Automatic detection of adverse events to predict drug label changes using text and data mining techniques

Area of Science:

• Pharmacovigilance
• Text Mining
• Data Science

Background:

• Adverse drug events (ADEs) require continuous monitoring.
• Traditional pharmacovigilance methods may miss emerging safety signals.
• Integrating diverse data sources can improve ADE detection.

Purpose of the Study:

• To evaluate the effectiveness of automated adverse event signal detection from text and open-source data.
• To assess the impact of these signals on predicting drug label changes.

Main Methods:

• Collected open-source adverse effect data from FAERS, Yellow Cards, and SIDER.
• Employed a shallow linguistic relation extraction system (JSRE) for ADE extraction from MEDLINE case reports.
• Utilized statistical methods for signal detection and prediction of label changes for 29 drugs.

Main Results:

• Successfully predicted 76% of drug label changes automatically.
• Identified 6% of drug label changes exclusively through text mining.
• JSRE precisely identified four ADEs from MEDLINE that were previously undetectable.

Conclusions:

• Automated prediction of drug label changes is feasible using data and text mining.
• Text mining is a mature technology capable of supporting pharmacovigilance.
• This approach enhances the ability to detect and respond to adverse drug events.