What question did this study set out to answer?

April 16, 2026Journal of Hepatocellular Carcinoma0 citationsOpen Access

Development and Validation of Predictive Machine Learning Models for Postoperative Recurrence and Microvascular Invasion in Hepatocellular Carcinoma Using Nuclear Magnetic Resonance Metabolomics

HTHongkun TanZhujiang Hospital YXYuyan XuZhujiang Hospital WLWenxuan LiuZhujiang Hospital

Key Points

The aim is to identify plasma metabolic biomarkers and develop predictive models for postoperative recurrence and microvascular invasion in hepatocellular carcinoma.
Conducted untargeted NMR metabolomic profiling of plasma from 92 HCC patients and 92 matched healthy controls
Identified differential metabolites and assessed their diagnostic performance using receiver operating characteristic curves
Developed predictive models using various machine learning algorithms including random forest and support vector machine
Validated models with performance metrics for clinical implementation
Identified 67 metabolites with significant alterations related to HCC
Notable biomarkers included acetic acid and dimethylsulfone with AUC of ≥ 0.954
The StepCox[forward] + random survival forest model achieved an AUC of 0.811 for recurrence prediction
The support vector machine model demonstrated superior performance for MVI with an AUC of 0.957
Developed two online platforms for clinical application and validation

Abstract

Hongkun Tan,1, Yuyan Xu,1, Wenxuan Liu,1, Yaohong Wen,1 Cheng Zhang,1 Chunming Wang,1 Luhao Chi,1 Hangyu Liao,1 Shunjun Fu,1 Lei Cai,1 Hongbo Guo,2 Mingxin Pan1 1Department of Hepatobiliary Surgery II, General Surgery Center, Zhujiang Hospital, Southern Medical University, Guangzhou, Peopleâs Republic of China; 2Neurosurgery Center, Zhujiang Hospital, Southern Medical University, Guangzhou, Peopleâs Republic of ChinaThese authors contributed equally to this workCorrespondence: Hongbo Guo, Email guohongbo911@126.com Mingxin Pan, Email panmx@smu.edu.cnBackground: Hepatocellular carcinoma (HCC) has a poor prognosis, necessitating better diagnostic tools. Nuclear magnetic resonance (NMR)-based metabolomics has emerged as a powerful tool for cancer biomarker discovery, yet its application in HCC prognosis remains underexplored. This study aimed to identify plasma metabolic biomarkers for the diagnosis and prognosis of HCC, and to develop predictive models for postoperative recurrence and microvascular invasion (MVI) to enhance clinical management.Methods: We performed untargeted NMR metabolomic profiling of plasma from 92 HCC patients and 92 matched healthy controls. Differential metabolites were identified, and their diagnostic performance was assessed using receiver operating characteristic curves. Predictive models for postoperative recurrence and MVI were developed and validated using multiple machine learning algorithms, such as random forest, support vector machine, and gradient boosting machine models.Results: Significant metabolic differences were identified, with 67 metabolites and blood-lipid indicators showing marked alterations. Acetic acid, dimethylsulfone, glycerol, glycine, and low-density lipoprotein (LDL)-3 cholesterol exhibited the highest discriminatory power (area under the curve AUC â¥ 0.954). Regarding HCC recurrence prediction, the StepCoxforward + random survival forest model achieved an AUC of 0.811 and was an independent prognostic indicator (multivariate Cox HR = 1.20, 95% CI: 1.11â 1.30, P < 0.001). Regarding MVI prediction, the support vector machine model demonstrated superior performance (AUC = 0.957). Calibration curve, decision curve, and SHapley Additive exPlanations (SHAP) analyses confirmed model robustness and clinical utility. Two online platforms were developed for clinical implementation.Conclusion: This study developed and validated NMR-based prognostic and MVI prediction models for HCC, offering valuable tools for precision management. Their clinical value warrants further validation in larger prospective cohorts.Keywords: hepatocellular carcinoma, metabolomics, metabolites, microvascular invasion, NMR

Perguntar à IA

Bookmark

View Full Paper