Unified Machine Learning and Survival Analysis Framework for Liver Cirrhosis Patient Classification and Mortality Risk Stratification

Abstract

Chronic liver disease and its progression to cirrhosis remain major global health challenges owing to delayed diagnosis, complex progression patterns, and limited predictive tools for patient-specific prognosis. This study proposes a comprehensive machine learning-based framework for classifying the clinical status of patients with cirrhosis and predicting survival outcomes, enabling proactive care. Using a real-world dataset of 418 patients with demographic, clinical, biochemical, and histological features, we developed a two-stage pipeline: (1) classification of patient status into censored, transplantation, and death categories using ensemble learning models and (2) survival modeling to estimate time-to-event risk and stratify patients by mortality probability. Data preprocessing included the rigorous handling of missing values, encoding, and class balancing with SMOTE to mitigate bias. Among the models tested, Random Forest achieved the highest predictive accuracy (86.4 %) and robust generalization (macro AUC <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=0.963$</tex>). The Cox Proportional Hazards model achieved a strong concordance index (0.89), identifying albumin, platelet count, and histological stage as key predictors of mortality. Kaplan-Meier analysis and risk stratification further revealed heterogeneity in survival trajectories, enabling the early detection of high-risk patients. This framework demonstrates both predictive power and interpretability, offering a clinically meaningful tool for early risk assessment and individualized patient monitoring in clinical practice. These findings highlight the potential of machine learning-driven survival modeling to support evidence-based interventions, reduce cirrhosis-related mortality, and enhance healthcare resource allocation.

Ähnliche Arbeiten