Programmatic Theme: Clinical Research Informatics

Abstract: Medications are rarely mentioned as causes of allergy in clinical notes. We focus on allergen extraction to identify medications mentioned specifically as causing allergies. To overcome drawbacks of sequential machine learning classifiers, we propose two-stage instance filtering including dictionary lookup and SVM classification. Allergy-related lexical clues were employed to improve SVM classification. Our experimental results showed that the two-step approach extracted more allergen concepts from clinical notes without losing accuracy.

Learning Objective: Improve allergen extraction from clinical text with multistage instance filtering

Authors:

Youngjun Kim (Presenter)
Medical University of South Carolina

Stephane Meystre, Medical University of South Carolina