Predicting mortality in liver transplant candidates with eight variables: A machine learning study integrating SHAP and supervised clustering
Health Data Science
image: (A) Cluster 1, (B) Cluster 2, (C) Cluster 3, (D) Cluster 4, (E) Cluster 5, (F) Cluster 6, and (G) Cluster 7. Mortality probability for each medoid is in bold. Features in red increase risk (white arrows pointing right), while features in blue decrease risk (white arrows pointing left).
Credit: Nesma Houmani
Background
For patients with hepatocellular carcinoma (HCC) awaiting liver transplantation, mortality risk arises from two directions: hepatic dysfunction and tumour progression. The overlap complicates prognosis.
Existing scoring systems each address one end. Child–Pugh assesses hepatic functional reserve; the Albumin–Bilirubin (ALBI) grade reflects liver function; the Model for End-Stage Liver Disease (MELD) and its variants MELD-Na and MELD 3.0 centre on short-term mortality risk; alpha-fetoprotein (AFP) scores lean toward tumour biology. The paper notes that while these tools are widely used, they often fail to deliver precise risk assessment.
Methods
The study proposes a machine learning–based scoring system combining ensemble learning with SHapley Additive exPlanations (SHAP). SHAP provides both global and local explanation: the importance of variables across the population, and the basis for an individual case's prediction.
The design further embeds SHAP values in a UMAP (Uniform Manifold Approximation and Projection) space and performs supervised clustering there to infer latent subgroup structure.
Results
A system built on LightGBM outperformed conventional scores using only the 8 relevant variables selected by SHAP analysis. The authors note these variables correspond to the dual-risk problem the study set out to address.
Supervised clustering identified 7 subgroups with increasing mortality risk levels, and the contribution of risk factors differed across subgroups.
Implications
The authors argue that, in contrast to existing studies, the approach offers an integrative data-driven framework for handling the dual risk challenge faced by HCC patients with hepatic dysfunction. It also provides a tool for more precise risk evaluation that may guide treatment decisions and help monitor patient progression.
Limitations
This release does not report sample size, cohort source, or follow-up duration, as these methodological parameters were not retrieved from the source text. Readers intending any clinical use should consult the original article for full methods and the authors' stated scope of applicability.
Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.