Drug-Drug Interaction Extraction-Based System: an NLP Approach

Purpose: Poly-medicated patients, especially those over 65, have increased. Multiple drug use and inappropriate prescribing increase drug-drug interactions, adverse drug reactions, morbidity, and mortality. This issue was addressed with several CDSS alerts. Health professionals have not followed these systems due to their poor alert quality and incomplete databases. Methods: Recent research shows a growing interest in using Text Mining via NLP to extract drug-drug interactions from unstructured data sources to support clinical prescribing decisions. NLP text mining and machine learning classifier training for drug relation extraction were used in this process. Results: In this context, the proposed solution allows to develop an extraction system for drug-drug interactions from unstructured data sources. The system produces structured information, which can be inserted into a database that contains information acquired from three different data sources. Conclusion: The architecture outlined for the drug-drug interaction extraction system is capable of receiving unstructured text, identifying drug entities sentence by sentence, and determining whether or not there are interactions between them.