A Machine-Learning model for robust prediction of Sepsis-Induced Coagulopathy in Critically Ill Patients with Sepsis

Jia Sun¹Lixin Zhang²Zhaotang Gong²Hongling Ma²Dan Wu²Rihan Wu²Guleng Siri^2*

• ¹Department of Pharmacy, Baotou Medical College, Baotou, China
• ²Department of Pharmacy, Inner Mongolia People's Hospital, Hohhot, China

Abstract: Introduction: Sepsis-induced coagulopathy (SIC) is a common disease in patients with sepsis. It denotes higher mortality rate and poorer prognosis in septic patients. This study aimed to develop a practical machine-learning models to predict the risk of SIC in critically ill patients with sepsis. Methods: In this retrospective cohort study, we extracted patients from the Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ) database and the Inner Mongolia Autonomous Region People's Hospital database. Sepsis and SIC was defined based on Sepsis 3.0 criteria and criteria was developed based on International Society of Thrombosis and Haemostasis (ISTH) respectively. We compared 9 machine learning (ML) models with Sequential Organ Failure Assessment (SOFA) score in terms of SIC prediction ability. Optimal model selection was based on the superior performance metrics exhibited by the model on the training dataset, internal validation dataset, and external validation dataset. Results: Of 15, 479 patients in MIMIC-IV included in the final cohort, a total of 6036 (38.9%) patients developed SIC during sepsis. Seventeen features were selected to construct ML prediction models. The Gradient Boosting Machine (GBM) model was deemed optimal as it achieves high predictive accuracy and reliability across the training, internal, and external validation sets. The area under the curve of GBM model were 0.773 (95% CI 0.765–0.782) in the training set, 0.730 (95% CI 0.715–0.745) in the internal validation set and 0.966 (95% CI 0.938–0.994) in the external validation set. Shapley Additive explanations (SHAP) values illustrated the prediction results, indicating that total bilirubin, red cell distribution width (rdw), systolic blood pressure (sbp), heparin, and blood urea nitrogen (bun) were risk factors for progression to sepsis-induced coagulopathy in septic patients. Conclusions: We developed an optimal and operably ML models which was able to predict the risk of SIC in septic patients better than scoring models.