In a large multi-center retrospective study using the MIMIC-IV and eICU-CRD databases, researchers developed an interpretable machine learning model to predict sepsis-induced coagulopathy (SIC) within 7 days of ICU admission.

After rigorous feature selection with LASSO, random forest, and Boruta methods, a LightGBM model using just 13 readily available clinical variables achieved excellent performance: AUROC 0.885 internally and 0.831 on external validation – outperforming traditional scores.

Key predictors included INR, platelet count, lactate, SOFA score, blood pressure, and comorbidities such as heart failure.

Using SHAP values for interpretability and deployed as an interactive free web app, the tool empowers clinicians to identify high-risk patients early, potentially enabling timely interventions and better outcomes in this life-threatening complication of sepsis.

This model offers strong predictive ability with clinical interpretability for early SIC detection and targeted intervention.

Free Tool: Shiny APP for Sepsis-Induced Coagulopathy

Read More