A drug-adverse event extraction algorithm to support pharmacovigilance knowledge mining from PubMed citations.

Area of Science:

• Pharmacovigilance
• Medical Informatics
• Computational Biology

Background:

• Adverse drug events (ADEs) pose significant risks to patient safety.
• A standardized, accessible knowledgebase of drug-ADE relationships is needed for effective ADE detection.
• Medical literature, particularly PubMed citations, is a rich source for identifying drug-ADE pairs.

Purpose of the Study:

• To develop and validate a method for automatically determining drug-specific adverse event (AE) causation from PubMed citations.
• To create a computable knowledgebase of drug-ADE relationships.
• To assess the method's performance in identifying neutropenia and myocardial infarction as AEs for various drugs.

Main Methods:

• A drug-ADE classification model was developed and initially trained to detect neutropenia.
• The classification method was applied to 76 drugs to identify neutropenia causation.
• The method was further validated by applying it to 48 drugs to detect myocardial infarction causation.

Main Results:

• The automated method demonstrated high performance in identifying drug-ADE relationships.
• Area Under the Receiver Operating Characteristic (AUROC) scores of 0.93 for neutropenia detection and 0.86 for myocardial infarction detection were achieved.
• The study successfully validated the method's applicability to different adverse events and drug sets.

Conclusions:

• The developed automated method is effective for extracting drug-ADE relationships from biomedical literature.
• This approach can contribute to building a valuable, publicly available knowledgebase for ADE detection and patient safety.
• The findings support the use of computational methods for pharmacovigilance and knowledge discovery from scientific text.