Prediction of intensive care unit requirement and in-hospital mortality in Fournier's gangrene: a comparative analysis of conventional statistical and machine learning models.

内容摘要

This study aimed to evaluate the need for intensive care unit (ICU) admission and identify factors associated with in-hospital mortality (IHM) in patients with Fournier's gangrene (FG) using traditional statistical methods complemented by machine learning-based models. This retrospective cohort study included surgically treated FG patients at a tertiary referral center. Demographic, clinical, laboratory, and perioperative variables were analyzed. Established prognostic indices, including the Fournier's Gangrene Severity Index, Uludağ Fournier's Gangrene Severity Index, Laboratory Risk Indicator for Necrotizing Fasciitis, Systemic Immune-Inflammation Index, platelet-to-lymphocyte ratio (PLR), and neutrophil-to-lymphocyte ratio, were evaluated. Patients were stratified according to ICU requirement and survival status. Multivariable logistic regression was performed to identify independent predictors. In addition, exploratory machine learning models, including k-Nearest Neighbors (KNN), Random Forest (RF), Support Vector Machine, and Decision Tree algorithms, were applied to assess predictive performance. Multivariable logistic regression analysis revealed that PLR and heart failure (HF) were independent predictors of ICU requirement. Regarding IHM, PLR remained the only independent predictor. In the exploratory ML analysis, KNN and RF showed AUC values of 0.886 and 0.873 for ICU prediction, and 0.787 and 0.765 for IHM prediction, respectively. However, given the limited sample size and low number of outcome events, these performance estimates should be interpreted cautiously. This study highlights the prognostic relevance of inflammatory markers, particularly PLR, and comorbid conditions including HF, chronic kidney disease, cerebrovascular disease and concurrent malignancy, in disease severity and IHM in FG. Machine learning-based models showed promising performance, although these findings should be considered preliminary and require validation in larger, multicenter cohorts.