Extract adverse drug events (ADE)

This model is engineered for the extraction of adverse drug events (ADEs) from unstructured clinical texts, leveraging several components finely tuned for this purpose.

Use the provided Streamlit playground application to test this service.

- Entity Recognition: Initially, the model accurately identifies entities related to adverse events (such as rash, nausea) and drug mentions within the text.

- Assertion Status Detection: Subsequently, it assigns an assertion status (e.g., present, negated, historical, hypothetical) to each identified ADE entity, taking into account the surrounding context.

- Relation Extraction: The final step involves the detection of relationships between the extracted ADE entities and drug entities, thereby identifying pairs of medication and associated adverse events.
- Document Classification: The model then classifies the entire document, discerning whether it contains a report of an ADE. This classification aids in filtering documents more likely to possess relevant ADE information.

The model proficiently correlates drugs with their respective adverse reactions, discerning if an event is drug-induced.

**Covered entities**: DRUG, ADE.

Covered assertion statuses: absent, present, conditional, associated_with_someone_else, hypothetical, possible.

Covered classes: ADE, noADE

Relations: 0 (absent), 1 (present).

搜集汇总

背景与挑战

背景概述

该模型专用于从非结构化临床文本中提取药物不良事件(ADE)，通过实体识别、断言状态判断、关系抽取和文档分类四步流程，准确关联药物与其不良反应。覆盖DRUG/ADE实体、6种断言状态和2类关系，支持区分药物诱发事件。