Liu Maojie, Jiang Longfeng, Yang Juan, Yao Yao, Puyang Xuerong, Ge Xinyuan, Lu Jing, Zhang Lu, Yan Yuqian, Shen Hongbing, Song Ci
Department of Epidemiology.
Departments of Infectious Disease.
J Clin Gastroenterol. 2025 Apr 25. doi: 10.1097/MCG.0000000000002166.
The development of accurate noninvasive tests to identify individuals with metabolic dysfunction-associated steatohepatitis (MASH) and liver fibrosis is of great clinical importance. In this study, we aimed to develop 2 noninvasive diagnostic models on the basis of routine clinical and laboratory data, using machine learning, to identify patients with MASH and significant fibrosis (fibrosis stages 2 to 4), respectively.
This analysis included the training (n=456) and the validation (n=105) sets of patients who underwent liver biopsy and laboratory testing for liver disease at 2 hospitals in China. Logistic regression, random forest, support vector machine, and the XGBoost algorithm were used to construct models, respectively. The best diagnostic models for MASH and significant fibrosis were compared with 7 existing noninvasive scoring systems including AAR, AST to platelet ratio index (APRI), BARD score, fibrosis-4 (FIB-4), fibrotic non-alcoholic steatohepatitis (NASH) index (FNI), homeostatic model assessment of insulin resistance (HOMA-IR), and non-alcoholic fatty liver disease fibrosis score (NFS). Performance was estimated by the area under the receiver operating characteristic curve (AUROC).
The final noninvasive diagnostic model integrated 19 indicators derived from routine clinical and laboratory tests. The XGBoost models exhibited superior performance in MASH and significant fibrosis with an improved AUROC value (MASH, 0.670, 95% CI 0.530-0.811; significant fibrosis, 0.713, 95% CI 0.611-0.815) compared with other noninvasive scoring systems in the validation set.
Utilizing machine learning can assist in diagnosing MASH and significant fibrosis based on clinical epidemiological information with good diagnostic performance.
开发准确的非侵入性检测方法以识别代谢功能障碍相关脂肪性肝炎（MASH）和肝纤维化患者具有重要的临床意义。在本研究中，我们旨在基于常规临床和实验室数据，利用机器学习分别开发2种非侵入性诊断模型，以识别MASH患者和显著纤维化（纤维化2至4期）患者。
该分析纳入了在中国2家医院接受肝活检及肝病实验室检测的患者的训练集（n = 456）和验证集（n = 105）。分别使用逻辑回归、随机森林、支持向量机和XGBoost算法构建模型。将MASH和显著纤维化的最佳诊断模型与7种现有的非侵入性评分系统进行比较，包括天冬氨酸氨基转移酶与碱性磷酸酶比值（AAR）、天冬氨酸氨基转移酶与血小板比值指数（APRI）、BARD评分、纤维化-4（FIB-4）、纤维化非酒精性脂肪性肝炎（NASH）指数（FNI）、胰岛素抵抗稳态模型评估（HOMA-IR）和非酒精性脂肪性肝病纤维化评分（NFS）。通过受试者操作特征曲线下面积（AUROC）评估性能。
最终的非侵入性诊断模型整合了19个源自常规临床和实验室检测的指标。在验证集中，与其他非侵入性评分系统相比，XGBoost模型在MASH和显著纤维化方面表现出更优性能，AUROC值有所提高（MASH为0.670，95%CI为0.530 - 0.811；显著纤维化为0.713，95%CI为0.611 - 0.815）。
利用机器学习可基于临床流行病学信息辅助诊断MASH和显著纤维化，诊断性能良好。
