LivMarX: An Optimized Low-Cost Predictive Model Using Biomarkers for Interpretable Liver Cirrhosis Stage Classification

Abstract

Liver cirrhosis, a progressive and irreversible condition characterized by the replacement of healthy liver tissue with scar tissue, presents a persistent challenge in healthcare. Leveraging a comprehensive dataset of 424 patients, including controlled trial data from 312 patients and real-world follow-up information from an additional 112 patients sourced from the Mayo Clinic trial on primary biliary cirrhosis, this study refines patient demographics and integrates biomarkers for a detailed analysis. Our paper introduces LivMarX, a novel model that utilises advanced machine learning algorithms within an interdisciplinary framework and strives to stage Liver Cirrhosis using biomarkers instead of images. Our study employs meticulous feature engineering techniques, the creation of synthetic variables and categorizations to unveil critical relationships between patient characteristics and disease stages. To further refine performance, hyperparameter optimization is implemented, combining Genetic Algorithm, Optuna, and GridSearchCV. The Random Forest Classifier in LivMarX outperformed other models with an accuracy of 84.33%, and post-optimization, an accuracy improvement of 86%. LivMarX demonstrates an AUC of 0.95, showing that it provides reliable staging of liver cirrhosis. By relying on common blood tests instead of expensive imaging, this offers a cost-effective and comfortable approach to Liver Cirrhosis diagnosis. This study positions LivMarX as a potential model for accurately classifying Liver Cirrhosis stages, particularly in areas with limited access to complex imaging equipment.

Keywords

• Liver cirrhosis
• machine learning
• health determinants
• healthcare
• disease prediction