Practical Machine Learning Model for Early and Accurate Prediction of Disseminated Intravascular Coagulation Before...

Area of Science:

• Medical Informatics
• Clinical Medicine
• Machine Learning in Healthcare

Background:

• Sepsis management requires early intervention to prevent complications like overt disseminated intravascular coagulation (DIC).
• Anticoagulant therapy is most effective when initiated before the onset of overt DIC.
• Accurate prediction of sepsis-induced overt DIC is crucial for timely treatment.

Purpose of the Study:

• To develop and evaluate machine learning models for the early prediction of sepsis-induced overt DIC.
• To identify key predictive features from electronic medical records for overt DIC development.

Main Methods:

• A multi-center retrospective observational study of adult sepsis patients without overt DIC at day 1.
• Development of three machine learning models (minimum, compact, full) using XGBoost and GBM algorithms.
• Evaluation of models on a 20% test set, assessing prediction accuracy using AUC-ROC and AUC-PR.

Main Results:

• Out of 7,532 sepsis patients, 766 developed overt DIC within 7 days.
• XGBoost and GBM models demonstrated high prediction accuracy, with AUC-ROC values of 0.916 (full model).
• The full XGBoost model achieved 80% recall with 14.4% precision.

Conclusions:

• Machine learning models can accurately predict overt DIC in sepsis patients at a clinically relevant level.
• Early prediction enables timely intervention with anticoagulant therapy.
• These models offer a promising tool for improving sepsis management and patient outcomes.