Machine Learning-Based Prediction of 28-Day Mortality in Patients with Sepsis-Induced Cardiomyopathy: An Integrated Analysis of Inflammatory Biomarkers
Abstract
Background Sepsis-induced cardiomyopathy (SICM) is a common complication of sepsis and is associated with poor short-term outcomes. However, effective tools for early risk stratification remain limited. This study aimed to develop a machine learning (ML)–based model to predict 28-day mortality in patients with SICM. Methods A single-center retrospective cohort study was conducted, including 762 patients with SICM admitted to the intensive care unit (ICU). Clinical variables, including inflammatory biomarkers, were collected. The dataset was divided into training and validation sets at a ratio of 7:3. Feature selection was performed using least absolute shrinkage and selection operator (LASSO) regression. Multiple ML algorithms, including Random Forest (RF), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Logistic Regression (LR), Light Gradient Boosting Machine (LightGBM), and Multilayer Perceptron (MLP), were used to develop prediction models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and Brier score. Shapley Additive Explanations (SHAP) were applied to interpret the model. Results A total of 762 patients were included, with a 28-day mortality rate of 31.5%. Among all models, RF achieved the best performance (AUC = 0.88) with good calibration (Brier score = 0.149). SHAP analysis identified albumin (ALB), N-terminal pro-B-type natriuretic peptide (NT-proBNP), age, pH (PH), systolic blood pressure (SBP), and neutrophil-to-lymphocyte ratio (NLR) as key predictors of mortality. A web-based interactive nomogram was constructed to generate individualized risk estimates tailored to each patient’s clinical profile. Conclusion The ML-based model demonstrated good predictive performance and interpretability for 28-day mortality in SICM patients. This model may serve as a useful tool for risk stratification and clinical decision-making. Further external validation in multicenter prospective cohorts is required before broader clinical implementation.