Clinical text mining for efficient extraction of drug-allergy reactions

Abstract

This work focuses on the extraction of allergic drug reactions in electronic health records. The goal is to annotate a sub-class of cause-effect events, those in which drugs are causing allergies. Little work has carried out in this field, seldom for Spanish clinical text mining, which is, indeed, the aim of this work. We present two approaches: a rule-based method and another one based on machine learning. Both approaches incorporate semantic knowledge derived from FreeLing-Med, a software explicitly developed to parse texts in the medical domain. Having recognised the medical entities for a given record, the challenge stands on triggering the underlying allergies. To this end, the knowledge is expressed as a set of semantic, syntactic and structural features. Our best system, based on machine learning, obtained a precision of 0.90 with a recall of 0.87, outperforming a rule-based approach.