Text mining for the Vaccine Adverse Event Reporting System: medical text classification using informative feature...

Area of Science:

• Medical informatics
• Computational epidemiology
• Pharmacovigilance

Background:

• The US Vaccine Adverse Event Reporting System (VAERS) collects crucial post-vaccination safety data.
• Manual review of VAERS reports by medical officers is time-consuming.
• Standardized case definitions, like those from the Brighton Collaboration, aid report classification.

Purpose of the Study:

• To demonstrate a multi-level text mining approach for automated classification of VAERS reports.
• To assess the potential of text mining to reduce human workload in analyzing vaccine safety data.

Main Methods:

• A corpus of 6034 H1N1 vaccine VAERS reports was created, labeled as positive or negative for anaphylaxis.
• Text mining techniques extracted keywords and patterns, forming three feature sets.
• Rule-based and machine learning classifiers (boosted trees, weighted SVM) were trained and validated.

Main Results:

• Rule-based, boosted trees, and weighted SVM classifiers demonstrated strong macro-recall.
• The rule-based classifier achieved high sensitivity (79.05%) and specificity (94.80%).
• Some high-performing classifiers had a higher mean misclassification error rate.

Conclusions:

• Validated results indicate the feasibility of developing effective medical text classifiers for VAERS reports.
• Combining text mining with feature selection is a promising strategy for automating vaccine safety data analysis.
• This approach has the potential to considerably reduce reviewer workload in pharmacovigilance.