Investigation the role of the artificial intelligence technologies in predicting urine culture results using biochemical parameters

Urinary tract infections are among the most frequently encountered infections, with urine culture serving as the gold standard for laboratory diagnosis. However, obtaining culture results requires time, which may delay appropriate treatment decisions. This study aimed to evaluate the effectiveness of different artificial intelligence (AI) models in predicting urine culture results more rapidly by utilizing biochemical parameters. Our single-center, cross-sectional and methodological study was conducted by retrospectively scanning the health data of patients who had hemogram, complete urinalysis, urea, creatinine, and CRP tests performed on the same day as urine culture between January 2014 and December 2023.A total of 25,796 clinical samples were tested using five different machine learning algorithms, employing two variable sets: a full variable set consisting of 46 features and a reduced variable set consisting of 12 non-invasive features. Additionally, subgroup analyses were conducted based on age and sex.The best-performing algorithm for the full variable set was Random Forest (AUC: 0.909; accuracy: 0.867; sensitivity: 0.705; specificity: 0.930). The best-performing algorithm for the reduced variable set was also Random Forest (AUC: 0.897; accuracy: 0.856; sensitivity: 0.689; specificity: 0.912). XG Boost demonstrated the best performance in women over 65 years and males under 18 years.The use of machine learning algorithms alongside objective laboratory parameters has the potential to predict urine culture results with high specificity. The Random Forest algorithm outperformed other models in both full and reduced variable sets, while XG Boost also emerged as a viable option depending on age and sex-related factors. Keywords: Artificial Intelligence, Machine Learning, Urinary Tract Infection, Urine Culture