Explainable Mortality Prediction for Liver Transplant Candidates with Hepatocellular Carcinoma: A Supervised...

Area of Science:

• Hepatology and Transplant Surgery
• Artificial Intelligence in Medicine
• Oncology

Background:

• Accurate mortality prediction for liver transplant candidates with hepatocellular carcinoma (HCC) is a significant clinical challenge.
• Existing scoring systems (Child-Pugh, MELD, etc.) often lack precision due to the dual burden of liver dysfunction and tumor progression.
• A comprehensive approach is needed to effectively manage HCC patients.

Purpose of the Study:

• To develop an advanced machine learning-based scoring system for improved mortality risk prediction in HCC patients.
• To leverage Ensemble Learning and SHapley Additive exPlanations (SHAP) for a deeper understanding of mortality risk factors.
• To identify latent patient subgroups for more granular risk assessment.

Main Methods:

• An advanced machine learning model utilizing Ensemble Learning and SHAP was developed.
• SHAP values were embedded in Uniform Manifold Approximation and Projection (UMAP) space for supervised clustering.
• Latent subgroups were inferred to provide granular insights into variable contributions to mortality risk.

Main Results:

• The LightGBM-based system outperformed conventional scores, utilizing only 8 key variables identified by SHAP analysis.
• These selected variables effectively addressed the dual risk challenge of liver dysfunction and tumor progression.
• Supervised clustering identified 7 distinct subgroups with increasing mortality risk levels and detailed risk factor contributions.

Conclusions:

• The proposed data-driven framework offers an integrative approach to the dual risk challenge in HCC patients with liver dysfunction.
• This method provides a more precise risk evaluation tool compared to existing studies.
• The findings may guide treatment decisions and improve patient progression monitoring.