机器学习与休克指数衍生评分预测急性冠脉综合征患者造影剂肾病
这项回顾性病例对照研究纳入719例接受经皮冠状动脉介入治疗(PCI)的急性冠脉综合征(ACS)患者,分为造影剂肾病(CIN)组与非CIN组,利用机器学习算法分析临床、操作及血流动力学参数(包括休克指数)。采用梯度提升机(GBM)算法开发了新预测模型CIN-Predict 5,其中改良休克指数重要性评分最高(0.25)。CIN-Predict 5预测CIN的AUC为0.87,优于Mehran风险评分(AUC=0.75);模型预测值与院内死亡率(r=0.16,P<0.001)和住院时长(r=0.20,P<0.001)显著相关。作者认为该模型可为ACS后CIN高危患者的早期识别提供实用工具。
要点
- 在719例接受PCI的ACS患者中,改良休克指数是GBM算法中最重要的预测变量,重要性评分为0.25。
- CIN-Predict 5模型预测造影剂肾病的AUC为0.87,优于Mehran风险评分的AUC=0.75。
- CIN-Predict 5模型预测值与院内死亡率相关(r=0.16,P<0.001),与住院时长相关(r=0.20,P<0.001)。
- 该研究为回顾性病例对照研究,作者认为模型可为ACS后CIN高危患者的早期识别提供实用工具。
适用人群与场景
- 接受经皮冠状动脉介入治疗(PCI)的急性冠脉综合征(ACS)患者,共719例,来自回顾性病例对照研究。
研究设计与关键结果
- 研究类型为回顾性病例对照研究,纳入719例接受PCI的ACS患者,分为CIN组与非CIN组。
- 采用梯度提升机(GBM)算法开发CIN-Predict 5模型,纳入临床相关且统计学显著的变量,其中改良休克指数重要性评分最高(0.25)。
- CIN-Predict 5预测CIN的AUC为0.87,Mehran风险评分的AUC为0.75。
- CIN-Predict 5预测值与院内死亡率相关(r=0.16,P<0.001),与住院时长相关(r=0.20,P<0.001)。
证据类型、等级与局限
- 证据类型:回顾性病例对照研究。
- 局限:原文未报告局限性。
对床旁实践的改变
- 作者认为,基于休克指数和机器学习开发的GBM模型可为ACS后CIN高危患者的早期识别提供实用工具,有助于及时采取预防策略并改善临床决策。
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: Contrast-induced nephropathy (CIN) is a serious complication following acute coronary syndrome (ACS), leading to increased morbidity and mortality. Machine learning (ML), combined with parameters such as shock indices, can potentially improve CIN risk prediction by analyzing complex variable interactions and creating accessible, clinically applicable models. Methods: This retrospective case-control study included 719 ACS patients who underwent percutaneous coronary intervention (PCI). Patients were divided into two groups (CIN and non-CIN), and clinical, procedural, and hemodynamic parameters, including shock indices, were analyzed using machine learning algorithms. A new predictive model, CIN-Predict 5, was developed using the Gradient Boosting Machine (GBM) algorithm, incorporating clinically relevant and statistically significant variables. Correlations between model predictions and secondary outcomes, including in-hospital mortality and hospitalization duration, were evaluated. Results: Among the variables used in the GBM algorithm, the Modified Shock Index emerged as the most significant predictor, with an importance score of 0.25. The CIN-Predict 5 model achieved an AUC of 0.87, outperforming the Mehran Risk Score (Area Under the Curve (AUC) = 0.75) for predicting CIN. The secondary outcomes showed that CIN-Predict 5 correlated significantly with in hospital mortality (r = 0.16, P < 0.001) and hospitalization duration (r = 0.20, P < 0.001). Conclusions: The GBM-based model we developed, utilizing shock indices and derived through ML, provides a practical tool for early identification of high-risk CIN patients post-ACS, enabling timely preventive strategies and improving clinical decision-making.
参考来源
- Yavuz YE, Tatar S, Akıllı H, et al. MACHINE LEARNING AND SHOCK INDICES-DERIVED SCORE FOR PREDICTING CONTRAST-INDUCED NEPHROPATHY IN ACUTE CORONARY SYNDROME PATIENTS. Shock. 2025. PMID: 39965631. doi:10.1097/SHK.0000000000002567
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。