ORAKLE:使用深度学习对mAke30的最佳风险预测,用于AKI相关的败血症患者
Wonsuk Oh1,2, Marinela Veshtaj3,4, Ashwin Sawant1,2,5
1Charles Bronfman Institute for Personalized Medicine, Icahn School of Medicine at Mount Sinai, New York, NY, USA.
Critical care (London, England)
|May 26, 2025
概括
一个新的深度学习模型ORAKLE准确地预测了急性损伤 (AKI) 严重病患者在30天内发生的重大不良事件 (MAKE30). 这种动态模型改进了静态预测,从而实现了个性化的患者护理.
科学领域:
- 关键护理医学 关键护理医学
- 腎臟病學 (nephrology) 是一種醫學.
- 医疗保健中的人工智能
背景情况:
- 30天内发生的重大脏不良事件 (MAKE30) 是急性脏损伤 (AKI) 中关键的以患者为中心的结果.
- 现有的MAKE30预测模型是静态的,不考虑动态的临床变化.
- 需要先进的模型,能够适应不断变化的患者数据.
研究的目的:
- 介绍ORAKLE,这是一个新的深度学习模型,用于预测MAKE30.
- 利用不断变化的时间序列数据进行更准确的MAKE30预测.
- 为 AKI 管理提供个性化,以患者为中心的方法.
主要方法:
- 使用MIMIC-IV,SiCdb和eICU-CRD数据库进行了回顾性研究.
- 开发ORAKLE使用动态DeepHit框架进行时间序列生存分析.
- 将ORAKLE与Cox和XGBoost模型进行比较,通过Brier分数评估校准.
主要成果:
- 在所有队伍中,Oracle在预测MAKE30方面表现出卓越的表现,AUROCs从0.83到0.85.8不等.
- 在AUROC和AUPRC,Oracle的表现优于XGBoost和Cox模型.
- 观察到很好的模型校准,布赖尔分数为0.21.1.
结论:
- 甲骨文是一个强大的深度学习模型,用于预测AKI重症患者的MAKE30.
- 该模型使用动态时间序列数据捕获不断变化的患者轨迹和治疗效果.
- 甲骨文为AKI提供量身定制的风险评估和个性化管理策略.
相关概念视频
Acute Kidney Injury IV: Diagnostic Studies and Prevention
64
Accurate diagnosis and effective prevention are critical in managing Acute Kidney Injury (AKI), which is linked to high mortality rates ranging from 10% to 80%. Timely recognition of at-risk patients and careful monitoring can significantly reduce the likelihood of kidney damage.Diagnostic Assessments:The diagnostic process starts with a comprehensive medical history to identify prerenal, intrarenal, and postrenal causes.Prerenal causes, such as dehydration, hypotension, or blood loss, should...
64
Acute Kidney Injury I: Introduction
85
Introduction:Acute Kidney Injury (AKI) describes a swift decrease in kidney function occurring over hours to days, characterized by the kidneys' failure to remove waste products from the bloodstream. This leads to dangerous complications like metabolic acidosis, fluid overload, and electrolyte imbalances, such as hyperkalemia, which can cause life-threatening arrhythmias. AKI is common in both hospital and outpatient settings, often triggered by dehydration, sepsis, or exposure to nephrotoxic...
85


