机器学习识别儿童脓毒症微血管表型:对器官功能障碍与死亡率的预后意义
该研究为前瞻性脓毒症队列(2018–2024)的二次分析,按时间拆分推导队列(2018–2022)与验证队列(2023–2024),纳入239例患儿(中位年龄2.1岁),整合人口学、临床变量、生物标志物及舌下 videomicroscopy(GlycoCheck®)数据,采用混合数据因子分析与Ward.D2无监督聚类识别表型,主要终点为入科后72小时多器官功能障碍综合征(MODS)持续存在。共识别三种微血管表型:保留型(n=52)、部分代偿型(n=75)、失代偿型(n=112)。失代偿型较保留型72小时MODS风险更高(aHR 2.47;95%CI 1.04–8.44;p=0.042),MODS发生率分别为42%与10%,独立于年龄、感染灶及有创机械通气需求。死亡率呈阶梯式升高(保留型1.9%、部分代偿型12.1%、失代偿型17.2%;p=0.023)。内部验证分析重现了表型分配与结局关联,证实其时间稳定性。结论认为早期微血管表型分型可行,可增强风险分层。
要点
- 儿童脓毒症可识别三种微血管表型:保留型(n=52)、部分代偿型(n=75)、失代偿型(n=112),共239例。
- 失代偿型较保留型72小时MODS风险更高(aHR 2.47;95%CI 1.04–8.44;p=0.042),MODS率42% vs 10%。
- 死亡率呈阶梯式升高:保留型1.9%、部分代偿型12.1%、失代偿型17.2%(p=0.023)。
- 失代偿型以更年幼、多器官功能障碍与急性肾损伤更频繁、普遍机械通气、毛细血管密度最低为特征。
- 部分代偿型炎症激活显著,需更多血管活性药物与通气支持,PICU停留更长。
适用人群与场景
- 前瞻性脓毒症队列(2018–2024)的二次分析,纳入239例患儿,中位年龄2.1岁(IQR 0.5–11.1),52%为女性。
- 推导队列为2018–2022年,验证队列为2023–2024年。
研究设计与关键结果
- 研究类型为前瞻性脓毒症队列的二次分析,样本量239例,按时间拆分推导队列(2018–2022)与验证队列(2023–2024)。
- 采用混合数据因子分析降维后以Ward.D2无监督聚类,识别三种微血管表型:保留型(n=52)、部分代偿型(n=75)、失代偿型(n=112)。
- 主要终点为入科后72小时MODS持续存在;失代偿型较保留型风险更高(aHR 2.47;95%CI 1.04–8.44;p=0.042),MODS率分别为42%与10%,独立于年龄、感染灶及有创机械通气需求。
- 死亡率呈阶梯式升高:保留型1.9%、部分代偿型12.1%、失代偿型17.2%(p=0.023)。
- 保留型患者年龄较大、器官支持需求较少,微血管指标有利(毛细血管密度较高、糖萼轻度改变);部分代偿型炎症激活显著,需更多血管活性药物与通气支持、PICU停留更长,微循环示毛细血管密度中度降低与轻微糖萼改变;失代偿型以更年幼、多器官功能障碍与急性肾损伤更频繁、普遍机械通气、毛细血管密度最低及全面微血管衰竭为特征。
- 内部验证分析重现了表型分配与结局关联,证实其时间稳定性。
证据类型、等级与局限
- 证据类型:前瞻性脓毒症队列(2018–2024)的二次分析,采用时间拆分推导与验证队列,内部验证分析重现表型分配与结局关联。
- 局限:原文未报告局限性。
对床旁实践的改变
- 作者认为早期微血管表型分型可行,可增强风险分层;该分型整合床旁临床、生物标志物与微循环数据,识别从预后良好的保留状态到与器官功能障碍和死亡风险更高相关的失代偿状态。
提炼说明
- 本条依据论文摘要整理;摘要未涉及「诊断标准与关键阈值」「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 : Pediatric sepsis exhibits marked heterogeneity. Early integration of clinical, biomarker, and microcirculatory data may refine risk stratification and identify reproducible phenotypes associated with distinct outcomes. METHODS: We performed a secondary analysis of a prospective sepsis cohort (2018–2024). A temporal split defined derivation (2018–2022) and validation (2023–2024) cohorts, including demographics, clinical variables, biomarkers, and sublingual videomicroscopy (GlycoCheck®). Dimensionality reduction was performed using Factorial Analysis of Mixed Data, followed by unsupervised clustering with Ward.D2. The primary outcome was persistence of multiple organ dysfunction syndrome (MODS) at 72 h after admission. RESULTS: Among 239 patients (median age 2.1 years [IQR 0.5–11.1]; 52% female), three distinct microvascular phenotypes were identified: Preserved (n = 52), Partially Compensated (n = 75), and Decompensated (n = 112). Patients in the Preserved phenotype were older, required less organ support, and exhibited favorable microvascular indices, characterized by higher capillary density and mild glycocalyx alteration. Partially Compensated patients exhibited marked inflammatory activation, requiring greater vasoactive and ventilatory support and longer PICU stays. Microcirculatory assessment showed moderate reductions in capillary density and subtle glycocalyx changes, indicating a partially compensated response to inflammatory stress. Decompensated patients were predominantly younger and had more frequent multi-organ dysfunction, acute kidney injury, and universal mechanical ventilation, with the lowest capillary density and evidence of global microvascular failure. The risk of MODS within 72 h was highest in the Decompensated phenotype compared with the Preserved group (aHR 2.47; 95% CI 1.04–8.44; p = 0.042), with corresponding MODS rates of 42% and 10%, respectively, independent of age, infection focus, and the need for invasive mechanical ventilation. Mortality followed a stepwise gradient (Preserved 1.9%, Partially Compensated 12.1%, Decompensated 17.2%; p = 0.023). Internal validation analyses reproduced phenotype assignments and outcome associations, confirming their temporal stability. CONCLUSIONS: Three reproducible microvascular phenotypes in pediatric sepsis—Preserved, Partially Compensated, and Decompensated—were identified using bedside clinical, biomarker, and microcirculation data. These phenotypes represent distinct pathophysiologic and prognostic patterns, from preserved states with favorable outcomes to decompensated forms linked to higher risk of organ dysfunction and death. Early microvascular phenotyping is feasible and enhances risk stratification.
参考来源
- Beltrán JC, Fernández-Sarmiento J, Orjuela-Cañón AD, et al. Microvascular phenotypes in pediatric sepsis identified by machine learning: prognostic implications for organ dysfunction and mortality. Crit Care. 2026. PMID: 41645301. doi:10.1186/s13054-025-05785-x
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。