概述
FDA FAERS and WHO VigiBase contain spontaneous reports; those reports and statistical signals do not establish incidence or causation.
深入探讨
NLP can support extraction of suspected drug-event pairs from text, as demonstrated in a clinical narrative system evaluated in the 2018 n2c2 challenge. UMC describes VigiBase as a WHO database of reported events from health workers, the public and marketing authorisation holders. UMC signal management includes detection, validation, prioritisation, assessment and communication, combining computerized data mining with manual clinical assessment. ICH E2D(R1) minimum ICSR criteria are an event or observation, a suspect/interacting product, an identifiable patient and an identifiable reporter. A report or signal is not proof of causality.
现实世界的实施
A clinical NLP system can extract medication and adverse-event concepts and relations from narratives; one published n2c2 system was evaluated on that shared task.
MedDRA standardizes medical terms used in safety coding.
Disproportionality analysis screens report databases for drug-event reporting patterns that warrant assessment.
ICH E2D(R1) defines minimum criteria for an individual case safety report, while local rules determine reportability.
人工智能在药物安全和药物警戒领域的未来
Use models as support for extracting, coding or prioritising information, with documented validation and expert assessment. This is a cautious workflow recommendation; the cited sources do not establish that fully automated safety decisions are appropriate.
付诸实践
为什么这很重要
- 背景与规则. 行业背景决定了人工智能创意能否与现实接触。
- 质量控制. 领域约束会影响可接受的错误率和监督模型。
- 构建选择. 成功的部署使技术能力与一线工作流程保持一致。
留意
- 监管要求可能会使原本强大的原型失效。
- 历史数据可能会编码损害特定社区的偏见。
- 遗留系统可能会造成集成瓶颈和隐性成本。
后续步骤
- 让领域专家参与从问题框架到评估的整个过程。
- 在启动前设计审计跟踪和文档。
- 尽早验证合规性和安全义务。
- 分阶段推出，并具有明确的停止和回滚标准。
不断探索
常见问题
What is AI in Drug Safety and Pharmacovigilance?
Pharmacovigilance detects, evaluates and prevents suspected harm associated with medicines. FDA FAERS and WHO VigiBase contain spontaneous reports; those reports and statistical signals do not establish incidence or causation.
药物警戒的主要目标是什么？
药物警戒是一门监测药物不良反应并采取行动防止患者受到伤害的科学，尤其是在药物上市后。
In the published n2c2 clinical NLP system, what information did the system extract from narrative notes?
The paper describes a hybrid system for extracting medication and adverse-event concepts and relations from clinical narratives, evaluated in the 2018 n2c2 shared task.
不成比例分析在不良事件数据库中检测到什么？
比例报告比等不成比例的衡量标准会标记药物和经常意外出现的反应的组合，表明可能存在安全问题。
What is VigiBase, according to Uppsala Monitoring Centre?
UMC identifies VigiBase as the WHO global database of reported adverse events. A report is a suspicion and does not by itself confirm causality.
MedDRA 在药物警戒管道中的用途是什么？
MedDRA 是 NLP 系统将提取的反应映射到的标准化术语，因此可以对病例进行一致的比较和分析。