Background: Each day, millions of health consumers seek drug-related information on the Web. Despite some efforts in linking related resources, drug information is largely scattered in a wide variety of websites of different quality and credibility.
Methods: As a step toward providing users with integrated access to multiple trustworthy drug resources, we aim to develop a method capable of identifying drug's dosage form information in addition to drug name recognition. We developed rules and patterns for identifying dosage forms from different sections of full-text drug monographs, and subsequently normalized them to standardized RxNorm dosage forms.
Results: Our method represents a significant improvement compared with a baseline lookup approach, achieving overall macro-averaged Precision of 80%, Recall of 98%, and F-Measure of 85%.
Conclusions: We successfully developed an automatic approach for drug dosage form identification, which is critical for building links between different drug-related resources.