Background and Objective: Urinary tract infection (UTI) is one of the common infections that affects the urinary system. UTI can be detected using analysis of urine culture which is a time-consuming and error-prone procedure. The fast prediction of UTI helps to start antibiotic medication at the correct time and prior to the culture report. The present study aimed to assess the potential of artificial intelligence for UTI prediction. Materials and Methods: The current study was conducted based on the published data from a retrospective cohort study that was performed on 300, 000 human samples in Denmark. The performance of machine learning algorithms, including support vector machines, decision trees, linear discriminant analysis, and linear regression models, was investigated for UTI prediction according to historical and demographical features. Results: The obtained results pointed out that UTI prediction using a linear discriminant analysis model with an accuracy of 65.16±0.64% was possible. Among the features, age had a significant effect on UTI prediction where the model accuracy rates for elderly, adult, young, and infant cases were reported as 73.64, 86.25, 39.42, and, 60.60%, respectively. In addition, the results demonstrated that the classification performance was better for adults compared to younger participants. The most informative features were also suggested according to the neighborhood component analysis. Conclusion: The results of this study highlighted the potential of artificial intelligence for fast UTI prediction; nonetheless, the performance needs to be enhanced by the addition of other fast-accessible features.