Related Experiment Video
Updated: May 17, 2025
Tail Vein Transection Bleeding Model in Fully Anesthetized Hemophilia A Mice
Published on: September 30, 2021
Interpretable machine learning model for early prediction of disseminated intravascular coagulation in critically ill
Jintuo Zhou1, Yongjin Xie2, Ying Liu1
1Department of Pharmacy, Fujian Maternity and Child Health Hospital College of Clinical Medicine for Obstetrics and Gynecology and Pediatrics, Fujian Medical University, #18 Daoshan Road, Fuzhou, China.
Insights
Early prediction of disseminated intravascular coagulation (DIC) in critically ill children is crucial. Machine learning, particularly the XGB model, accurately identifies children at risk, enabling timely intervention and improved outcomes in the pediatric intensive care unit (PICU).
Area of Science:
- Pediatric critical care medicine
- Hematology
- Machine learning in healthcare
Background:
- Disseminated intravascular coagulation (DIC) is a life-threatening thrombo-hemorrhagic disorder in critically ill children.
- Accurate and efficient early prediction of DIC is essential for timely intervention and improved patient outcomes.
- Existing prediction methods require enhancement for critically ill pediatric populations.
Purpose of the Study:
- To develop and evaluate machine learning models for the early prediction of DIC in critically ill children.
- To identify key predictors for DIC development in this vulnerable patient group.
- To determine the optimal machine learning algorithm for clinical utility in pediatric intensive care units (PICUs).
Main Methods:
- Utilized a stepwise logistic regression model to select candidate predictors from demographics, comorbidities, laboratory findings, and therapies.
- Trained and evaluated six machine learning algorithms: logistic regression (LR), extreme gradient boosting (XGB), random forest (RF), support vector machine (SVM), decision tree (DT), and k-nearest neighbor (KNN).
- Assessed model performance using metrics including AUC, accuracy, specificity, sensitivity, PPV, NPV, precision, recall, and decision curve analysis (DCA). SHAP analysis was used for model interpretation.
Main Results:
- The study included 6093 critically ill children, with 11.2% developing DIC.
- The XGB model demonstrated superior performance with an AUC of 0.908, high specificity (0.859), PPV (0.978), and precision (0.969).
- SHAP analysis identified D-dimer, INR, PT, TT, and PLT count as the most significant predictors of DIC.
Conclusions:
- Machine learning models, particularly the XGB-based 'Alfalfa-PICU-DIC' model, offer a reliable tool for early DIC prediction in critically ill children.
- The XGB model exhibits superior clinical utility and accuracy compared to other evaluated algorithms.
- Timely intervention facilitated by accurate DIC prediction can significantly reduce the burden of the disorder in PICU patients.
Abstract:
Disseminated intravascular coagulation (DIC) is a thrombo-hemorrhagic disorder that can be life-threatening in critically ill children, and the quest for an accurate and efficient method for early DIC prediction is of paramount importance. Candidate predictors encompassed demographics, comorbidities, laboratory findings, and therapy strategies. A stepwise logistic regression model was employed to select the features included in the final model. Six machine learning algorithms-logistic regression (LR), extreme gradient boosting (XGB), random forest (RF), support vector machine (SVM), decision tree (DT), and k-nearest neighbor (KNN)-were employed to construct predictive models for DIC in critically ill children. Models were then evaluated by using area under the curve (AUC), accuracy, specificity, sensitivity, positive predictive value (PPV), negative predictive value (NPV), precision, recall and decision curve analysis (DCA). Interpretation of the optimal model was conducted using shapley additive explanations (SHAP). A total of 6093 critically ill children were encompassed in this study, of whom 681 (11.2%) developed DIC. The RF model exhibited the highest levels of accuracy (0.856), sensitivity (0.866), Kappa (0.472), NPV (0.423), and recall (0.866). However, the XGB model outperformed RF, LR, SVM, DT, and KNN in terms of AUC (0.908), specificity (0.859), PPV (0.978), and precision (0.969). Decision curve analysis (DCA) confirmed the superior clinical utility of the XGB model. Overall, the XGB model demonstrated superior clinical utility compared to RF, LR, SVM, DT, and KNN. We named the final model Alfalfa-PICU-DIC. SHAP analysis identified D-dimer, INR, PT, TT, and PLT count as the top predictors of DIC. Machine learning models can be a reliable tool for predicting DIC in critically ill children, which will facilitate timely intervention, thereby reducing the burden of DIC on patients in the pediatric intensive care unit (PICU).
