Lin Huapeng, Cheuk-Fung Yip Terry, Lee Hye Won, Meng Xiangjun, Che-To Lai Jimmy, Ahn Sang Hoon, Pang Wenjing, Lai-Hung Wong Grace, Zeng Lingfeng, Wai-Sun Wong Vincent, de Lédinghen Victor, Kim Seung Up
Department of Gastroenterology and Hepatology, Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China; Center for Digestive Diseases Research and Clinical Translation of Shanghai Jiao Tong University, Shanghai, China; Shanghai Key Laboratory of Gut Microecology and Associated Major Diseases Research, Shanghai, China; Medical Data Analytics Center, Department of Medicine and Therapeutics, The Chinese University of Hong Kong, Hong Kong.
Medical Data Analytics Center, Department of Medicine and Therapeutics, The Chinese University of Hong Kong, Hong Kong; State Key Laboratory of Digestive Disease, The Chinese University of Hong Kong, Hong Kong.
J Hepatol. 2025 Mar;82(3):456-463. doi: 10.1016/j.jhep.2024.09.020. Epub 2024 Sep 20.
BACKGROUND & AIMS: Direct-acting antivirals (DAAs) have considerably improved chronic hepatitis C (HCV) treatment; however, follow-up after sustained virological response (SVR) typically neglects the risk of liver-related events (LREs). This study introduces and validates the artificial intelligence-safe score (AI-Safe-C score) to assess the risk of LREs in patients without cirrhosis after successful DAA treatment.
The random survival forest model was trained to predict LREs in 913 patients without cirrhosis after SVR in Korea and was further tested in a combined cohort from Hong Kong and France (n = 1,264). The model's performance was assessed using Harrell's C-index and the area under the time-dependent receiver-operating characteristic curve (AUROC).
The AI-Safe-C score, which incorporated liver stiffness measurement (LSM), age, sex, and six other biochemical tests - with LSM being ranked as the most important among nine clinical features - demonstrated a C-index of 0.86 (95% CI 0.82-0.90) in predicting LREs in an external validation cohort. It achieved 3- and 5-year LRE AUROCs of 0.88 (95% CI 0.84-0.92) and 0.79 (95% CI 0.71-0.87), respectively, and for hepatocellular carcinoma, a C-index of 0.87 (95% CI 0.81-0.92) with 3- and 5-year AUROCs of 0.88 (95% CI 0.84-0.93) and 0.82 (95% CI 0.75-0.90), respectively. Using a cut-off of 0.7, the 5-year LRE rate within a high-risk group was between 3.2% and 6.2%, mirroring the incidence observed in individuals with advanced fibrosis, in stark contrast to the significantly lower incidence of 0.2% to 0.6% in a low-risk group.
The AI-Safe-C score is a useful tool for identifying patients without cirrhosis who are at higher risk of developing LREs. The post-SVR LSM, as integrated within the AI-Safe-C score, plays a critical role in predicting future LREs.
The AI-Safe-C score introduces a paradigm shift in the management of patients without cirrhosis after direct-acting antiviral treatment, a cohort traditionally not included in routine surveillance protocols for liver-related events. By accurately identifying a subgroup at a comparably high risk of liver-related events, akin to those with advanced fibrosis, this predictive model facilitates a strategic reallocation of surveillance and clinical resources.
直接抗病毒药物（DAAs）显著改善了慢性丙型肝炎（HCV）的治疗；然而，持续病毒学应答（SVR）后的随访通常忽略了肝脏相关事件（LREs）的风险。本研究引入并验证了人工智能安全评分（AI-Safe-C评分），以评估成功接受DAA治疗后无肝硬化患者发生LREs的风险。
采用随机生存森林模型，对韩国913例SVR后无肝硬化的患者发生LREs的情况进行预测，并在来自中国香港和法国的联合队列（n = 1264）中进一步测试。使用Harrell's C指数和时间依赖性受试者工作特征曲线下面积（AUROC）评估模型性能。
AI-Safe-C评分纳入了肝脏硬度测量（LSM）、年龄、性别和其他六项生化检测指标（LSM在九项临床特征中被列为最重要的指标），在外部验证队列中预测LREs的C指数为= 0.86（95%CI 0.82-0.90）。其3年和5年LRE的AUROC分别为0.88（95%CI 0.84-0.92）和0.79（9%CI 0.71-0.87），对于肝细胞癌，C指数为0.87（95%CI 0.81-0.92），3年和5年AUROC分别为0.88（95%CI 0.84-0.93）和0.82（95%CI 0.75-0.90）。以0.7为临界值，高危组5年LRE发生率在3.2%至6.2%之间，与晚期纤维化患者的发生率相近，而低危组的发生率显著较低，为0.2%至0.6%。
AI-Safe-C评分是识别无肝硬化但发生LREs风险较高患者的有用工具。纳入AI-Safe-C评分的SVR后LSM在预测未来LREs方面起着关键作用。
AI-Safe-C评分在直接抗病毒治疗后无肝硬化患者的管理中引入了范式转变，这一群体传统上未被纳入肝脏相关事件的常规监测方案。通过准确识别出与晚期纤维化患者风险相当高的肝脏相关事件亚组，该预测模型有助于对监测和临床资源进行战略重新分配。
