Development and Validation of an Early Risk Prediction Model for Sepsis-Induced Coagulopathy Based on Machine Learning.

ObjectiveTo develop and validate machine learning models for early prediction of sepsis-induced coagulation dysfunction (SIC) risk at intensive care unit (ICU) admission, before diagnostic criteria are fully met.MethodsThis retrospective cohort study enrolled 197 septic ICU patients. The cohort was randomly split into training (n=159, 80.7%) and internal validation sets (n=38, 19.3%). An independent external cohort of 91patients served for validation. Feature screening used least absolute shrinkage and selection operator (LASSO) followed by multivariate logistic regression (P<0.05). Four models were built and optimized via 10-fold cross validation. Models were evaluated using the area under the curve (AUC), calibration curve, and decision curve analysis (DCA); The optimal model was interpreted by shapley additive explanations (SHAP).Results57.9% of patients developed SIC. Prothrombin time (PT), diastolic blood pressure (DBP), activated partial thromboplastin time (APTT), D-dimer (DD), white blood cell count (WBC), renal insufficiency (RI), and partial pressure of carbon dioxide (pCO2) were identified as influencing factors for SIC. Prolonged PT and APTT, elevated DD, RI, and higher pCO2 increased SIC risk, while higher DBP and WBC were protective. The random forest (RF) model achieved the best predictive efficiency, with internal and external validation AUCs of 0.76 and 0.79.ConclusionThe RF model presents moderate predictive performance for SIC risk, suggesting its potential utility for early identification of high-risk patients upon ICU admission. Further prospective validation is needed before clinical implementation.