Answers are grounded in this paper and cite their sources.
Early prediction of prognosis is critical for optimizing care in intensive care unit (ICU) sepsis patients with atrial fibrillation (AF). This study aims to develop explainable machine learning (ML) models to predict 28-day in-hospital mortality risk for sepsis patients with AF. Data from sepsis patients with AF were extracted from three datasets: the Medical Information Mart for Intensive Care (MIMIC)-IV database, the MIMIC-III database, and the eICU Collaborative Research Database (eICU). The study endpoint was 28-day in-hospital mortality following ICU admission. Feature selection was performed using least absolute shrinkage and selection operator (LASSO), Boruta and recursive feature elimination (RFE). We developed eight distinct ML models and evaluated their performance comprehensively using various metrics. SHapley Additive exPlanations (SHAP) was employed to interpret the optimal predictive model. A total of 11,740 sepsis patients with AF were included in this study, with 8,374 from the MIMIC-IV database, 1,669 from the MIMIC-III database, and 1,697 from the eICU database. 18 variables were selected to construct prediction models. The Gradient Boosting Machine (GBM) model demonstrated robust predictive performance, with an area under the curve (AUC) and 95% confidence interval (CI) of 0.770 (0.749–0.790) in the internal validation set, 0.781 (0.756–0.807) in the MIMIC-III cohort, and 0.721 (0.691–0.750) in the eICU cohort. SHAP analysis revealed that respiratory failure, urine output, and RDW score were the three most important features for predicting 28-day in-hospital mortality risk in sepsis patients with AF. The simplified model containing only 12 features demonstrated comparable performance, achieving an AUC of 0.756 (0.735–0.777) in the internal validation set, 0.776 (0.750–0.802) in the MIMIC-III cohort, and 0.715 (0.685–0.745) in the eICU cohort. The final streamlined model was deployed on a dedicated website ( https://ml-model.shinyapps.io/gbmapp/ ) to assist physicians in clinical decision-making and timely intervention.
No takes yet. Share an insight, caveat, or question.
Sun et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: