Abstract

Despite medical advancements, accurately predicting and assessing cirrhosis risk remains complex. Established methods like MELD, FIB-4 Index, and ALBI score provide insights, yet new approaches like machine learning are needed. This study compares advanced machine learning with quantitative risk scores, focusing on hepatic biomarkers for cirrhosis risk assessment.

Methods - Three machine learning models (XGBoost, random forest, decision tree) and three risk scores (MELD, FIB-4 Index, ALBI score) were compared using hepatic biomarkers.

Results - Performance was evaluated using the area under the curve (AUC). XGBoost had the highest (0.95) discriminative capability among machine learning models, while MELD (0.89) outperformed risk scores, including the decision tree. Notable cirrhosis risk features were International Normalized Ratio (INR), albumin, age, prothrombin time (PT), total bilirubin, alanine aminotransferase (ALT). ALP, and indirect bilirubin.

Conclusions - Machine learning, leveraging pre-treatment hepatic biomarkers, excels in cirrhosis risk assessment, surpassing most risk scores. MELD stands out as a reliable risk assessment tool.

Keywords

Cirrhosis; Risk assessment; Hepatology; Machine learning; MELD Score; ALBI Score; FIB-4 Index

Introduction