Using machine learning methods to predict 28-day mortality in patients with hepatic encephalopathy

Area of Science:

• Medical Informatics
• Machine Learning in Medicine
• Critical Care Medicine

Background:

• Hepatic encephalopathy (HE) significantly increases mortality in cirrhosis patients.
• Developing predictive models for HE mortality is crucial for patient management.
• This study focuses on machine learning for predicting 28-day mortality in HE.

Purpose of the Study:

• To develop and validate machine learning models for predicting 28-day mortality in patients with HE.
• To identify independent risk factors for 28-day mortality in HE patients.
• To compare the performance of developed ML models against existing clinical scores.

Main Methods:

• Retrospective cohort study using the MIMIC-IV database.
• Patients were divided into training (70%) and validation (30%) cohorts.
• Recursive Feature Elimination (RFE) identified predictors within 24 hours of ICU admission.
• Area Under the Curve (AUC) and calibration curves assessed model performance.

Main Results:

• 601 HE patients identified; 18.64% experienced 28-day mortality.
• Key predictors included APSIII, SOFA, INR, TBIL, albumin, BUN, AKI, and mechanical ventilation.
• The Artificial Neural Network (NNET) model achieved the highest AUC (0.837) and demonstrated good calibration.
• NNET performance surpassed MELD and MELD-Na scores.

Conclusions:

• The NNET model shows superior discrimination for predicting 28-day mortality in HE patients.
• This ML model may enhance early detection and improve clinical outcomes for HE.
• Further external prospective validation is necessary to confirm the model's generalizability.