Shin Sug Kyun, Hur Ho, Cheon Eun Kyung, Oh Ock Hee, Lee Jeong Seon, Ko Woo Jin, Kim Beom Seok, Kwon YoungOk
Department of Internal Medicine, Nephrology Division, National Health Insurance Service Ilsan Hospital, Goyang, Korea.
Department of Surgery, National Health Insurance Service Ilsan Hospital, Goyang, Korea.
Yonsei Med J. 2017 Nov;58(6):1229-1236. doi: 10.3349/ymj.2017.58.6.1229.
Adverse drug events (ADEs) are associated with high health and financial costs and have increased as more elderly patients treated with multiple medications emerge in an aging society. It has thus become challenging for physicians to identify drugs causing adverse events. This study proposes a novel approach that can improve clinical decision making with recommendations on ADE causative drugs based on patient information, drug information, and previous ADE cases.
We introduce a personalized and learning approach for detecting drugs with a specific adverse event, where recommendations tailored to each patient are generated using data mining techniques. Recommendations could be improved by learning the associations of patients and ADEs as more ADE cases are accumulated through iterations. After consulting the system-generated recommendations, a physician can alter prescriptions accordingly and report feedback, enabling the system to evolve with actual causal relationships.
A prototype system is developed using ADE cases reported over 1.5 years and recommendations obtained from decision tree analysis are validated by physicians. Two representative cases demonstrate that the personalized recommendations could contribute to more prompt and accurate responses to ADEs.
The current system where the information of individual drugs exists but is not organized in such a way that facilitates the extraction of relevant information together can be complemented with the proposed approach to enhance the treatment of patients with ADEs. Our illustrative results show the promise of the proposed system and further studies are expected to validate its performance with quantitative measures.
药物不良事件（ADEs）会带来高昂的健康和经济成本，且随着老龄化社会中接受多种药物治疗的老年患者增多而有所增加。因此，医生识别导致不良事件的药物变得具有挑战性。本研究提出了一种新方法，该方法可基于患者信息、药物信息和既往ADE病例，就ADE致病药物提供建议，从而改善临床决策。
我们引入一种个性化的学习方法来检测具有特定不良事件的药物，其中使用数据挖掘技术为每个患者生成量身定制的建议。随着通过迭代积累更多的ADE病例，通过了解患者与ADE之间的关联，可以改进建议。在参考系统生成的建议后，医生可以相应地更改处方并报告反馈，使系统能够根据实际因果关系不断发展。
利用1.5年期间报告的ADE病例开发了一个原型系统，并且医生对从决策树分析获得的建议进行了验证。两个代表性案例表明，个性化建议有助于对ADE做出更及时、准确的反应。
当前系统中存在个别药物的信息，但未以便于共同提取相关信息的方式进行整理，所提出的方法可以对其进行补充，以加强对ADE患者的治疗。我们的示例结果显示了所提出系统的前景，预计进一步的研究将用定量方法验证其性能。
