Machine learning in primary biliary cholangitis: A novel approach for risk stratification.

Area of Science:

• Hepatology
• Medical Informatics
• Biostatistics

Background:

• Machine learning (ML) offers novel approaches for patient prognostication by identifying distinct subgroups.
• Primary biliary cholangitis (PBC) management can benefit from advanced analytical methods for sub-phenotyping and risk stratification.

Purpose of the Study:

• To explore the utility of machine learning (ML) in defining patient subgroups within primary biliary cholangitis (PBC).
• To assess ML's capability for risk stratification in PBC patients.
• To identify novel prognostic markers, including subtle variations in albumin levels.

Main Methods:

• An international PBC patient dataset was utilized, divided into training and validation cohorts.
• Unsupervised machine learning algorithms were applied to identify patient clusters based on clinical features.
• Standard survival analysis was performed alongside ML to compare prognostic accuracy.

Main Results:

• ML identified four distinct patient clusters with varying phenotypes and prognoses.
• Cluster 2 patients showed worse prognosis, differentiated by albumin levels near the normal limit.
• An increase in albumin above 1.2 times the lower limit of normal (LLN), particularly with ursodeoxycholic acid treatment, correlated with improved transplant-free survival.

Conclusions:

• Unsupervised ML successfully delineated four novel prognostic groups in PBC patients.
• Subtle variations in albumin levels within the normal range are significant prognostic indicators.
• Achieving a therapy-induced increase in albumin above 1.2 x LLN should be a therapeutic target in PBC management.