Machine learning-based prediction of septic shock in patients with obstructive pyelonephritis caused by ureteral stones

Obstructive acute pyelonephritis (OAPN) is a common condition that can rapidly become life-threatening. In particular, OAPN associated with ureteral stones (OAPN-US) should be treated by a diversion of the urinary tract. Notably, we focused on predicting septic shock, which can be highly fatal. Previous studies have suggested multiple variables for predicting septic shock, but most studies used the Sepsis-2 criteria, which were revised to the Sepsis-3 criteria in 2016. Furthermore, no studies have identified a confirmatory variable or proposed combination of multiple variables for the prediction of septic shock. This study aimed to determine the best combination of potential predictors of septic shock in patients with OAPN-US based on the Sepsis-3 criteria. Our approach was based on machine learning (ML) because previous studies have demonstrated high performance of ML and because it can easily identify the combined effects of multiple variables. We selected three variables from approximately 30 factors and compared five ML algorithms. Notably, the random forest algorithm demonstrated the best performance, with an area under the receiver operating curve of 0.9965 and accuracy of 0.9583. Among the selected factors, procalcitonin had the highest Gini importance score. To the best of our knowledge, this is the first study to use ML for the predication of septic shock. Excellent predictions were made after identifying the optimal combination of multiple predictors.