Exploring the Distributional Semantic Relation for ADR and Therapeutic Indication Identification in EMR

Extraction of relations and their semantic relations from a clinical text is significant to comprehend the actionable harmful and beneficial events between two clinical entities. Particularly to implement drug safety surveillance, two simplest but most important semantic relations are adverse drug reaction and therapeutic indication. In this paper, a method to identify such semantic relations is proposed. A large scale of nearly 1.6 million sentences over 50,998 discharge summary from Electronic Medical Records were preliminary explored. Our approach provided the three main contributions; (i) Electronic Medical Records characteristic exploration; (ii) OpenIE examination for clinical text mining; (iii) automatic semantic relation identification. In this paper, the two complementary information from public knowledge base were introduced as a comparative advantage over expert annotation. Then the set of relation patterns were qualified with 0.05 significant level. The experimental results show that our method can identify the common adverse drug reaction and therapeutic indication with the high lift value. Additionally, a novel adverse drug reaction and alternative drug for a specific symptom therapy are reported to support the comprehensive further drug safety surveillance. The paper clearly illustrates that our method is not only effortless from expert annotation, automatic pattern-specific semantic relation extraction, but also effective for semantic relation identification.