多算法共识分类识别三种急性肝衰竭亚型及其差异治疗反应:多数据库队列研究
该研究纳入六个国际重症监护数据库的2,691例成人急性肝衰竭患者,以MIMIC-IV为发现队列(n=1,227),五个验证队列共1,464例。研究整合十种聚类算法建立多算法共识急性肝衰竭分类(MCALFC),识别出三种表型:亚型1(28.3%)为危重血流动力学崩溃,亚型2(44.6%)为心血管功能障碍,亚型3(27.0%)为超急性肝坏死。三种亚型28天死亡率不同(以亚型1为参照,亚型2 HR 0.48,95%CI 0.40-0.59;亚型3 HR 0.39,95%CI 0.31-0.50),且治疗反应存在异质性:肾上腺素在亚型1有害(OR 1.68),在亚型2和3有保护作用(OR 0.47、0.34);右美托咪定在亚型3获益(OR 0.26)但在亚型2有害(OR 1.47);肾脏替代治疗在亚型3风险最高(OR 2.05)。作者认为该表型分类为急性肝衰竭精准管理提供了框架,挑战当前一刀切治疗范式。
要点
- 三种急性肝衰竭表型:亚型1(28.3%)危重血流动力学崩溃,MAP 78.9 mmHg、HR 113.5次/分、SAPS II 62.0、SOFA 12.2;亚型2(44.6%)MAP 69.1 mmHg但HR 84.3次/分;亚型3(27.0%)MAP 99.4 mmHg、ALT 1059.8 U/L、AST 1427.3 U/L。
- 以亚型1为参照,28天死亡风险亚型2 HR 0.48(95%CI 0.40-0.59)、亚型3 HR 0.39(95%CI 0.31-0.50),差异维持至一年随访。
- 肾上腺素在亚型1有害(OR 1.68),在亚型2和3有保护作用(OR 0.47、0.34)。
- 右美托咪定在亚型3获益(OR 0.26)但在亚型2有害(OR 1.47);肾脏替代治疗在亚型3风险最高(OR 2.05)。
- 该分类在六个国际数据库共2,691例成人中验证,作者认为可为急性肝衰竭精准管理提供框架。
适用人群与场景
- 成人急性肝衰竭患者,来自六个国际重症监护数据库:MIMIC-IV发现队列n=1,227,五个验证队列(MIMIC-III CareVue、eICU、HiRID、NWICU、SICdb)共n=1,464。
诊断标准与关键阈值
- 多算法共识急性肝衰竭分类(MCALFC)整合十种聚类算法,基于临床参数、干预措施、合并症和药物识别亚型。
- 亚型1(28.3%)为危重血流动力学崩溃:MAP 78.9 mmHg、HR 113.5次/分、SAPS II 62.0、SOFA 12.2,需器官支持。
- 亚型2(44.6%)为心血管功能障碍:MAP 69.1 mmHg、HR 84.3次/分。
- 亚型3(27.0%)为超急性肝坏死:MAP 99.4 mmHg、ALT 1059.8 U/L、AST 1427.3 U/L。
研究设计与关键结果
- 多数据库队列研究:发现队列MIMIC-IV n=1,227,五个验证队列共n=1,464,总样本2,691例成人急性肝衰竭。
- 三种表型28天死亡率不同:以亚型1为参照,亚型2 HR 0.48(95%CI 0.40-0.59),亚型3 HR 0.39(95%CI 0.31-0.50),差异维持至一年随访。
- 治疗反应异质性:肾上腺素在亚型1有害(OR 1.68),在亚型2和3有保护作用(OR 0.47、0.34);右美托咪定在亚型3获益(OR 0.26)但在亚型2有害(OR 1.47);肾脏替代治疗在亚型3风险最高(OR 2.05)。
- 采用SHAP分析(XGBoost)识别关键区分特征。
证据类型、等级与局限
- 证据类型:多数据库队列研究(发现队列加五个验证队列)。
- 局限:原文未报告局限性。
对床旁实践的改变
- 作者认为该表型分类为急性肝衰竭精准管理提供了框架,挑战当前一刀切治疗范式。
- 研究提示肾上腺素在亚型1可能有害(OR 1.68),在亚型2和3可能有保护作用(OR 0.47、0.34),作者据此提出治疗应基于表型。
- 研究提示右美托咪定在亚型3可能获益(OR 0.26)但在亚型2可能有害(OR 1.47),肾脏替代治疗在亚型3风险最高(OR 2.05)。
提炼说明
- 本条依据论文摘要整理;摘要未涉及「处置流程与药物剂量」「红旗征与不能漏的情况」「指南或共识出处」,故未列出这些栏目;全文请点「阅读原文」。
英文原文(摘要)· 点击展开
BACKGROUND: Acute liver failure (ALF) is a heterogeneous syndrome with high mortality. Current prognostic models fail to capture pathophysiological heterogeneity, necessitating refined patient stratification and personalized therapeutic strategies.
METHODS: We analyzed 2,691 adult patients with ALF from six international critical care databases (MIMIC-IV as discovery cohort, n = 1,227; five validation cohorts: MIMIC-III CareVue, eICU, HiRID, NWICU, and SICdb, n = 1,464). We developed a Multi-algorithm Consensus-based Acute Liver Failure Classification (MCALFC) approach integrating ten clustering algorithms to identify distinct subtypes based on clinical parameters, interventions, comorbidities, and medications. We validated these subtypes across databases, performed survival analyses, conducted SHAP analysis using XGBoost to identify key distinguishing features, and evaluated treatment response heterogeneity.
RESULTS: Three robust ALF phenotypes emerged: Subtype 1 (28.3 %) - "critical hemodynamic collapse" with severe cardiovascular instability (MAP 78.9 mmHg, HR 113.5 bpm), highest illness severity (SAPS II: 62.0, SOFA: 12.2), and organ support requirements; Subtype 2 (44.6 %) - "cardiovascular dysfunction" with profound hypotension (MAP 69.1 mmHg) but low heart rate (84.3 bpm); and Subtype 3 (27.0 %) - "hyperacute hepatic necrosis" with preserved hemodynamics (MAP 99.4 mmHg) but severe hepatocellular injury (ALT 1059.8 U/L, AST 1427.3 U/L). Subtypes demonstrated distinct survival trajectories maintained through one-year follow-up (28-day mortality: Subtype 2 HR 0.48, 95 % CI: 0.40-0.59; Subtype 3 HR 0.39, 95 % CI: 0.31-0.50 vs. Subtype 1). Treatment responses varied significantly: epinephrine was harmful in Subtype 1 (OR 1.68) but protective in Subtypes 2 and 3 (OR 0.47, 0.34); dexmedetomidine benefited Subtype 3 (OR 0.26) but harmed Subtype 2 (OR 1.47); renal replacement therapy showed highest risk in Subtype 3 (OR 2.05).
CONCLUSION: ALF comprises three distinct phenotypes with unique pathophysiological features, prognostic trajectories, and treatment responses. This phenotypic classification, validated across multiple international databases, provides a framework for precision medicine in ALF management and challenges the current one-size-fits-all treatment paradigm.
参考来源
- Yang Y, Yang W, Tang B, et al. Multi-algorithm consensus classification identifies three distinct acute liver failure subtypes with differential treatment responses: a multi-database cohort study. J Adv Res. 2025. PMID: 40505959. doi:10.1016/j.jare.2025.06.019
供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。