Coagulation Risk Prediction in Patients With Liver Failure: Integrated Meta-Analysis and Machine Learning Model Study

Area of Science:

• Hepatology and critical care medicine
• Biomedical engineering and artificial organ development
• Data science and machine learning in healthcare

Background:

• Liver failure frequently causes severe coagulation dysfunction, a major clinical challenge.
• Artificial liver support systems (ALSS) show promise in improving coagulation parameters.
• Predictive modeling for coagulation dynamics in ALSS remains underexplored.

Purpose of the Study:

• To assess the impact of ALSS on coagulation function in liver failure.
• To develop a dynamic machine learning model for predicting coagulation parameter improvements.
• To identify key predictors of coagulation abnormalities during ALSS treatment.

Main Methods:

• Systematic literature review and meta-analysis of 18 studies (1771 patients).
• Evaluation of ALSS effects on international normalized ratio (INR), prothrombin time (PT), activated partial thromboplastin time (APTT), and fibrinogen.
• Development of machine learning models (logistic regression, XGBoost, random forest, LSTM) using ICU data.

Main Results:

• ALSS significantly improved INR, PT, APTT, and fibrinogen (P<.05).
• Treatment efficacy varied across different ALSS modalities.
• The random forest model achieved the highest predictive performance (AUC 92.12%), with dynamic INR as a key predictor.

Conclusions:

• ALSS effectively improves coagulation parameters in liver failure patients, with variable efficacy by modality.
• A machine learning model accurately predicts coagulation dysfunction risk, supporting early recognition and personalized management.
• The model serves as a dynamic risk assessment tool to aid clinical evaluation and nursing interventions.