The following is a summary of “A novel post-percutaneous nephrolithotomy sepsis prediction model using machine learning,” published in the February 2024 issue of Urology by Shen et al.
This retrospective study aimed to develop a machine learning-based predictive model for sepsis following percutaneous nephrolithotomy (PCNL) to identify high-risk patients and facilitate early intervention by urologists.
The study included 694 patients who underwent PCNL, and a predictive model for sepsis was constructed using machine learning based on 22 preoperative and intraoperative parameters.
Sepsis occurred in 45 patients (16 males, 29 females) out of the total cohort. The dataset was randomly divided into 80% training and 20% validation sets using 100-fold Monte Carlo cross-validation. The selected variables were found to be highly independent. The model demonstrated robust predictive performance for postoperative sepsis (AUC = 0.89, sensitivity 87.8%, specificity 86.9%, and accuracy 87.4%). The top 10 contributing variables to the model included preoperative midstream urine bacterial culture, sex, days of preoperative antibiotic use, urinary nitrite, preoperative blood white blood cell (WBC) count, renal pyogenesis, staghorn stones, history of ipsilateral urologic surgery, cumulative stone diameters, and renal anatomic malformation.
Their developed predictive model exhibits suitability for estimating sepsis risk after PCNL, providing a valuable tool for reducing sepsis incidence through timely intervention.
Source: bmcurol.biomedcentral.com/articles/10.1186/s12894-024-01414-x