Kuan Yi Kiat Isaac, Kok Yixin Jamie, Liu Nigel Sheng Hui, Ong Brandon Jin An, Chee Ying Jie, Xu Chuanhui, Chow Minyang, Ramanathan Kollengode, Dalan Rinkoo, Ho Prahlad, Fan Bingwen Eugene
Department of Haematology, Tan Tock Seng Hospital, Singapore.
Yong Loo Lin School of Medicine, National University of Singapore, Singapore.
Res Pract Thromb Haemost. 2025 Jul 24;9(5):102984. doi: 10.1016/j.rpth.2025.102984. eCollection 2025 Jul.
Artificial Intelligence (AI) and machine learning (ML) are transforming hemostasis and thrombosis care, with applications spanning disease detection, risk assessment, laboratory testing, patient education, personalized medicine, and drug development. This narrative review explores AI's clinical utility and limitations across these 6 domains.
A comprehensive search of PubMed, Embase, and Scopus (up to February 2025) was conducted using terms related to AI, thrombosis, and hemostasis. Peer-reviewed, English-language studies were included, supplemented by manual and reference screening. Of 84 studies included, 38 focused on risk assessment, 16 on diagnostics, and others on personalized medicine, drug development, and patient engagement.
AI demonstrated high accuracy in diagnosing thrombotic events via imaging and electronic health record analysis, although sensitivity gaps persisted for complex cases. In laboratory settings, AI outperformed manual review in detecting errors (eg, sample mislabeling and clotted specimens). Risk stratification models surpassed traditional scores (eg, CHADS-VASc) in predicting thromboembolism, yet inconsistently performed in cancer-associated thrombosis. Personalized anticoagulation dosing and genetic severity prediction in hemophilia highlighted AI's precision. Chatbots and adherence tools have enhanced patient education while AI-driven drug discovery identified novel anticoagulants and repurposed existing therapies. Limitations included variable external validation, "black box" interpretability issues, and dataset biases.
AI offers significant promise for improving diagnostics, risk prediction, and individualized therapy in thrombosis and haemostasis. Future integration depends on transparent, validated, and equitable AI systems embedded within clinical workflows.
人工智能（AI）和机器学习（ML）正在改变止血和血栓形成护理领域，其应用涵盖疾病检测、风险评估、实验室检测、患者教育、个性化医疗和药物开发。本叙述性综述探讨了AI在这六个领域的临床实用性和局限性。
使用与AI、血栓形成和止血相关的术语，对PubMed、Embase和Scopus（截至2025年2月）进行了全面检索。纳入了同行评审的英文研究，并辅以手动筛选和参考文献筛选。在纳入的84项研究中，38项聚焦于风险评估，16项聚焦于诊断，其他研究则涉及个性化医疗、药物开发和患者参与。
AI通过影像学和电子健康记录分析在诊断血栓形成事件方面显示出高准确性，尽管复杂病例的敏感性仍存在差距。在实验室环境中，AI在检测错误（如样本错误标记和凝血标本）方面优于人工审核。风险分层模型在预测血栓栓塞方面超过了传统评分（如CHADS-VASc），但在癌症相关血栓形成中的表现不一致。血友病中的个性化抗凝剂量和基因严重程度预测突出了AI的精准性。聊天机器人和依从性工具增强了患者教育，而AI驱动的药物发现则识别出了新型抗凝剂并重新利用了现有疗法。局限性包括外部验证的可变性、“黑箱”可解释性问题和数据集偏差。
AI在改善血栓形成和止血的诊断、风险预测和个体化治疗方面具有巨大潜力。未来的整合取决于嵌入临床工作流程中的透明、经过验证且公平的AI系统。
