一种深度学习方法,可以预测长期接受抗凝治疗的患者随时间推移的出血风险
Soroush Shahryari Fard1, Theodore J Perkins2, Philip S Wells1
1The Ottawa Hospital Research Institute, The Ottawa Hospital, Ottawa, Ontario, Canada; Department of Medicine, University of Ottawa, Ottawa, Ontario, Canada.
Journal of thrombosis and haemostasis : JTH
|April 20, 2024
概括
使用患者随访数据的深度学习模型显著改善了接受抗凝治疗的患者的大型出血预测. 这种方法解释了不断变化的临床变化,增强了风险分层,超出了基线预测因素.
科学领域:
- 心血管医学 心血管医学
- 医疗保健中的人工智能
- 临床风险预测预测
背景情况:
- 目前用于预测抗凝治疗患者大出血的临床模型仅依赖基线数据.
- 这些模型未能纳入基线后发生的动态临床变化和事件,这可能会改变出血风险.
- 从不规则的时间序列后续数据开发预测模型是一个重大挑战.
研究的目的:
- 通过整合患者时间序列后续数据,证明深度学习在改善大出血预测方面的有效性.
- 为接受长期抗凝治疗的患者开发更准确的风险分层工具.
主要方法:
- 一项纵向队列研究,涉及2542名患者超过8年,其中118人经历严重出血.
- 四个基于神经网络的机器学习模型使用基线,后续或组合数据集进行了训练.
- 模型性能与使用30%数据的六个现有临床模型的修改版本进行了评估.
主要成果:
- 一组使用基线和后续数据的前和循环神经网络实现了最佳性能.
- 顶级模型在检测主要出血时表现出61%的灵敏度和82%的特异性.
- 这种深度学习模型的表现优于传统的临床模型,通过接受器操作特征曲线 (82%) 和精度回忆曲线 (14%) 下的区域的优越区域来证明这一点.
结论:
- 纳入时间序列后续数据显著提高了在延长抗凝治疗的患者大出血的预测.
- 深度学习方法为服用抗凝血药物的患者更准确,更动态的风险评估提供了一个有希望的途径.
相关概念视频
Anticoagulant Drugs: Low-Molecular-Weight Heparins
689
Hemostasis is a crucial process that prevents excessive blood loss from damaged blood vessels. It involves various mechanisms such as vasoconstriction, platelet adhesion and activation, and fibrin formation. The importance of each mechanism depends on the type of vessel injury. In contrast, thrombosis is the abnormal formation of a blood clot within the blood vessels, leading to potential complications if the clot obstructs blood flow. Thrombosis can be caused by increased coagulability of the...
689
Disorders of Hemostasis
881
Hemostasis, the process that stops bleeding after a blood vessel injury, is crucial for maintaining the integrity of the circulatory system. However, disorders of hemostasis can disrupt this delicate balance, leading to either excessive clotting or bleeding. These disorders can be broadly classified into thromboembolic disorders and bleeding disorders.
Thromboembolic Disorders
Two factors primarily cause thromboembolic conditions.
Thromboembolic Disorders
Two factors primarily cause thromboembolic conditions.
881


