Controlling nutritional status score predicts posthepatectomy liver failure: an online interpretable machine learning
Jun Yuan1, Rui Qing Zhang1, Qiang Guo1
1Department of Hepatobiliary and Echinococcosis Surgery, Digestive and Vascular Surgery Center, The First Affiliated Hospital.
European Journal of Gastroenterology & Hepatology
|April 10, 2025
Summary
The Controlling Nutritional Status (CONUT) score significantly impacts posthepatectomy liver failure (PHLF) in hepatocellular carcinoma (HCC) patients. A machine learning model using the CONUT score accurately predicts PHLF risk, aiding clinical decision-making.
Area of Science:
- Hepatobiliary surgery
- Oncology
- Medical informatics
Background:
- Posthepatectomy liver failure (PHLF) is a critical complication following liver resection for hepatocellular carcinoma (HCC).
- Accurate preoperative prediction of PHLF is essential for patient management and improving surgical outcomes.
- The Controlling Nutritional Status (CONUT) score, a nutritional index, has not been extensively evaluated for its role in predicting PHLF.
Purpose of the Study:
- To investigate the association between the CONUT score and the incidence of PHLF in patients undergoing hepatectomy for HCC.
- To develop and validate a machine learning (ML) model for the early identification of individuals at high risk of developing PHLF.
Main Methods:
- A cohort of 464 HCC patients undergoing hepatectomy was analyzed, with data split into training (n=324), test (n=94), and validation (n=46) groups.
- Univariate logistic regression and least absolute shrinkage and selection operator (LASSO) regression were used for feature selection in the training set.
- Nine ML algorithms were employed to build predictive models, with the optimal model selected, interpreted using SHapley Additive exPlanations (SHAP), and deployed online.
Main Results:
- PHLF occurred in 8.9% of patients in the training group.
- The Light Gradient Boosting Machine (LightGBM) model, incorporating the CONUT score, demonstrated strong predictive performance with an AUC of 0.927 in the training group.
- The model achieved robust performance across test and validation groups, with AUCs of 0.703 and 0.808, respectively, indicating its generalizability.
Conclusions:
- The CONUT score is a significant predictor of PHLF after hepatectomy for HCC.
- The developed LightGBM-based ML model exhibits high accuracy and clinical utility in predicting PHLF, facilitating risk stratification and personalized patient care.
