Urinary tract infections (UTIs) are common pathologies that require empirical therapy with broad-spectrum antibacterial drugs before obtaining urine culture results. This is one of the factors in the development of antibacterial resistance. Urinary tract infections (UTIs) are most often caused by Enterobacteraceae. It is generally accepted that the most common uropathogen is Escherichia coli, but according to different studies, its percentage in the proportion of UTI pathogens varies from 30 to 95% and depends on factors such as UTI localization, the presence of complicating factors, etc. According to WHO, Enterobacteriaceae resistant to carbapenems and cephalosporins are the most dangerous infections, requiring huge treatment costs [1]. In this regard, an extremely important issue is the search for ways to identify the causative agent of UTIs before prescribing treatment. Artificial intelligence allows us to solve problems in healthcare that previously seemed insoluble, including predicting the resistance of microorganisms to antibiotics [2]. 
Creating a machine learning model for predicting UTI pathogens is the aim of our study.

The results of urine culture tests of 4117 patients treated by urologists in 2024 were analyzed. The following clinical data were used as factors for the analysis: age, gender, UTI localization (upper or lower urinary tract (UTT), presence of UTI drainage, complicated UTI or not, number of previous hospitalizations, causative agent of UTI.
Considering the natural sensitivity and resistance of UTI pathogens [3], the pathogens were divided into four main groups: gram-positive bacteria, non-fermenting gram-negative bacteria, Escherichia coli and Enterobacteriaceae, the overwhelming majority of which was represented by Klebsiella pneumonia, which in recent years has been called a "rapidly growing threat to public health" [4]. Klebsiella pneumonia is rapidly taking one of the leading positions among the causative agents of UTI, including those resistant to broad-spectrum antibiotics, topping the WHO list of resistant pathogens, a critical priority. Klebsiella pneumonia has become one of the main uropathogens of hospital-acquired infections, displacing Pseudomonas aeruginosa. 
In this study, we used deep machine learning software written in Phyton with the following classifiers: PyTorch-based multilayer perceptron and Random Forest model. To evaluate the final performance of the model, a part of the dataset (20%) was not used during training and was used only for testing purposes, and the sample was random. To find the best classifier, we consider some key metrics such as accuracy, area under the curve.

Random Forest model, multilayer percepton showed quite good data, but the best results in accuracy and AUC metrics were achieved by using the multilayer percepton classifier (0.734 and 0.768, respectively). Experimental results show that the multilayer perceptron model is able to differentiate UTI pathogens with an accuracy of 73.4% based on data that is fairly quick and easy to collect: age, gender, UTI localization (upper or lower urinary tract (UTT), presence of UTI drainage, complicated UTI or not, number of previous hospitalizations.

The use of artificial intelligence to predict the pathogen of UTI allows us to predict the results  of urine culture test. Understanding which pathogen caused the UTI episode in each specific case will allow us to most accurately select an antibacterial drug before receiving the results of bacteriological testing of urine. At the same time, in our opinion, it is very important to take into account both the natural resistance of the suspected microorganism and the resistance based on the local model for assessing antibiotic resistance.

Early detection of the pathogen and knowledge of its antibiotic resistance profile is critical for a favorable outcome. The introduction of machine learning methods can significantly help in identifying the uropathogen before receiving the results of urine culture test and choosing empirical antibiotic treatment.

Sati, Hatim & Tacconelli, Evelina & Carrara, Elena & Savoldi, Alessia & Garcia-Vello, Pilar & Zignol, Matteo & Cameron, Alexandra. (2024). WHO Bacterial Priority Pathogens List, 2024.
Martínez-Agüero S, Mora-Jiménez I et al., Machine Learning Techniques to Identify Antimicrobial Resistance in the Intensive Care Unit, Entropy 21 (2019), 603.
European committee on Antimicrobial  Susceptibility Testing (EUCAST). Expert rules, intrinsic resistance and exceptional phenotypes]. 2016;3(1). www.eucast.org.