Scholarlink

<p>Sepsis-associated acute kidney injury (SA-AKI) is a highly heterogeneous syndrome with variable clinical trajectories and outcomes. Conventional staging systems, such as KDIGO, fail to capture this heterogeneity, limiting risk stratification and personalized management. We aimed to identify reproducible SA-AKI subphenotypes using unsupervised machine learning on routinely collected clinical data and to characterize their clinical profiles and prognostic implications. We conducted a retrospective cohort study of 9,029 adult patients with SA-AKI from the MIMIC-IV v3.1 database. Within the 24-hour period before and after the diagnosis of SA-AKI, 19 core clinical variables—including vital signs and laboratory values—were extracted. A multi-algorithm consensus clustering framework integrating K-means, hierarchical clustering, and K-medoids was applied to identify subphenotypes. Cluster stability was assessed using consensus strength, entropy, and silhouette analysis, with visualization via t-SNE. Key discriminative features were identified using XGBoost with SHapley Additive exPlanations (SHAP). Primary outcome was in-hospital mortality; secondary outcome was 30-day survival. Four robust SA-AKI subphenotypes were identified: Subphenotype I (Hypoinflammatory): preserved oxygenation (SpO<sub>2</sub>≈100%), low inflammation (WBC 8.9 × 10<sup>9</sup>/L), hospital mortality 16.4%; Subphenotype II (Stable): highest urine output (470 mL/6&#xa0;h), lowest mortality (14.3%); Subphenotype III (Hyperdynamic): elevated heart/respiratory rates, impaired oxygenation (SpO<sub>2</sub>≈96%), mortality 21.6%; Subphenotype IV (Critical): severe renal dysfunction (creatinine 2.2&#xa0;mg/dL, urine output 253.5 mL/6&#xa0;h), coagulopathy, metabolic acidosis, and highest mortality (34.2%).Kaplan-Meier analysis confirmed significant differences in 30-day survival across subphenotypes (log-rank <i>P</i> &lt; 0.0001), with Subphenotype IV showing the poorest prognosis (median survival: ~18 days). Unsupervised consensus clustering reveals four clinically meaningful SA-AKI subphenotypes with markedly different outcomes. The “Critical” subphenotype is identifiable by routine parameters and warrants early aggressive intervention. This framework supports precision phenotyping for risk stratification and future targeted trials in SA-AKI.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-algorithm consensus clustering identifies four subphenotypes in sepsis-associated acute kidney injury

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月3日

查看全文
  • Shuang Chen,
  • Xiancheng Xu,
  • Guang Li,
  • Qingzhan Zeng,
  • Canlin Li,
  • Xiaoyue Li,
  • Shaohong Li,
  • Heng Li

摘要

Sepsis-associated acute kidney injury (SA-AKI) is a highly heterogeneous syndrome with variable clinical trajectories and outcomes. Conventional staging systems, such as KDIGO, fail to capture this heterogeneity, limiting risk stratification and personalized management. We aimed to identify reproducible SA-AKI subphenotypes using unsupervised machine learning on routinely collected clinical data and to characterize their clinical profiles and prognostic implications. We conducted a retrospective cohort study of 9,029 adult patients with SA-AKI from the MIMIC-IV v3.1 database. Within the 24-hour period before and after the diagnosis of SA-AKI, 19 core clinical variables—including vital signs and laboratory values—were extracted. A multi-algorithm consensus clustering framework integrating K-means, hierarchical clustering, and K-medoids was applied to identify subphenotypes. Cluster stability was assessed using consensus strength, entropy, and silhouette analysis, with visualization via t-SNE. Key discriminative features were identified using XGBoost with SHapley Additive exPlanations (SHAP). Primary outcome was in-hospital mortality; secondary outcome was 30-day survival. Four robust SA-AKI subphenotypes were identified: Subphenotype I (Hypoinflammatory): preserved oxygenation (SpO2≈100%), low inflammation (WBC 8.9 × 109/L), hospital mortality 16.4%; Subphenotype II (Stable): highest urine output (470 mL/6 h), lowest mortality (14.3%); Subphenotype III (Hyperdynamic): elevated heart/respiratory rates, impaired oxygenation (SpO2≈96%), mortality 21.6%; Subphenotype IV (Critical): severe renal dysfunction (creatinine 2.2 mg/dL, urine output 253.5 mL/6 h), coagulopathy, metabolic acidosis, and highest mortality (34.2%).Kaplan-Meier analysis confirmed significant differences in 30-day survival across subphenotypes (log-rank P < 0.0001), with Subphenotype IV showing the poorest prognosis (median survival: ~18 days). Unsupervised consensus clustering reveals four clinically meaningful SA-AKI subphenotypes with markedly different outcomes. The “Critical” subphenotype is identifiable by routine parameters and warrants early aggressive intervention. This framework supports precision phenotyping for risk stratification and future targeted trials in SA-AKI.