Association between clinical characteristics within 6 h of ICU admission and 30-day mortality risk in immunocompromised sepsis patients: development and validation of a machine learning model based on the MIMIC-IV database
Zhipeng Cheng, Xiuqing Ma, Weiying Duan, Zeyu Mou, Zhixin Liang
Objective To develop and validate a machine learning model for predicting 30-day mortality in immunocompromised sepsis patients using clinical data within 6 h of ICU admission. Methods This retrospective cohort study utilized data from the MIMIC-IV and eICU databases. Adult immunosuppressed patients meeting Sepsis-3 criteria were enrolled, and clinical indicators within 6 h of ICU admission were extracted. Variable selection was performed using multiple testing correction and recursive elimination algorithms, ultimately incorporating 10 feature variables. Predictive models were constructed using seven machine learning algorithms: Logistic Regression, Decision Tree, Random Forest, XGBoost, LightGBM, Support Vector Machine (SVM), and Artificial Neural Network (ANN). Model discriminative capabilities were assessed using Area Under the Receiver Operating Characteristic Curve (AUC), calibration curves, and Decision Curve Analysis (DCA). The best-performing model underwent external validation using eICU database data and SHAP interpretability analysis. Results A total of 2,494 immunosuppressed sepsis patients were included, with a 30-day mortality rate of 33.4%. The final prediction model included 10 feature variables: Weight, APS-III score, Urine output, Prothrombin Time (PT), Blood Urea Nitrogen (BUN), SOFA score, Red Blood Cell count (RBC), Platelet count (PLT), Age, and Mean Corpuscular Hemoglobin Concentration (MCHC). Among all tested machine learning algorithms, the Support Vector Machine (SVM) model demonstrated the best predictive performance, with an AUC of 0.794 (95% CI: 0.761–0.826) in the validation set. Calibration curves and DCA showed good consistency between predicted probabilities and actual risk. External validation on the eICU dataset ( n = 1,280) yielded an AUC of 0.847, indicating stable predictive performance and strong generalizability. Conclusion This study successfully developed an SVM-based prediction model that effectively predicts 30-day mortality risk in immunosuppressed sepsis patients using only 10 easily obtainable clinical indicators available within 6 h of admission. This model shows promise as a practical tool for early identification of high-risk immunosuppressed sepsis patients and assisting in personalized treatment decisions in clinical settings.