Pharmacovigilance using clinical notes.

Area of Science:

• Pharmacovigilance and Health Informatics

Background:

• Electronic health records (EHRs) are increasingly adopted, presenting an opportunity to leverage unstructured clinical notes for drug safety.
• Current pharmacovigilance methods can be enhanced by utilizing the rich data within free-text EHRs.

Purpose of the Study:

• To develop and demonstrate novel methods for extracting drug-adverse event and drug-drug interaction information from unstructured EHR data.
• To enable early detection and analysis of adverse events using a previously untapped data source.

Main Methods:

• Developed methods to annotate unstructured clinical notes and transform them into a deidentified patient-feature matrix.
• Encoded the patient-feature matrix using standardized medical terminologies.
• Applied the processed data for detecting drug-adverse event associations and adverse events linked to drug-drug interactions.

Main Results:

• Successfully flagged adverse events early, often preceding official alerts.
• Demonstrated the ability to filter spurious signals by adjusting for potential confounding factors.
• Enabled the compilation of adverse event prevalence information.

Conclusions:

• Analyzing large volumes of free-text clinical notes offers a powerful, untapped resource for drug safety surveillance.
• Data mining of EHR free text facilitates hypothesis generation and rapid analysis of suspected adverse event risks.
• These novel methods enhance the capabilities of pharmacovigilance by utilizing unstructured clinical data.