A Machine Learning-Based Prognostic Model for Sepsis-Associated Liver Injury Using Routine Indicators.

Area of Science:

• * Critical care medicine
• * Machine learning in healthcare
• * Hepatology

Background:

• * Sepsis-associated liver injury (SALI) affects ~40% of sepsis cases, contributing to high mortality.
• * Current prognostic models for SALI are lacking, hindering timely clinical interventions.
• * Precise prognostic tools are needed to guide treatment and reduce mortality in SALI patients.

Purpose of the Study:

• * To develop and validate a machine learning (ML)-based prognostic model for SALI.
• * To utilize conventional biomarkers for predicting SALI patient outcomes.
• * To guide clinical decision-making and potentially reduce SALI-related mortality.

Main Methods:

• * Retrospective analysis of 307 SALI patients, split into training (80%) and validation (20%) sets.
• * Feature selection using LASSO regression on hematological, liver/renal function, and coagulation parameters.
• * Nine ML algorithms were trained, with the Random Forest model selected for performance evaluation via AUC and SHAP for interpretability.

Main Results:

• * Red blood cell distribution width-coefficient of variation (RDW-CV), anion gap (AG), and high-sensitivity cardiac troponin (hs-cTn) were key prognostic factors.
• * The Random Forest model achieved an AUC of 0.816 in the validation set and 0.781 in external validation.
• * SHAP analysis provided interpretability for the model's predictions.

Conclusions:

• * The developed Random Forest model demonstrates potential for guiding clinical decisions in SALI management.
• * Further external validation is necessary before widespread clinical implementation.
• * The model offers a promising tool for improving prognostic accuracy in sepsis-associated liver injury.