Clinical and Molecular Hepatology (Jul 2025)

Fibrosis-4plus score: a novel machine learning-based tool for screening high-risk varices in compensated cirrhosis (CHESS2004): an international multicenter study

  • Bingtian Dong,
  • Ruiling He,
  • Shenghong Ju,
  • Yuping Chen,
  • Ivica Grgurevic,
  • Jianzhong Ma,
  • Ying Guo,
  • Huizhen Fan,
  • Qiang Yan,
  • Chuan Liu,
  • Huixiong Xu,
  • Anita Madir,
  • Kristian Podrug,
  • Jia Wang,
  • Linxue Qian,
  • Zhengzi Geng,
  • Shanghao Liu,
  • Tao Ren,
  • Guo Zhang,
  • Kun Wang,
  • Meiqin Su,
  • Fei Chen,
  • Sumei Ma,
  • Liting Zhang,
  • Zhaowei Tong,
  • Yonghe Zhou,
  • Xin Li,
  • Fanbin He,
  • Hui Huan,
  • Wenjuan Wang,
  • Yunxiao Liang,
  • Juan Tang,
  • Fang Ai,
  • Tingyu Wang,
  • Liyun Zheng,
  • Zhongwei Zhao,
  • Jiansong Ji,
  • Wei Liu,
  • Jiaojiao Xu,
  • Bo Liu,
  • Xuemei Wang,
  • Yao Zhang,
  • Qiong Yan,
  • Hui Liu,
  • Xiaomei Chen,
  • Shuhua Zhang,
  • Yihua Wang,
  • Yang Liu,
  • Li Yin,
  • Yanni Liu,
  • Yanqing Huang,
  • Li Bian,
  • Ping An,
  • Xin Zhang,
  • Shaoting Zhang,
  • Jinhua Shao,
  • Xiangman Zhang,
  • Wei Rao,
  • Chaoxue Zhang,
  • Christoph Frank Dietrich,
  • Won Kim,
  • Xiaolong Qi

DOI
https://doi.org/10.3350/cmh.2024.0898
Journal volume & issue
Vol. 31, no. 3
pp. 881 – 898

Abstract

Read online

Background/Aims A large percentage of patients undergoing esophagogastroduodenoscopy (EGD) screening do not have esophageal varices (EV) or have only small EV. We evaluated a large, international, multicenter cohort to develop a novel score, termed FIB-4plus, by combining the fibrosis-4 (FIB-4) score, liver stiffness measurement (LSM), and spleen stiffness measurement (SSM) to identify high-risk EV (HRV) in compensated cirrhosis. Methods This international cohort study involved patients with compensated cirrhosis from 17 Chinese hospitals and one Croatian institution (NCT04546360). Two-dimensional shear wave elastography-derived LSM and SSM values, and components of the FIB-4 score (i.e., age, aspartate aminotransferase, alanine aminotransferase, and platelet count [PLT]) were combined using machine learning algorithms (logistic regression [LR] and extreme gradient boosting [XGBoost]) to develop the LR-FIB-4plus and XGBoost-FIB-4plus models, respectively. Shapley Additive exPlanations method was used to interpret the model predictions. Results We analyzed data from 502 patients with compensated cirrhosis who underwent EGD screening. The XGBoost-FIB-4plus score demonstrated superior predictive performance for HRV, with an area under the receiver operating characteristic curve (AUROC) of 0.927 (95% confidence interval [CI] 0.897–0.957) in the training cohort (n=268), and 0.919 (95% CI 0.843–0.995) and 0.902 (95% CI 0.820–0.984) in the first (n=118) and second (n=82) external validation cohorts, respectively. Additionally, the XGBoost-FIB-4plus score exhibited high AUROC values for predicting EV across all cohorts. The FIB-4plus score outperformed the individual parameters (LSM, SSM, PLT, and FIB-4). Conclusions The FIB-4plus score effectively predicted EV and HRV in patients with compensated cirrhosis, providing clinicians with a valuable tool for optimizing patient management and outcomes.

Keywords