人工智能在代谢功能障碍相关脂肪性肝病中的三级应用框架
本文为叙述性综述,提出按决策主体将AI在MASLD诊疗中的应用分为患者、卫生系统和政策三个层级。患者层面AI可用于纤维化评估、影像与病理判读、多组学整合及HCC风险分层;系统与政策层面可支持病例发现和资源配置。现有研究多为单中心回顾性,尚无随机试验证实AI改善临床结局。
要点
- AI在MASLD患者层面可用于纤维化评估、影像与病理判读及HCC风险分层。
- 目前尚无经验证的AI工具能在治疗前预测治疗反应。
- 多数研究为单中心回顾性,缺乏外部验证和随机试验证据。
核心结论
- AI在MASLD诊疗的三个层级均展现潜在价值,但尚无随机试验证实AI指导的管理可改善临床结局。
研究设计与样本
- 叙述性综述,按AI输出所服务的决策主体分为患者、卫生系统和健康政策三个层级进行讨论。
主要发现
- 患者层面,机器学习和深度学习方法在纤维化评估、组织病理与影像判读、多组学整合、治疗反应预测及肝细胞癌风险分层中展现潜在应用。
- 在部分选定数据集中,若干模型表现与常规无创检测或专家判读相当或更优。
- 目前尚无经验证的工具可在治疗前预测治疗反应。
- 卫生系统层面,AI可支持人群病例发现、从非结构化健康记录中提取临床相关信息、预后评估及优化确证检测与专科资源分配。
- 政策层面,AI辅助的疾病负担建模与风险分层可帮助指导人力配置、资源分配、报销与治疗覆盖决策。
临床怎么用
- 现有证据尚不足以支持将AI工具常规用于MASLD临床决策,应用应限于研究或明确界定的决策场景。
- 临床使用前需关注AI工具的外部验证情况、报告规范及公平性问题。
局限性
- 多数现有研究为回顾性研究,且常来自单中心,外部验证有限。
- 尚无随机试验证明AI指导的管理可改善MASLD临床结局。
- 治理、透明度和公平性框架仍不完善。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Metabolic dysfunction-associated steatotic liver disease (MASLD) is highly prevalent worldwide, yet most affected individuals remain undiagnosed, and pathways for noninvasive risk stratification and linkage to specialty care remain inconsistently implemented. With the emergence of pharmacologic therapies for metabolic dysfunction-associated steatohepatitis, accurate patient identification, disease staging, treatment selection, and longitudinal monitoring have become increasingly consequential. In this narrative review, we examine the evolving role of artificial intelligence (AI) across 3 levels of MASLD care, defined by where an AI output informs a decision and who acts on it: the individual patient, the health system, and health policy. At the patient level, machine-learning and deep-learning approaches have demonstrated potential applications in fibrosis assessment, histopathologic and imaging interpretation, multiomics integration, treatment-response prediction, and hepatocellular carcinoma risk stratification. Across selected datasets, several models have performed comparably to or better than conventional noninvasive tests and expert interpretation, although no validated tool currently predicts treatment response before therapy. At the health-system level, AI may support population-level case finding, extraction of clinically relevant information from unstructured health records, prognostic assessment, and more efficient allocation of confirmatory testing and specialty care. At the policy level, AI-informed disease-burden modeling and risk stratification may help guide decisions regarding workforce capacity, resource allocation, reimbursement, and treatment coverage. However, most available studies are retrospective, frequently originate from single centers, and have limited external validation. No randomized trial has yet demonstrated that AI-guided management improves clinical outcomes in MASLD, whereas governance, transparency, and equity frameworks remain underdeveloped. Realizing the clinical value of AI will require prospective multicenter studies embedded within clearly defined MASLD care decisions, validation across genetically and socioeconomically diverse populations and health system settings, standardized reporting and risk-of-bias assessment, and sustained attention to implementation, governance, and equity.
引用
- Kalligeros M, Papatheodoridi M, Henry L, et al. Artificial Intelligence in Metabolic Dysfunction-Associated Steatotic Liver Disease: A 3-Level Framework Across Clinical, Health System, and Policy Decisions. Clin Gastroenterol Hepatol. 2026. PMID: 42822578. doi:10.1016/j.cgh.2026.09.034
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。