XGBoost (eXtreme Gradient Boosting) Can Predict Organisms Growing in Urine Culture from the Emergency Department

Source: PubMed Central Open Access, NCBI / U.S. National Library of Medicine

Introduction Urinary tract infections are common in the emergency department (ED) but are frequently misdiagnosed and mismanaged. We sought to determine whether eXtreme Gradient Boosting (XGBoost), an open-source machine-learning library, could predict the organisms growing in urine cultures ordered from the ED. Methods We developed XGBoost algorithms to retrospectively examine 62,963 Mayo Clinic ED encounters between January 1, 2017–December 31, 2021, during which a urinalysis and urine culture were performed. The model used 1,303 patient variables. All patient ages were included. Data were from the electronic health record and available to the clinician during the patient encounter. Results For the most common bacteria growing in urine culture, XGBoost was able to predict the presence of a member of thefamily with an area under the receiver operating curve (AUC) of 0.90 and an accuracy of 0.79. The model predicted the presence of 10 different bacterial genera with an AUC of 0.70–0.88 and an accuracy of 0.87–0.99. Furthermore, XGBoost was able to predict whether the urine culture would report Gram-positive or Gram-negative bacteria with an AUC of 0.81 and 0.90, respectively, and an accuracy of 0.85 and 0.86, respectively. The model predicted whether yeast would be reported with an AUC of 0.84 and an accuracy of 1.00. Discussion XGBoost can predict the bacterial genus and Gram-staining results of the bacteria growing in urine cultures.