The use of medications has a central role in health care provision, yet on occasion, it may injure the person taking them as result of adverse drug events. A correct drug choice must be modulated to acknowledge both patients' status and drug-specific information. However, this information is locked in free-text and, as such, cannot be actively accessed and elaborated by computerized applications. The goal of this work lies in extracting content (active ingredient, interaction effects, etc.) from the Summary of Product Characteristics, focusing mainly on drug-related interactions, following a machine learning based approach. We compare two state of the art classifiers: conditional random fields with support vector machines. To this end, we introduce a corpus of 100 interaction sections, hand annotated with 13 labels that have been derived from a previously developed conceptual model. The results of our empirical analysis demonstrate that the two models perform well. They exhibit similar overall performance, with an overall accuracy of about 91%.
Summary of Product Characteristics content extraction for a safe drugs usage.
QUAGLINI, SILVANA
2012-01-01
Abstract
The use of medications has a central role in health care provision, yet on occasion, it may injure the person taking them as result of adverse drug events. A correct drug choice must be modulated to acknowledge both patients' status and drug-specific information. However, this information is locked in free-text and, as such, cannot be actively accessed and elaborated by computerized applications. The goal of this work lies in extracting content (active ingredient, interaction effects, etc.) from the Summary of Product Characteristics, focusing mainly on drug-related interactions, following a machine learning based approach. We compare two state of the art classifiers: conditional random fields with support vector machines. To this end, we introduce a corpus of 100 interaction sections, hand annotated with 13 labels that have been derived from a previously developed conceptual model. The results of our empirical analysis demonstrate that the two models perform well. They exhibit similar overall performance, with an overall accuracy of about 91%.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
