Urine Flow Cytometry Parameter Cannot Safely Predict Contamination of Urine—A Cohort Study of a Swiss Emergency Department Using Machine Learning Techniques

Background: Urine flow cytometry (UFC) analyses urine samples and determines parameter counts. We aimed to predict different types of urine culture growth, including mixed growth indicating urine culture contamination. Methods: A retrospective cohort study (07/2017–09/2020) was performed on pairs of urine samples and urine cultures obtained from adult emergency department patients. The dataset was split into a training (75%) and validation set (25%). Statistical analysis was performed using a machine learning approach with extreme gradient boosting to predict urine culture growth types (i.e., negative, positive, and mixed) using UFC parameters obtained by UF-4000, sex, and age. Results: In total, 3835 urine samples were included. Detection of squamous epithelial cells, bacteria, and leukocytes by UFC were associated with the different types of culture growth. We achieved a prediction accuracy of 80% in the three-class approach. Of the n = 126 mixed cultures in the validation set, 11.1% were correctly predicted; positive and negative cultures were correctly predicted in 74.0% and 96.3%. Conclusions: Significant bacterial growth can be safely ruled out using UFC parameters. However, positive urine culture growth (rule in) or even mixed culture growth (suggesting contamination) cannot be adequately predicted using UFC parameters alone. Squamous epithelial cells are associated with mixed culture growth.