An Explainable Machine Learning Framework for Mortality Risk Prediction of Liver Cirrhosis Patients in the U.S. Healthcare System

Abstract

Liver cirrhosis represents a significant source of morbidity and mortality within the United States healthcare system, placing a substantial burden on the U.S. biomedical and clinical care sector and increasing demand for reliable, system-relevant risk assessment tools. Although previous machine learning–based studies have demonstrated promising predictive performance, their limited interpretability, black-box decision-making, and insufficient alignment with U.S. clinical workflows have restricted widespread adoption. To address these challenges, this study presents an explainable and clinically interpretable machine learning framework for mortality risk prediction of liver cirrhosis patients within the U.S. healthcare system using routinely collected clinical and treatment-related data. A publicly available U.S. cirrhosis dataset was analyzed, and a Random Forest classifier was developed and rigorously evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis, with particular emphasis on minimizing false negative predictions to enhance patient safety. To overcome the transparency limitations of earlier approaches, SHapley Additive exPlanations (SHAP) were integrated in probability space to provide both global and patient-level interpretability. Experimental results demonstrate strong predictive performance while consistently identifying clinically meaningful risk factors, including age, ascites, edema, hepatomegaly, spider angiomas, and treatment type, reinforcing the clinical reliability of the proposed framework within U.S. healthcare environments.